Where capital flows tells you where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

Two moods showed up in the same week. Investors poured early money into nuclear fusion and small modular reactors, frontier energy bets priced at valuations that assume a decade of patience, while one of China’s mid-tier battery makers raised a down round at a third below its prior valuation. Long-patience bets at the frontier, a squeeze in the commodity middle.

Where the Money Went

107 deals crossed the wire this week. Of those, 31 disclosed a dollar figure, for a combined total of roughly $1.21 billion (~¥8.7 billion). Treat that as a floor, not a ceiling: most Chinese rounds, especially the early ones, never publish an amount, so the real number is well above what the disclosed deals add up to.

🧠 AI (23 deals)

The busiest sector by deal count, and the one where chips, not models, drew the money. Several rounds converged on one idea: make AI computing cheaper by routing around Nvidia rather than trying to match it head-on. Sparse-computing chipmaker Moffett AI took ¥1 billion (~$139M); its chips skip the math the network does not need, since most of the numbers flowing through a neural network are zeros that contribute nothing, so the chip does more useful work per watt. A cluster of smaller chip startups raised on the same cost-per-answer logic. Disclosed AI capital came to about $251M (~¥1.8B).

⚡ New Energy & Materials (16 deals)

The week’s largest disclosed pool, about $317M (~¥2.3B), spread across battery materials, solid-state electrolytes, and specialty materials. The standout was a markdown rather than a milestone (see Spotlight), but capital also kept flowing toward the next battery chemistry rather than today’s.

🔬 Semiconductors (11 deals)

Mostly early and mostly cheap, around $26M (~¥187M) disclosed, but strategically dense: memory-maker-backed inference chips, compute-in-memory designs (chips that compute where data is stored, avoiding the memory-to-processor shuttling that is one of the largest energy costs in AI hardware), and chip-making equipment. Listed component suppliers showed up as investors, buying a window into the startups building around them.

🤖 Robotics (20 deals)

Money concentrated in components, not finished robots. Force sensors, robot hands, the interchangeable tools a robot arm swaps at its wrist, and the data used to train machines all drew more capital than any finished humanoid. The largest disclosed robotics round, ¥1 billion (~$139M), went to a force-control arm maker rather than a humanoid brand. Disclosed robotics capital was about $235M (~¥1.7B).

💡 Frontier Tech (6 deals)

Small in count, outsized in ambition: two nuclear-fusion startups, a small modular reactor designer, a neutral-atom quantum company (one of the leading approaches to a practical quantum computer, trapping individual atoms with laser beams as the basic units of calculation), and a brain-spine implant maker. Pre-revenue, pre-prototype, and priced like proven leaders. Disclosed capital was only about $42M (~¥300M), but here the headline figures are the valuations, not the cash in.

🚀 Aerospace & Defense (7 deals)

Electric air taxis and satellite laser-communication startups led here, about $146M (~¥1.05B) disclosed. The biggest was a ¥1 billion (~$139M) round for a maker of passenger electric aircraft that take off vertically like a helicopter but run on battery motors.

⚙️ Other Adv. Manufacturing (24 deals)

The catch-all sector and the second busiest, about $196M (~¥1.4B) disclosed: industrial lasers, 3D printing, optical and inspection equipment, and consumer hardware. A ¥900 million (~$125M) raise for an industrial-laser business was the largest.

Spotlight

Sunwoda EVB (欣旺达动力)

Series C, ¥1.68 billion (~$233M), ¥25 billion (~$3.5B) pre-money valuation

The power-battery arm of a listed cell maker, raising at a steep discount.

Sunwoda EVB is the electric-vehicle battery subsidiary of Shenzhen-listed Sunwoda Electronic, and one of China’s larger EV-battery makers, ranked sixth domestically by installed volume in 2025. This week it raised ¥1.68 billion from 13 investors at a pre-money valuation of ¥25 billion. The prior round valued the business at roughly ¥36.4 billion, so this round prices it about 31% lower.

The decline maps the broader battery correction. China makes far more battery capacity than its cars and grid can absorb, and a multi-year price war has crushed margins for everyone outside the two giants, CATL and BYD, which together still hold about two-thirds of the domestic market. Sunwoda EVB ran up roughly ¥8.8 billion (~$1.2B) in cumulative losses across 2021 to 2025, and turned its first quarterly profit, a thin ¥13 million (~$1.8M), only in the first quarter of 2026. The investor rotation matters as much as the valuation cut: the round is led by the investment arms of state banks (ICBC, ABC, and China Post), while the carmaker investors from its 2022 round, NIO, Xpeng, Li Auto, SAIC, and GAC, do not appear among the new backers. State capital stepping in where strategic carmakers stepped out is the shape of a sector correction on a single cap table.

Investors: ICBC Capital (工银资本), ABC International (农银投资), and China Post and Shenzhen state vehicles, among 13 total.

Announcement

Zhibang Lithium (智邦锂电)

Strategic round (de-facto Series A), ¥500 million (~$69M), ¥2.5 billion (~$347M) valuation

A two-year-old solid-state battery materials startup, and a billionaire’s first hard-tech battery bet.

Zhibang Lithium, founded in Quzhou, Zhejiang in March 2024, makes solid-state electrolytes, the component that replaces the flammable liquid inside a normal lithium battery with a solid one, which is the central unlock for safer, denser next-generation batteries. It works across all three competing chemistries (oxide, polymer, and sulfide). It raised ¥500 million at a ¥2.5 billion valuation. The money comes through a fund whose largest backer, at about 82%, is Yangshengtang, the holding company of Zhong Shanshan, the bottled-water billionaire behind Nongfu Spring.

This is Yangshengtang’s first bet on a hard-tech battery-materials company, and it arrives while commodity cell makers like Sunwoda EVB raise at markdowns: capital with no battery history is chasing the chemistry that comes next. The technical anchor is the team. The company’s chief technologist, Xu Xiaoxiong, was chief scientist for solid-state battery research at Ganfeng Lithium, one of the world’s largest lithium players, where he built the program from 2017. With solid-state still years from mass production, this is a wager on where the battery industry goes next.

Investors: Kunshan Gewuzhizhi Fund (昆山格物致知, Yangshengtang-backed), with Quzhou state capital and Xianfeng Changqing (险峰长青).

Announcement

Xingyun Integrated Circuit (行云集成电路)

Pre-A rounds, over ¥400 million (~$56M) cumulative, latest tranche led by Biwin

A Beijing chip startup trying to make AI answers cheap by dropping the most expensive memory.

Xingyun, founded in Beijing in 2023, designs chips for AI inference, the act of running a trained model to answer a query, as opposed to the far heavier work of training it in the first place. Its pitch is a cost end-run. High-end AI chips like Nvidia’s rely on high-bandwidth memory (HBM), fast, costly memory stacked right on top of the processor to feed it data quickly enough to keep it busy, and US export controls now largely deny China access to it. Xingyun instead builds around commodity memory, the cheaper LPDDR and NAND chips found in phones and storage drives. Those chips are slower, but inference draws on memory in shorter, more predictable bursts than training does, so the company claims the trade buys a memory-cost reduction of one to two orders of magnitude. Its all-in-one inference box, called Brown Ant (褐蚁), already runs open models like Alibaba’s Qwen3. This week listed memory-maker Biwin (佰维存储) extended its backing, bringing Xingyun’s Pre-A funding to over ¥400 million.

Ji Yu (季宇), the founder, came out of Huawei’s “Genius Youth” recruitment program, holds a Tsinghua doctorate, and worked on the compiler and architecture for Huawei’s Ascend AI chips, the most developed homegrown alternative to Nvidia’s chip-and-software stack. Biwin’s continued backing shows a listed memory company building around the thesis that cheap Chinese AI compute will run on the memory it already sells, not the HBM it cannot buy.

Investors: Biwin (佰维存储); earlier Pre-A tranches led by 5Y Capital (五源资本), SAIF Partners (赛富投资), and Primavera Capital (春华资本).

Announcement

Lightwheel (光轮智能)

Latest round (labeled Series A+), hundreds of millions of RMB, led by Ant Group

The simulation-data supplier behind some of the world’s best-known robots.

Lightwheel, based in Beijing, builds synthetic training data for robots and self-driving systems: photorealistic virtual worlds where a robot can practice picking up a cup or folding a towel millions of times before it touches the real object, far faster and cheaper than physical trials. This week it raised a new round, reported in the hundreds of millions of RMB and led by Ant Group, with China Renaissance (华兴资本) as financial adviser. The company reached unicorn status (a valuation above $1 billion) in an earlier round this spring; this round’s valuation was not disclosed.

Synthetic training data is infrastructure robot builders need regardless of which hardware wins, provided training in simulation transfers reliably to real machines, and Lightwheel’s customer list suggests the industry has made that bet. It is global and blue-chip: Nvidia names it a partner and connects it to GR00T, Nvidia’s general-purpose robot-brain model, software pre-trained to handle a wide range of physical tasks that robot builders adapt rather than build from scratch, and Lightwheel’s simulated worlds are part of what GR00T trains inside. Nvidia also lists Google DeepMind and Figure among Lightwheel’s users, and the company reports over ¥100 million (~$14M) in revenue and, per its own announcement, around $100 million in first-quarter orders. CEO Xie Chen (谢晨) previously ran autonomous-driving simulation at Nvidia, directed simulation at Cruise, and led it at NIO. Lightwheel is one of the few Chinese startups selling this layer to Western robot leaders, not only Chinese ones.

Investors: Ant Group (蚂蚁集团, lead), CCB Investment (建投投资); China Renaissance (华兴资本) as financial adviser. Reported customers beyond those Nvidia confirms (1X, Toyota Research Institute, Bosch) are company-stated.

Announcement

Qiaotian Intelligent (桥田智能)

Series A+ (second A+), around ¥100 million (~$14M), led exclusively by the National Machine Tool Fund

Not a humanoid, the quick-change tooling that robot factories run on.

Qiaotian, founded in Shanghai in 2016, makes robot end-effectors and the hardware around them: the quick-change “wrists” that let a robot arm swap tools in seconds, magnetic die-change systems, grippers, and industrial connectors. It is unglamorous, essential plumbing for any automated production line. This week it raised around ¥100 million in its second Series A+ round, and the investor is the whole point. The National Machine Tool Industry Investment Fund (国家工业母机产业投资基金), a roughly ¥150 billion state fund created to reduce China’s dependence on foreign precision-manufacturing equipment, took the round on its own.

A national fund whose mandate is the machine tools that make everything else, writing an exclusive check into an end-effector maker, treats quick-change tooling as strategic industrial infrastructure, and it fits the week’s broader pattern of state vehicles leading in hard tech. The company says its core business grew 30 to 40% year over year in early 2026, with customers including Geely; the business is profitable, low-profile, and now state-anointed.

Investors: National Machine Tool Industry Investment Fund (国家工业母机产业投资基金, exclusive).

Announcement

Junhe Atomic (钧合原子)

Angel round, hundreds of millions of RMB, ~¥2 billion (~$280M) valuation

A startup designing small nuclear reactors, founded by the man who sold China’s flagship reactor abroad.

Junhe Atomic, based in Shanghai, designs small modular reactors (SMRs): compact, factory-built nuclear reactors meant to be cheaper and faster to deploy than the giant gigawatt plants, and increasingly pitched worldwide as a clean answer to the electricity demand of AI data centers. The company raised an angel round of hundreds of millions of RMB from more than a dozen investors, with funds earmarked for the conceptual design of its first reactor and team building. The roughly ¥2 billion valuation comes from investor databases and is not confirmed in the announcement; treat it as an early indication on a concept-stage company, not a verified figure.

The founder is why this round attracted more than a dozen investors. Tian Jiashu (田佳树) spent 30-plus years in China’s nuclear industry, served as a deputy chief engineer at both of the country’s nuclear giants, CNNC and CGN, and ran Hualong International Nuclear Power, the joint venture created to sell China’s flagship Hualong One reactor abroad. SMRs are where Western startups like Oklo and X-energy have drawn heavy investor and government backing; China’s are mostly state-run, which makes a privately funded SMR designer with this pedigree rare. A near-term catalyst sits in the background: the state-built Linglong One, the world’s first commercial land-based SMR, is targeting commissioning in 2026, which would de-risk the regulatory path for private players too.

Investors: Lingang Sci-Tech (临港科创), Eastern Bell Capital (东方富海), and Cohesion Capital (聚合资本), among more than a dozen.

Announcement

Dongsheng Fusion (东昇聚变)

New round (labeled Pre-A), hundreds of millions of RMB, co-led by CMC Capital’s carbon-neutrality fund

A fusion startup chasing the cleaner, harder reaction, incubated at a top Shanghai university.

Dongsheng Fusion, based in Shanghai and incubated by Fudan University’s Institute of Modern Physics, is pursuing a particularly ambitious flavor of nuclear fusion. Most fusion efforts fuse deuterium and tritium, which throws off torrents of neutrons that damage the reactor and make it radioactive. Dongsheng targets a deuterium-helium-3 reaction that produces almost no neutrons, which in principle allows a smaller machine that could sit closer to a city. It uses high-temperature superconducting magnets and AI-driven plasma control, and runs an experimental device named Chenguang (晨光). This week it raised a new round, reported in the hundreds of millions of RMB, co-led by the carbon-neutrality fund of CMC Capital.

The deal is a study in how China is pricing frontier science. The founder, Fudan fusion physicist Xu Min (许敏), was recruited in 2025 specifically to build this team; the company was founded only in mid-2025 and raised its angel round in January 2026, drawing a syndicate of Sequoia China, IDG Capital, and Hillhouse. Investor databases put this round’s valuation near ¥3 billion, but no announcement confirms it, and for a company with no demonstrated plasma result and barely a year of history, that figure reflects the team’s pedigree, not any measure of technical progress. Fusion is the highest-conviction, longest-duration bet on this list.

Investors: CMC Capital (CMC资本) carbon-neutrality fund (co-lead); angel backers Sequoia China (红杉中国), IDG Capital (IDG资本), Hillhouse (高瓴), and Fudan Science Park (复旦科创).

Announcement

Also on the Radar

Tianji Intelligent (天机智能). Series B, ¥1 billion (~$139M). Force-control robot arms; one of the week’s largest robotics rounds, co-led by Meituan and Hillhouse, signaling a delivery giant positioning to own the arms behind its automation. Valued at ¥9 billion (~$1.25B).

Moffett AI (墨芯人工智能). Series C, ¥1 billion (~$139M). AI chips built on “sparse computing,” which skips the zero-value math in a neural network to do more work per watt. Valued at ¥6 billion (~$833M).

Beijing Link-Touch (蓝点触控). Series C+, hundreds of millions of RMB. Six-axis force sensors, the touch-sensitive joints that let robots handle delicate objects. Backed this round by carmaker and chip-industry strategics including SAIC. Valued at ¥6 billion (~$833M).

Volant Aerotech (沃兰特航空). Series C+, ¥1 billion (~$139M). Passenger electric vertical-takeoff-and-landing aircraft (air taxis). Among the better-funded Chinese eVTOL makers, valued at ¥15 billion (~$2.08B).

Shenfu Jianxing (神复健行). Angel, ¥300 million (~$42M). Implantable brain-spine interface aiming to restore movement after paralysis by rebuilding the signal path between brain and spinal cord. Led by Cathay Capital.

MatriQ (原子矩阵). Series A, hundreds of millions of RMB. Neutral-atom quantum computing, one of the leading hardware approaches to a useful quantum machine. Backed by Monolith and Vertex.

HeyGears (黑格科技). Series C, ¥300 million (~$42M). Industrial 3D printing, with a strong dental-printing position and a majority of revenue from overseas markets.

Xiandao Laser (先导激光). Series A, ¥900 million (~$125M). The laser-systems business of industrial group Xiandao; one of the week’s largest manufacturing rounds, funded by its parent.

Shuowei Photonics (烁威光电). Pre-A, ¥100 million (~$14M). Perovskite solar cells engineered for spacecraft. Perovskite can be made as thin, lightweight films rather than the rigid silicon wafers in normal panels, generating more power per gram, which matters when every kilogram sent to orbit costs a fortune.

Vefan Intelligent (维泛智能). Seed, hundreds of millions of RMB. Brain-inspired chips meant to be a robot’s onboard “brain,” pairing standard processing with spiking neural networks that fire in pulses only when something changes, the way an eye reacts to motion, using a fraction of the power of always-on chips.

The Bigger Picture

The battery glut reaches the cap table

China’s battery industry became the go-to example of its manufacturing scale. This week showed the other side of that scale. Sunwoda EVB, the sixth-largest domestic battery maker by 2025 installed volume, raised at a valuation about 31% below its prior round, with state banks in and the 2022 carmaker investors out.

The underlying problem is overcapacity. In the first eight months of 2024, China produced 419.7 GWh of power batteries but installed only 219.2 GWh in vehicles, an installation rate of just 52%, meaning roughly half the batteries made that year sat in warehouses rather than cars. That gap between output and demand is what forces prices down: factories keep running to cover their fixed costs, inventory piles up, and cell makers cut prices to move it. Prices fell accordingly: as the price of lithium carbonate, the key raw material, collapsed through 2024, battery-cell prices dropped with it. For a manufacturer whose costs are mostly fixed plant and equipment, a sustained price cut on every cell sold can erase a year of profit fast. The pain is not evenly shared. CATL and BYD together still held about 65% of China’s installed volume in 2025, and run their plants at 80 to 90% utilization with healthy margins. The second tier runs at 60 to 70%, and by early 2026 several were near breakeven or in the red on their battery divisions, carrying debt above 60% of assets.

That gap is why this week’s two battery deals point in opposite directions. The commodity-cell maker takes a markdown and a state-bank backstop. The two-year-old solid-state startup, with no revenue, attracts a billionaire’s first battery bet at a ¥2.5 billion valuation. Capital is shifting away from the part of the market that has become a commodity and toward the chemistry that might reset the game. The battery sector is the clearest case of what happens when production outruns demand: output keeps growing, prices fall, and only the leaders with utilization high enough to absorb it stay profitable.

China’s race to build a non-Nvidia compute stack

Xingyun builds inference chips around cheap commodity memory instead of HBM; others raised this week on sparse computing and compute-in-memory designs. The common thread is cost: do AI computing more cheaply, and do it without the parts the US will not sell.

That instinct has a powerful tailwind. US export controls that took effect in late 2024 cut China off from the leading-edge high-bandwidth memory that high-end AI chips depend on, turning memory substitution from a clever idea into a structural necessity. And Huawei is building the rest of the alternative stack in public. In August 2025 it open-sourced CANN, the software layer that lets developers program its Ascend AI chips. That is a direct answer to Nvidia’s CUDA, the programming environment Nvidia has spent nearly two decades turning into the default language of AI development; because nearly every AI tool and library was written for CUDA first, moving to a rival chip has meant rewriting that whole software stack, which is the real lock-in. Opening CANN lets any developer adapt it, the first step toward building the same kind of gravity. In March 2026 Huawei took its Atlas 950 supercomputing system to MWC in Barcelona, its first global showcase of a cluster built entirely on Huawei silicon, interconnect, and in-house memory.

The startups raising this week sit downstream of that push. They do not need to beat Nvidia on raw performance; they need to make Chinese AI workloads run acceptably on parts China can actually obtain and afford. Inference, the day-to-day work of answering queries, is far less memory-hungry than training, which is exactly why a wave of startups is attacking it with cheaper memory and novel architectures. The bet across the sector is the same: in a market walled off from the best imported hardware, the winner is whoever produces acceptable inference at the lowest cost on parts China can actually source.

Deal data compiled from Chinese corporate announcements, financial news, and investment databases. Amounts and valuations are as reported and may be incomplete. For informational purposes only. Not investment advice. Verify all figures independently before acting on them.

Capital flows trace where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

The week’s largest disclosed round went to a rocket company, not an AI lab. After StepFun’s $2.5B Pre-IPO absorbed all the attention last week, capital spread wide and the most interesting deals turned up in the gaps: a Starship-architecture rocket in aerospace, a chip company building China’s answer to Nvidia’s NVLink, and a quantum computing angel round with SMIC’s foundry arm at the table.

Where the Money Went

85 deals across seven sectors. 11 disclosed exact amounts totaling $263M (~¥1.9B). Aerospace, which recorded zero deals last week, came back with the week’s biggest disclosed round at ¥500M (~$69M). The stage distribution continues the 2026 barbell: seed and angel bets at one end, a handful of larger rounds at the other, and a thin middle where Series A and B capital would normally concentrate.

🧠 AI (19 deals)

The deals cluster around infrastructure: the systems that sit between raw GPU hardware and the AI models that consume it. OriginFlow (渊澈太初) raised hundreds of millions of RMB at Pre-A for electromyography data infrastructure, capturing the electrical signals from human muscles to train robots to replicate dexterous hand movements. Approaching.AI (趋境科技) raised at a similar scale for the compute-management layer that sits between raw GPU clusters and the AI companies that rent them, scheduling which jobs run on which chips, at what priority, and at what cost. The company claims to process over a trillion tokens per day for customers including Zhipu and Kimi. DeepCybo (深度机智) raised hundreds of millions at Pre-A for PhysBrain, a system that trains robots by watching video of humans performing tasks rather than requiring expensive real-robot training data. HiDream.ai (智象未来) completed a double close, Series B and B+, reportedly with Hefei government funds backing the company across both rounds.

⚡ New Energy & Materials (10 deals)

International stories dominate. Zendure (征拓), covered in Spotlight below, raised a Series B for home energy storage systems already selling across Europe. Yanhe Solar (炎和科技) raised a Series A+ for thin, flexible perovskite solar cells, a next-generation solar technology that works in low light and can be embedded directly into consumer electronics. Yanhe Solar’s cells are already shipping to Lenovo, Signify (the company behind Philips lighting), and Dreame across more than 50 countries. Shanghai Energy Bridge (能量桥) raised at Pre-A for superconducting power distribution, a technology that transmits electricity with near-zero resistance, targeted at AI data centers where power delivery is becoming the binding constraint on how many GPUs you can run.

🔬 Semiconductors (16 deals)

The largest sector by deal count after AI. Tanwei Chiplink (探微芯联), covered in Spotlight, raised hundreds of millions at Pre-A for GPU interconnect chips. WingSemitech (隼瞻科技) raised close to ¥100M (~$14M) at angel for a chip design platform built on RISC-V, the open-source processor architecture that Chinese companies can use without licensing from any American company. Shenzhen Rayhertz Technology (镭赫技术) raised a Series A for semiconductor inspection equipment, the tools that scan chips during manufacturing to catch defects. Four angel-stage semiconductor deals reflect continued seed-level capital formation across the supply chain, from materials to packaging to sensors.

🤖 Robotics (15 deals)

Components and platforms over complete robots. AGILINK (临界点), covered in Spotlight, raised at A+ for dexterous robot hands that supply multiple humanoid manufacturers. Botshare (擎天租) raised at A+ for a platform that rents industrial and service robots by the month rather than requiring factories and warehouses to buy them outright, valued at ¥7B (~$972M), reportedly five months after founding. SEAHI Robotics (世航智能) raised at A+ for maritime robots that clean ship hulls and inspect underwater infrastructure, with the Ministry of Industry and Information Technology’s National SME Development Fund classifying the company as strategic technology. RoboParty (萝博派对), reportedly founded by a 21-year-old, raised tens of millions of dollars at angel for an open-source humanoid robot with Xiaomi backing across multiple rounds.

💡 Frontier Tech (11 deals)

Three brain-computer interface companies raised in this week’s data. See The Bigger Picture below. Juliang Guangqi (矩量光启), covered in Spotlight, raised ¥200M (~$28M) at angel for superconducting quantum computing. NeuraInsight raised a Series A from MiraclePlus (奇绩创坛), the Y Combinator-affiliated accelerator, for a structured database of neuroscience research and brain-activity data used by AI labs studying how biological intelligence works. Three synthetic biology deals covered companies engineering microbes to produce proteins, nutrients, and specialty chemicals through fermentation, an industrial process that can replace petroleum-derived ingredients in food, cosmetics, and materials with bio-manufactured alternatives.

🚀 Aerospace & Defense (7 deals)

Back from zero deals last week. Astronstone (宇石空间), covered in Spotlight, raised ¥500M (~$69M) at Series A for a reusable rocket, the week’s largest disclosed round. TransFuture Aviation (华羽先翔) raised hundreds of millions at Pre-A for an electric vertical-takeoff aircraft, with Ant Group leading what is reportedly its first investment in the eVTOL sector. Beijing Interstellar Development (星际开发) raised a Series A for a reusable orbital cargo vehicle, eight years after founding and still without a first flight.

⚙️ Other Advanced Manufacturing (7 deals)

No single thread across seven deals spanning optical eyewear, 3D printing materials, water treatment membranes, and networking equipment. Shenzhen Zhongjian Technology (众见科技) raised tens of millions at Series A for optical eyewear, backed by SEE Fund and AfterShokz (the bone-conduction headphone company).

Spotlight

Astronstone (宇石空间)

Series A, ¥500M (~$69M), ¥3B (~$417M) valuation

The engineer who helped build China’s heaviest operational rockets is now building a private Starship analog with chopstick catch.

Astronstone is the second chopstick-recovery rocket company we’ve covered in three weeks. In the May 2 edition of The China Round, we spotlighted Cosmoleap (大航跃迁), which raised ¥500M (~$69M) at Series A for a 70-meter, 18-ton-to-LEO vehicle with tower-catch testing underway since November 2024. Now Astronstone has raised the same amount, at the same stage, for the same core architecture: stainless steel airframe, liquid oxygen and methane propellant, and tower-based chopstick recovery. Both target first flight in 2027. Two companies raising ¥500M Series A rounds within three weeks for the same technical approach is a signal that Chinese investors see the Starship architecture as proven enough to back with scale capital, not just one contrarian bet.

The architecture matters because each element solves a specific cost problem. Stainless steel is heavier than carbon fiber but dramatically cheaper to manufacture and more tolerant of the extreme temperatures during reentry. Methane burns cleaner than kerosene, reducing engine refurbishment between flights. Chopstick tower recovery, catching the returning booster with mechanical arms rather than landing it on legs, eliminates the weight penalty of landing gear. The combination makes rockets cheap enough to treat as reusable infrastructure rather than expendable hardware. What distinguishes Astronstone is the founder. Tang Wen (唐文) holds a PhD in aerospace engineering from Tsinghua University and spent years at the First Academy of China Aerospace Science and Technology Corporation (CASC), where he worked on propulsion and structural systems for the CZ-5 and CZ-7 launch vehicles. The CZ-5 is China’s heaviest operational rocket. The CZ-7 is the cargo workhorse that supplies China’s space station. Tang’s experience is in the specific engineering discipline, liquid rocket propulsion and structures, that a Starship-class vehicle demands. The company has raised ¥1B (~$139M) across four rounds in one year, with first flight targeted for the first half of 2027. Landspace (蓝箭航天), currently the furthest ahead among Chinese reusable rocket startups, flew its Zhuque-3 on a test mission in late 2025. SpaceX filed its S-1 on May 20 and is set to list in June at a reported valuation of $1.5 trillion to $2 trillion, the largest IPO in history. Two Chinese companies raising ¥500M each for the same core architecture in the same month that SpaceX prices its public offering leaves little ambiguity about where the technology stands.

Investors: Kunlun Capital (昆仑资本, co-lead), Gaorong Capital (高榕创投, co-lead), and 11 others including Xingxiang Capital (兴湘资本).

Announcement

Tanwei Chiplink (探微芯联)

Pre-A, hundreds of millions RMB, ¥1.5B (~$208M) valuation

One of the first Chinese companies building a clean-room alternative to the high-speed links that let Nvidia’s GPUs talk to each other inside AI training clusters.

Tanwei Chiplink builds GPU interconnect chips. When companies train large AI models, they connect thousands of GPUs into a single cluster where each chip needs to communicate with every other at extreme speed. Nvidia dominates this layer through NVLink (the chip-to-chip protocol) and NVSwitch (the switching chip that routes traffic between GPUs), creating a lock-in that goes beyond the GPU itself. Even if a Chinese company builds a competitive AI processor, it still needs the high-speed fabric to connect them. Tanwei’s products, ACCLink and ACCSwitch, target this gap directly: the protocol IP delivers 50-nanosecond latency, fast enough that thousands of chips can coordinate without creating a bottleneck, and can scale to clusters of 4,096 GPUs, large enough to train frontier AI models. The team spun out of Tsinghua University’s Brain-Inspired Computing Research Center.

Inspur Information (浪潮信息), China’s largest AI server manufacturer, invested strategically, a real customer pulling the technology into its server product line. Inspur’s own statement cited “cooperation prospects on smart computing product lines” and noted that Tanwei’s product planning “benchmarks Nvidia’s NVLink+NVSwitch.” Three Beijing government funds, Capital Technology Development Group (首都科技发展集团), Zhongguancun Science City (中关村科学城), and Guozhong Capital (国中资本), all participated, treating GPU interconnect as AI sovereignty infrastructure. Tanwei is not alone in this race: Rongxin Zhiyuan (容芯致远) raised a separate angel round the same week for optical interconnect targeting the same layer, and Hygon (海光信息) published an open interconnect standard last year. But Tanwei is among the first dedicated pure-play chip companies attacking the problem, and Inspur’s strategic investment suggests real engineering engagement behind the capital.

Investors: ShenJing Jiacheng (盛景嘉成, lead), Zhengxuan Investment (正轩投资), Inspur Information (浪潮信息, strategic), iSoftStone (软通动力, strategic), Huagai Capital (华盖资本), Fengnian Capital (丰年资本), Capital Technology Development Group (首都科技发展集团), Zhongguancun Science City (中关村科学城), Guozhong Capital (国中资本), Jinpu Investment (金浦投资), and others.

Announcement

Zendure (征拓)

Series B, hundreds of millions RMB, ¥1.5B (~$208M) valuation

A Chinese home energy company ranked first in France and first among independent brands in Germany.

Zendure builds home energy management systems, the hardware and software that let households store solar energy, sell electricity back to the grid during peak pricing, and manage their home’s power consumption automatically. According to company figures, Zendure ranks first in France by total energy storage sales, first among independent brands in Germany, and top three in the Netherlands. Per company, revenue has grown approximately 200% year over year. The company launched PowerHub in Dusseldorf on May 11, a plug-in hub that connects rooftop solar, battery storage, and household loads into one managed system, with shipping starting July 2026. Its ZenWave subscription service automatically buys electricity when wholesale prices drop and sells stored energy back when prices spike.

The round was co-led by Han’s Laser (大族激光), a Shenzhen-listed industrial laser and precision manufacturing company with a market cap exceeding ¥100B (~$14B), alongside DeLink Capital (德联资本). A strategic investment from a laser manufacturer points toward supply chain and hardware production cooperation.

The structural tailwind is European electricity volatility. Wholesale electricity prices rose across the EU in 2025, with widening intraday spreads between solar-generation midday and peak evening demand. That price gap is what residential storage monetizes, and Zendure’s software automates the arbitrage. The EU deployed 27.1 GWh of new battery storage in 2025, a 45% increase and the twelfth consecutive record year, yet over 90% of EU battery manufacturing capacity targets EVs, leaving residential storage largely open to Chinese manufacturers.

Investors: Han’s Laser (大族激光, co-lead), DeLink Capital (德联资本, co-lead).

Announcement

AGILINK (临界点)

Series A+, hundreds of millions RMB, ¥8B (~$1.1B) valuation

The company that makes the hands for China’s humanoid robots raised three rounds in 28 days and landed an LG design win at CES.

AGILINK makes dexterous robot hands, the multi-fingered grippers that allow humanoid robots to pick up objects, turn handles, and perform the kind of manipulation tasks that fixed industrial grippers cannot. The company was spun off from AgiBot (智元机器人) in January 2026, inheriting AgiBot’s existing hand product line and the engineering team behind it. The products it inherited and continues to produce include the OmniHand series: the OmniHand (10 active degrees of freedom, ¥9,800 or ~$1,360), the OmniHand Touch (10 degrees of freedom with tactile sensing, ¥14,800 or ~$2,060), and the OmniHand Pro (12 degrees of freedom). The company shipped thousands of units per quarter in 2025, per its fundraising materials. At CES 2026, LG’s first humanoid robot, CLOiD, used AGILINK’s hands as its core manipulation hardware, a design win that validated the “sell to every body maker” strategy.

AGILINK completed three rounds in 28 days, from founding on January 14 to closing its Series A on February 11. Better to bet on the hands that go into every humanoid robot than to pick a single body maker. As the humanoid sector matures, the market is pricing the component layer, the parts every robot needs, higher than many of the robot makers themselves. AGILINK is not the only company investors are backing at this layer. LinkerBot (灵心巧手), covered in the May 2 edition, has shipped over 10,000 dexterous hands and raised seven rounds in 18 months at a valuation above ¥15B (~$2.1B). The logic behind spinning out hands as an independent company is that humanoid robot makers are competing fiercely on price and form factor, and most will not survive the shakeout. But every survivor needs hands. A standalone hand company can sell to all of them without being locked to one body maker’s fate, and investors can back the category without picking a winner. AGILINK at ¥8B and LinkerBot at ¥15B+ both sit above multiple humanoid body companies that are their customers. The A+ round, recorded in mid-May, was not accompanied by a public announcement. The Series A in February was led by a major internet company, with co-investors including Baidu Ventures (百度风投), Yunfeng Capital (云锋基金), Synstellation Capital, and Joyson Electronics (均胜电子).

Investors (Series A, Feb 2026): Major internet company (lead, unnamed), Baidu Ventures (百度风投), Yunfeng Capital (云锋基金), Synstellation Capital, Joyson Electronics (均胜电子), and others.

Announcement (Series A, Feb 2026)

Juliang Guangqi (矩量光启)

Angel, ¥200M (~$28M), ¥1B (~$139M) valuation

A quantum computing startup raised one of China’s largest angel rounds with a semiconductor foundry’s venture arm at the table, and a team forged at Sandia National Laboratories.

Juliang Guangqi is building a full-stack superconducting quantum computer: not just the quantum processor chip but the entire system, including cryogenic control hardware that cools circuits to near absolute zero, error-correction algorithms that compensate for the fragility of quantum states, and a quantum operating system. Founded in mid-2025 by Yu Wenlong (于文龙), who holds a physics PhD from Georgia Tech and an applied physics degree from the University of Science and Technology of China (USTC), with post-doctoral work at Sandia National Laboratories, one of the US Department of Energy’s premier weapons and technology research facilities. A co-founder reportedly worked at NIST, the US National Institute of Standards and Technology, on superconducting chip processes. The company claims to have completed China’s first quantum chip produced on an 8-inch industrial wafer, the standard disk size used in high-volume semiconductor factories, a manufacturing milestone marking the shift from lab-built prototypes to factory-scale production.

¥200M (~$28M) is an unusually large angel round for quantum computing, a field where most Chinese startups raise single-digit millions at seed stage. The investor that matters most is SMIC Capital (中芯聚源), the venture arm of Semiconductor Manufacturing International Corporation, China’s most advanced chip foundry. Quantum computing chips require sophisticated semiconductor fabrication, and having the foundry’s own investment arm at the table opens a potential path to manufacturing access. The deal has drawn comparisons in Chinese press to a “Fairchild moment”, a reference to Fairchild Semiconductor, the 1957 startup whose manufacturing innovations became the fabrication backbone that other companies built on. The analogy is specifically about foundry access: the argument is that Juliang Guangqi’s SMIC-aligned quantum chip process could become the manufacturing standard other Chinese quantum companies depend on. It is an ambitious claim for a company less than a year old, but the combination of US national laboratory experience, Chinese university pedigree, and foundry alignment is the kind of team composition that makes investors pay up early.

Investors: Junshan Capital (钧山资本, co-lead), Heli Capital (和利资本, co-lead), SMIC Capital (中芯聚源), Yixin Investment (翌昕投资), Yirong Capital (翊荣资本), Qiancheng Capital (千乘资本), Detong Capital (德同资本).

Announcement

Also on the Radar

Botshare (擎天租). Series A+, hundreds of millions RMB, ¥7B (~$972M) valuation. A robot rental platform that sells access to industrial and service robots by the month rather than requiring factories and warehouses to buy them outright. Previously featured in the May 2 edition at Pre-Series A and ¥4.2B (~$580M) valuation with four rounds in 105 days. Now at ¥7B and reportedly six rounds since December 2025. The company reportedly launched an international rental platform in 13 countries in April 2026, with overseas rental prices reportedly running six times the domestic rate.

Yanhe Solar (炎和科技). Series A+, hundreds of millions RMB, ¥1B (~$139M) valuation. Thin, flexible perovskite solar cells that work indoors at low light levels, embedded into consumer electronics rather than competing with rooftop solar panels. Products shipping in 50+ countries across 100+ brand supply chains, including Lenovo, Signify, and Dreame. CES 2026 showcase included solar-powered e-paper displays, smart glasses, and cameras. A 100MW production line has been operational in Changde, Hunan since August 2025.

HiDream.ai (智象未来). Series B+, ¥100M (~$14M), ¥5.5B (~$764M) valuation. An image generation model, the kind of AI that creates pictures from text descriptions, large enough to compete with Midjourney and DALL-E. A Series B followed by a B+ in close succession, reportedly with Hefei government funds across both rounds. Per company materials, over one million monthly global users.

TransFuture Aviation (华羽先翔). Pre-A, hundreds of millions RMB, ¥1.5B (~$208M) valuation. Electric vertical-takeoff aircraft. Ant Group (蚂蚁集团) led the round, reportedly its first eVTOL investment. A fintech company with no prior aerospace history moving into air mobility suggests the sector is being repriced from deep-tech moonshot to near-term infrastructure play. Investors: Ant Group (蚂蚁集团, lead), Fosun Venture (复星创富), Bian’an Shidai (彼岸时代).

SEAHI Robotics (世航智能). Series A+, undisclosed amount, ¥3.5B (~$486M) valuation. Maritime robots for ship hull cleaning and underwater inspection. Previously featured in the May 2 edition with simultaneous A+ and A++ rounds and proprietary thrusters rated for 10,000 operational hours versus a 300-400 hour industry standard. The National SME Development Fund, a MIIT-backed policy instrument, entered this round. Investors: GSR Ventures (金沙江创投), Shanghe Dynamics (上河动量), Vertex Ventures (祥峰投资), and others.

OriginFlow (渊澈太初). Pre-A, hundreds of millions RMB, ¥1.5B (~$208M) valuation. Electromyography data infrastructure for training robot manipulation. The company captures electrical signals from human muscles during tasks like grasping and assembly, then converts those signals into training data for robot arms and hands. Investors: BlueRun Ventures (蓝驰创投), Vitalbridge (绿洲资本), and seven others.

RoboParty (萝博派对). Angel, tens of millions USD (~¥144M+), ¥1.95B (~$271M) valuation. Open-source humanoid robot. The founder is reportedly 21 years old. Xiaomi (小米) has reportedly backed the company across multiple rounds. Over $40M (~¥288M) raised pre-revenue for a company whose strategy is to publish its robot design openly, letting a global developer community build applications on top and train the robot’s AI with real-world footage from thousands of users. Investors: Shunwei Capital (顺为资本, lead), Xiaomi Group (小米集团).

DeepCybo (深度机智). Pre-A, hundreds of millions RMB, ¥1.5B (~$208M) valuation. PhysBrain 1.0, a system that trains robots from egocentric human video, meaning the robot learns by watching footage shot from a person’s perspective rather than requiring expensive real-robot demonstrations. Reportedly three rounds in three months and heavily oversubscribed. Investors: Zhongguancun Development Group (中关村发展集团), Puhua Capital (普华资本), Oriental Fortune Ocean (东方富海), and others.

InfiMaker (无限工坊). Series A, close to ¥100M (~$14M), ¥600M (~$83M) valuation. Consumer desktop CNC machines, the kind of precision cutting and carving tools that let hobbyists and small manufacturers work with wood, metal, and plastic at home. The founding team are reportedly DJI alumni. Backed by Meituan (美团) and Kunlun Capital (昆仑资本).

Brainsmart Technology (脑思科技). Angel, tens of millions USD (~¥72M+), ¥325M (~$45M) valuation. Brain-computer interface electrodes and neural decoding systems. One of three BCI companies that raised in this week’s data. See The Bigger Picture. Investors: Kangju Capital (康君资本, co-lead), BlueRun Ventures (蓝驰创投, co-lead), Fosun Health Capital (复健资本).

The Bigger Picture

China Approved the World’s First Commercial Brain Implant, and the Capital Followed

Three brain-computer interface companies raised funding in this week’s data, the downstream effect of a regulatory and policy cascade that compressed months of dealmaking into a sprint.

On March 13, 2026, China’s National Medical Products Administration granted the world’s first commercial approval for an implantable brain-computer interface, a coin-sized wireless electrode array made by Neuracle Medical Technology (神络医疗) that reads signals from the brain’s motor cortex and restores hand movement in paralyzed patients. Two days later, the National Healthcare Security Administration assigned the product an insurance reimbursement code, opening the payment pathway that had blocked hospital adoption. Approval on a Monday, insurance coding by Wednesday. Brain-computer interfaces were reclassified from research bet to healthcare market in a single week.

In August 2025, the Ministry of Industry and Information Technology and six other agencies published a national BCI roadmap targeting major technical breakthroughs by 2027 and a complete domestic supply chain by 2030. In March 2026, the Government Work Report explicitly named BCI as a “future industry” to be cultivated alongside quantum computing and 6G. By Q1 2026, the sector had recorded 17 deals totaling ¥3.8B (~$528M), already exceeding the total deal value for all of 2025 in a single quarter.

The global comparison is instructive. Neuralink has implanted more patients and its N1 chip goes deeper into the brain, using intracortical electrodes that penetrate brain tissue rather than sitting on the surface. Its Blindsight vision-restoration implant received FDA Breakthrough Device designation in 2024. But Neuralink remains in clinical trials with no commercial approval, no insurance pathway, and no hospital distribution channel. China’s NMPA moved first, the investment followed across three companies in one week, and the sector is accelerating.

The Component Layer: When the Picks-and-Shovels Companies Outprice the Miners

When you don’t know which gold miner will strike it rich, sell shovels. That logic is playing out across this week’s data in both robotics and AI infrastructure, and the market is pricing the shovel companies aggressively.

AGILINK makes the dexterous hands that go into humanoid robots from multiple manufacturers. Tanwei Chiplink builds the interconnect chips that let GPUs communicate inside AI training clusters. OriginFlow captures electrical signals from human muscles and converts them into training data for robot arms. Three companies, three different technical layers, one shared positioning: they supply critical components to an entire category rather than competing as a finished product.

AGILINK, shipping thousands of hands per quarter with an LG design win at CES, commands a ¥8B (~$1.1B) valuation. That sits above multiple humanoid robot body companies that are its customers. The logic is straightforward: if five humanoid makers each capture 20% of the market, a component supplier embedded across all five captures 100%. AGILINK’s CES partnership with LG proves the strategy works internationally, not just within China’s robotics supply chain.

Tanwei Chiplink applies the same logic one layer deeper. Nvidia controls the AI training stack not just through GPUs but through NVLink and NVSwitch, the high-speed interconnect fabric that lets those GPUs coordinate. Chinese AI chip companies can build competitive processors, but without a domestic interconnect layer, every Chinese AI cluster still depends on Nvidia’s communication backbone. Tanwei’s ¥1.5B (~$208M) valuation at Pre-A, with Inspur investing strategically, prices it as AI sovereignty infrastructure.

OriginFlow sits at the data layer. Training a robot to grasp objects, fold towels, or assemble components currently requires either expensive physical demonstrations with real robots, or simulation environments that don’t fully capture how human hands actually move. OriginFlow’s approach, strapping sensors to human arms and recording the electrical signals muscles produce during real tasks, creates a scalable data pipeline that any robotics company can use.

The proto supply chain visible in this week’s data reads: data acquisition (OriginFlow) to compute scheduling (Approaching.AI) to manipulation hardware (AGILINK) to finished robots. Capital is concentrating at each layer independently rather than flowing to vertically integrated robot companies, a pattern that echoes how the smartphone supply chain matured: TSMC, Qualcomm, and Corning became more valuable than most of the phone brands they supplied.

Deal data compiled from Chinese corporate announcements, financial news, and investment databases. Amounts and valuations are as reported and may be incomplete. For informational purposes only. Not investment advice. Verify all figures independently before acting on them.

Where capital flows tells you where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

Where the Money Went

Last week’s 46 deals reflected the May Day holiday. This week brought 103 across seven sectors, confirming the Q2 2026 funding cycle is fully underway.

103 deals, 28 with disclosed amounts. Those 28 disclosed a combined $3.2B (~¥23.1B), dominated by StepFun’s ~$2.5B Pre-IPO, which alone accounts for 78%. Strip that out and the remaining 27 deals disclosed ~$706M (~¥5.1B). The true weekly total is higher. Most deals don’t publish amounts, and the disclosed figure understates actual capital deployed.

Stage distribution skews heavily early. Angel and seed rounds make up the majority across every sector, with few Series A or B companies and a handful of mega-rounds at the top. The barbell shape that has characterized China’s deep tech funding all year is intact.

🤖 Robotics (23 deals)

The week’s largest deal count, and the most fragmented. No single round dominates the way StepFun does in AI. Vbot (维他动力) raised the headline ticket at ~¥500M (~$69M) Pre-A for a consumer robot dog, but the 23 deals cover consumer companion robots, industrial embodied AI, construction robots, lawn mowers, kitchen robots, wearable exoskeletons, and component suppliers. Fourteen of the 23 were angel or seed-stage, five Pre-A or A, and four strategic or B+ transactions. One straggler: ZeroErr (零差云控, robot system core components, C+, close to ¥100M, ~$14M).

🧠 AI (18 deals, including 3 stragglers)

StepFun at ~$2.5B Pre-IPO dwarfs everything else in the sector. 奕行智能 (EVAS, RISC-V AI chips) took a strategic investment from China Mobile’s chain fund; Motion Brain (眸深智能) raised ¥300M (~$42M) Pre-A for an embodied AI brain targeting humanoid robots; ComNergy Tech (共绩科技) raised close to ¥100M (~$14M) Pre-A for idle GPU compute aggregation. Three AI stragglers were sub-¥10M angel rounds.

🔬 Semiconductors (14 deals)

SemiDrive headlines with a $100M (~¥720M) Series C. China’s most-deployed automotive SoC maker is now pivoting toward humanoid robots. 象帝先计算 raised hundreds of millions in a Series B as its near-death 2024 story converts to an IPO-prep 2026 story. RiVAI Technology (睿思芯科) closed a B+ for high-end RISC-V server processors. Fermion Technology (费勉仪器) closed a C+ for precision scientific instruments used in semiconductor and quantum research. Five of the 14 deals have disclosed amounts; government-affiliated funds led or co-led at least eight.

⚡ New Energy & Materials (16 deals)

The largest sector by deal count after robotics. Lithium Source (常州锂源) took ¥440M (~$61M) strategic investment for LFP cathode expansion. Xinta Green Energy (信塔绿能) raised ¥100M (~$14M) at angel. Multiple early-stage deals in solid-state batteries, power electronics, carbon nanomaterials, and hydrogen water tech. Greater Bay Technology (巨湾技研), the ultra-fast charging battery maker, took an undisclosed strategic round.

🚀 Aerospace & Defense (8 deals)

Zero last week, eight this week. All angel or early stage, with a combined $14M in disclosed capital. Beijing Hengxing Power (恒星力量) raised ¥100M (~$14M) at angel for space energy research; Cathay Innovation and 厚雪资本 co-led. Shenzhen Zhiyu Aerospace (执宇航天) raised tens of millions (~$7M+) at angel for small reusable launch vehicles. Shenzhen Qingwei Aviation (庆为航空) closed a Pre-A for fly-by-wire avionics and smart cockpit displays. Vision Aero (维新宇航) raised tens of millions (~$7M+) at angel for a 7-seat eVTOL targeting emergency response. Four additional deals covered satellite propulsion, aerospace electrical interconnects, and microwave remote sensing.

⚙️ Other Advanced Manufacturing (15 deals)

公大激光 / Gongda Laser headlines with a Series C, hundreds of millions RMB, for its green fiber laser technology, more on this in Spotlight. Yiktong (亦唐科技) raised hundreds of millions in a B+ for high-speed SMT pick-and-place equipment. KeenCool (云酷智能) raised close to ¥100M (~$14M) for liquid cooling equipment for AI data centers. One straggler: 时耘科技 (intelligent robots, angel).

💡 Frontier Tech (9 deals)

微元合成 / Microcyto leads with ¥1.5B (~$208M), a synbio startup that used the round to acquire a ¥900M+/year traditional factory. 雷鸟创新 / Rayneo, the global #1 AR glasses brand by unit volume, closed a C+ round of over ¥1B (~$139M) in January 2026, captured in this week’s data. Two quantum computing startups, Hangzhou Wuwen Quantum (无问清芯) and GGQuanta (国光量子), and a photonic computing chip startup, PhotonCore (光子芯力), raised angel rounds. Genlight (佳量脑科学) shows a C round at an implied ¥60B (~$8.3B) valuation and has no verifiable public presence.

Spotlight

StepFun (阶跃星辰)

Pre-IPO, ~$2.5B (~¥18B), valuation undisclosed

China’s most aggressive LLM fundraiser is heading for Hong Kong, and the companies betting on it build the phones.

StepFun builds large language models, the AI systems behind tools like ChatGPT and DeepSeek, developing everything from the training infrastructure to the final products. Founder Jiang Daxin (姜大昕) spent 16 years at Microsoft Research Asia, the Beijing lab that trained a generation of China’s AI researchers, before leaving in 2023 to start the company. He holds a PhD from SUNY Buffalo and spent a year as an assistant professor at Nanyang Technological University in Singapore before joining Microsoft. His 2008 best application paper at SIGKDD, one of computer science’s top data mining conferences, on query suggestion was one of the earliest works showing that user behavior data could improve search in ways no engineer could manually design. StepFun’s core products include Step-2, a language model with over a trillion parameters (the internal connections that determine what a model knows and how well it reasons), using a mixture-of-experts design that routes each query to the most relevant specialist sub-networks rather than running every parameter on every question, dramatically reducing cost; Step-Video-T2V, a 30-billion-parameter video generation model that was the best-performing in its class at its February 2025 release; and Step-Audio, a 130-billion-parameter model that handles the full speech pipeline from understanding spoken language to generating natural speech. The company publishes its model code freely on developer platforms, letting anyone download and build on them to lock in developer adoption before the IPO.

The Pre-IPO round closed at approximately $2.5B (~¥18B), with a filing target for Hong Kong by end of June and a year-end listing expected. In January 2026, the company raised a ¥5B+ B+ round led by Shanghai State Capital Pioneer Fund (上国投先导基金), with Tencent following on. Yin Qi (印奇), who co-founded facial recognition company Megvii (旷视科技) and navigated it through a multi-year sanctions saga to a Hong Kong listing, became chairman that same month, suggesting StepFun is treating the IPO as its primary near-term mission. Huaqin Technology (华勤技术) and Longcheer Technology (龙旗科技), the contract manufacturers that design and build most of the world’s Android phones for brands like Samsung and Xiaomi, both invested in the Pre-IPO. ZTE, a major Chinese telecom infrastructure company, also joined and has already embedded StepFun models in mass production inside the Nubia Z80 Ultra. All three manufacture or assemble the devices that would distribute StepFun’s models at scale. If StepFun’s AI ends up embedded in hundreds of millions of handsets, these companies want equity exposure before the IPO.

StepFun’s Pre-IPO, with smartphone ODMs as investors, reflects a broader sorting in China’s LLM cohort. Companies that can scale toward public listings are moving to do so, while those that can’t are pivoting to narrower applications or device-level partnerships.

Investors (Pre-IPO): Huaqin Technology (华勤技术), Longcheer Technology (龙旗科技), ZTE (中兴通讯), and others. B+ round (January 2026): Shanghai State Capital Pioneer Fund (上国投先导基金, lead), Tencent.

Announcement

SemiDrive (芯驰科技)

Series C, $100M (~¥720M), ~¥13B (~$1.8B) valuation

China’s most widely deployed automotive chip maker closes its C round and starts talking to humanoid robot OEMs.

SemiDrive makes the chips that run China’s car cabins, gateways, and control systems. Its X9 series handles what the auto industry calls intelligent cockpit: the dashboard screens, voice assistants, and multi-display environments inside modern cars. Its G9 series manages vehicle gateway functions, routing signals between different vehicle subsystems. Its E3 microcontroller handles smart control tasks like power management and sensor coordination. Together, the product family covers the vehicle’s full electrical and electronic architecture, the hardware spine that connects every sensor, screen, and control system in a modern car. The company has shipped 12 million units, built 100+ mass-production vehicle models into its customer base, and holds relationships with over 260 customers, including 90%+ of China’s domestic automakers and a selection of international automakers. It was founded in 2018.

The Series C was led by Suzhou State Investment (苏产投), with Beijing Yizhuang National Investment (亦庄国投), 京国瑞, and three other investors co-investing. Shaanxi Heavy Truck (陕汽集团) subsidiary 陕汽鸿德投资 joined as a new strategic shareholder, a truck maker integrating up the supply chain. The round values the company at approximately ¥13B (~$1.8B). SemiDrive’s D-series chips are now targeting humanoid robot applications. Automotive-grade chips, tested for vibration, temperature extremes, and long-life reliability in ways consumer electronics chips are not, are a natural fit for industrial robots that operate in harsh real-world environments. The company that supplies cockpit silicon to 90%+ of China’s car makers is positioning for the same role in the robot market.

Investors: Suzhou State Investment (苏产投, lead), Yizhuang National Investment (亦庄国投), 京国瑞, Beijing Municipal Advanced Manufacturing Fund (北京市先进制造基金), Xi’an Finance Group (西安财金), 益中亘泰. Financial advisor: 多维资本.

Announcement

Gongda Laser (公大激光)

Series C, hundreds of millions RMB (~$28M+), ¥1.5B (~$208M) valuation

The company that makes China’s most powerful green fiber lasers just found a second market in the cooling systems inside AI data centers.

Gongda Laser makes short-wavelength fiber lasers, specifically green lasers, which operate at wavelengths that infrared fiber lasers cannot reach. Green light is what you need when you’re welding copper, cutting lithium battery electrodes, or processing silicon in semiconductors, because copper and silicon absorb green wavelengths far more efficiently than infrared. Founded in 2019 by Zhang Fan (张帆), a PhD with 16 years in fiber laser research, the company claims over 80% of the domestic green fiber laser market, a figure repeated in trade press interviews. Per its Series C announcement, it has developed a 4kW continuous green fiber laser, enough power to cut and weld industrial copper at production speed, a level that no known Chinese or Western competitor has reached in this wavelength and form factor. (IPG Photonics, the global fiber laser leader, tops out at around 100W in green; Trumpf offers 3kW at green wavelengths via a disk laser, a different architecture at a different price point.) Revenue exceeded ¥100M in 2023.

Gongda’s sub-company 希禾增材 (Xihe Additive) uses copper-based metal 3D printing to manufacture cooling structures and heat spreaders for AI data center hardware. Copper conducts heat better than aluminum, but it’s harder to process, which is where a high-power green laser becomes essential.

The company has deployed over 100 copper 3D printing units commercially. 禾创致远, affiliated with Zhongji Innolight (中际旭创), Nvidia’s primary manufacturer of optical transceivers, the components that handle high-speed data transmission between servers in AI data centers, invested alongside 华胥基金 (affiliated with Sany Heavy Industry / 三一重工). Dazheng Capital (达晨财智) and Photosynthesis Ventures (光合创投) co-led.

Investors: Dazheng Capital (达晨财智, co-lead), Photosynthesis Ventures (光合创投, co-lead), 禾创致远 (Zhongji Innolight / 中际旭创 affiliated), 华胥基金 (三一集团 affiliated), 力合科创. Huachuang Capital (华创资本) continued from prior rounds.

Announcement

Microcyto (微元合成)

New financing round, ¥1.5B (~$208M), valuation undisclosed

A synthetic biology startup used its ¥1.5B financing round to acquire a sugar alcohol factory generating nearly ¥1B in annual revenue.

Microcyto uses microbial fermentation to manufacture bio-based compounds, essentially programming microorganisms to act as chemical factories, producing sugar alcohols, natural pigments, and functional ingredients from biological feedstocks rather than petroleum or chemical synthesis. Founded in December 2021 by Dr. Liu Bo, who did his doctorate at the CAS Institute of Microbiology, the company has developed a one-step fermentation pathway to allulose, a rare sugar with minimal calories, that China’s National Health Commission approved for commercial sale in 2025, making it the first company in China with regulatory clearance for this production method. Its fermentation process for mannitol, a sugar alcohol widely used in food and pharmaceuticals, converts 99% of the raw input material into product at 99.9% purity.

Part of the ¥1.5B (~$208M) financing is long-term bank debt from Xingye Bank’s (兴业银行) Beijing branch. Banks size credit facilities against physical assets and revenue, so the loan itself is a form of due diligence on the factory being acquired. The money is buying Yusweet (豫鑫糖醇), an Anyang, Henan manufacturer of xylitol, erythritol, and related sugar alcohols operating since 1996 that generated approximately ¥919M in revenue in 2024. Per the company’s announcement, Yusweet is among the world’s leading xylitol producers. Yusweet chairman Zhang Qibin (张其宾) co-invested in Microcyto as part of the deal, becoming a shareholder in the company buying his factory and aligning incentives across the transition.

Synthetic biology companies in China have historically either built new factories from scratch or partnered with contract manufacturers for pilot batches. Microcyto is acquiring an existing industrial asset and upgrading it with bio-manufacturing technology. The factory provides immediate scale; the technology provides margin improvement and the ability to expand into new products that chemical synthesis cannot access.

Investors: Zhang Qibin (张其宾, Yusweet chairman), Xingyin Investment (兴银投资, Xingye Bank’s ¥100B AUM investment subsidiary), Tan Ruiqing (谭瑞清, former Longbai Group vice chairman, 30+ years in chemical manufacturing).

Announcement

Vbot (维他动力)

Pre-A, ~¥500M (~$69M), valuation undisclosed

The ex-Horizon Robotics team that built autonomous driving chips for cars is now shipping autonomous robot dogs to consumers, having just raised the largest single ticket in Chinese consumer robotics history.

Vbot makes consumer robot dogs. The founding team assembled three of the people who built Horizon Robotics, the company that supplies AI chips to Chinese automakers: CEO Yu Yinan (余轶南) was Horizon’s VP and a founding member since 2015; CTO Song Wei (宋巍) was chief architect of Horizon’s software platform; co-founder Zhao Zhelun (赵哲伦) was head of intelligent driving products at Li Auto. The trio left and applied autonomous driving engineering discipline to a product for the home. The Vbot大头BoBo, their first product, runs 128 trillion operations per second of onboard computing power, a level normally found in self-driving car systems rather than consumer gadgets, and does not require a remote control, using vision and AI to navigate independently.

The company was founded in late 2024 and is already shipping. The first 500 units came off the production line and into customers’ hands on May 8th. May’s target is 1,500 total; June’s production target is 2,500+ units per month. The Pre-A of close to ¥500M (~$69M) is, per the company and multiple Chinese tech publications, the largest single financing ticket raised by any consumer robotics company in China. The round had eleven institutional investors, including investors who were present at the seed round continuing with conviction. Xinhua, China’s state news agency, published an editorial this week titled “Capital Accelerates Into Embodied AI, Delivery Capability Becomes Valuation Anchor” (资本加速涌入具身智能，交付能力成估值锚点), with Vbot’s round and shipping timeline as a central example.

The company is separately building a full-size humanoid robot, led by newly hired VP Qin Hailong. The software and hardware architecture overlaps significantly with the robot dog, and the AI stack is designed for reuse across form factors. Deployment data from the robot dog fleet feeds directly into humanoid development. More units in the field means more real-world situations to learn from, the same compounding data advantage that powers Tesla’s self-driving improvements.

Context on the market: Unitree (宇树科技), the domestic leader in quadruped robots, shipped approximately 23,700 units in 2024, representing roughly 69.75% of the global market by unit volume, per GGII, a Chinese robotics industry research firm. Vbot is entering a market that Unitree currently dominates, at a consumer price point (sub-¥10,000, or under $1,400) designed to expand the category rather than compete directly for Unitree’s existing industrial and research customers.

Investors: Dongfang Jiafu (东方嘉富, lead), Huatai Zijin (华泰紫金, lead), Fosun Ruizheng (复星锐正), SAIC-affiliated Shangqi Capital (尚颀资本), 明荟致远. Continuing from prior rounds: Hillhouse Creative Ventures (高瓴创投), Cathay Innovation (凯辉基金), Today Capital (今日资本), BV Baidu Ventures (百度风投), Corsair Capital (渶策资本), 柏睿资本. Financial advisor: Kaihui Capital (高鹄资本).

Announcement

Also on the Radar

• Rayneo (雷鸟创新) — C+, >¥1B (~$139M+), ¥8B (~$1.1B) valuation. Augmented reality glasses that overlay digital information onto the wearer’s real-world view. Global #1 by unit volume (per Counterpoint Research 2025: ~32% global market share, 3.8x overseas sales growth in 2025, Time Magazine Best Inventions 2025 Special Mention). China Mobile (中国移动链长基金) and China Unicom (中国联通联创创新基金) co-invested, the first time two state telecoms have jointly backed the same smart glasses company.

• EVAS / 奕行智能 — Strategic, hundreds of millions RMB (~$28M+), ¥10B (~$1.4B) valuation. AI chips built on RISC-V, an open-source chip architecture that frees Chinese chipmakers from US licensing dependencies. Epoch series in mass production since January 2026. China Mobile’s industry chain fund, which invests in companies building out its supply chain, made the strategic investment this week, following a Series B led by four Beijing government AI funds earlier in 2026.

• Ola Dimensions (欧拉万象) — Angel, hundreds of millions RMB (~$28M+), ¥1B (~$139M) valuation. Founded March 2026. Two rounds in 60 days. Sole founder Zhou Shunbo: former Huawei “genius recruit” (华为天才少年), company-level chief expert in robotics, first Chinese recipient of a best paper honor from IEEE, the world’s largest engineering professional society, in robotics and automation. A deliberate counter-thesis to the perfectionism in consumer robotics: ship an imperfect version one, collect real-world data, iterate rapidly. Two rounds backed by China Merchants Group Ventures (招商局创投) and SAIF Partners (赛富投资基金), among others.

• Lumos Robotics (鹿明机器人) — A1+A2 (announced together), hundreds of millions RMB (~$28M+), ~¥3B valuation. Mitsubishi Electric China led both consecutive rounds. The company’s FastUMI Pro, a tool that captures the movement data robots need to learn tasks, cuts training data collection time by 5x and cost by 80% according to the company, and is claimed to be in use by two-thirds of global robotics AI research labs. A Japanese industrial giant securing access to Chinese robotics AI training tools through back-to-back equity stakes.

• XDXCT / 象帝先计算 — Series B, hundreds of millions RMB (~$28M+), ¥9B (~$1.25B) valuation. A domestic maker of high-performance GPUs, the processors that train and run AI models. Near-death in 2024: accounts frozen in August, hundreds of employees laid off, an investor protection clause triggered that penalizes the company for missing performance targets. Rescued by a new investor (Anfu Technology) in February 2025. Now: stock restructuring planned for H2 2026, IPO advisor signed. The 40% valuation haircut from peak was the price of survival.

• Fangshi Robot (方石机器人) — Series A, close to ¥100M (~$14M), ¥500M (~$69M) valuation. Construction robots: spray-coating, floor tile laying, ground leveling. Seven years old. 500+ projects completed, 15M+ square meters covered. Deployed in 10+ countries across the Middle East, Southeast Asia, and East Asia. Fortune 500 construction companies as customers. Only now taking its first institutional round.

• Picocom (比科奇微电子) — C+, ¥100M (~$14M), ¥500M (~$69M) valuation. Chips that handle the core signal processing in 5G cell towers, built for open radio access networks, an architecture that lets mobile operators mix equipment from different vendors instead of buying locked-in systems from a single supplier. President Peter Claydon’s prior company Picochip dominated the market for small cell base stations, the compact antennas that extend mobile coverage inside buildings and dense urban areas, holding approximately 70% global market share before its $75M acquisition. Beijing and Bristol R&D centers. Now expanding into broadband terminals for low-earth orbit satellite networks like Starlink and China’s equivalent constellations.

• Changyao Innovation (长曜创新) — A+, tens of millions RMB (~$7M+), ¥150M (~$21M) valuation. Lawn mower robots that navigate without buried boundary wires, using six cameras and a depth sensor. Its AIRSEEKERS TRON raised $2.2M from approximately 1,400 backers on Kickstarter in 30 days. US and Europe offices established. Midea-system investor Yingfeng Environment (盈峰环境) led this round. Born on Kickstarter, sold to Western buyers before any domestic launch narrative.

• Greater Bay Technology / GBT (巨湾技研) — Strategic, undisclosed, ¥15B (~$2.1B) valuation. Batteries that can fully charge in 5-15 minutes while storing enough energy per kilogram to power a full-size passenger car, not just a scooter. Production customers span GAC’s passenger lineup (Aion, Trumpchi, Hycan), commercial truck makers including Shaanxi Auto and FAW Jiefang, and EHang’s electric vertical-takeoff aircraft. Five-region manufacturing strategy underway.

• Motion Brain (眸深智能) — Pre-A, ¥300M (~$42M), ¥1.8B (~$250M) valuation. AI software that gives humanoid robots the ability to generate fluid movements, understand the 3D space around them, and run complex AI models directly on the robot’s own hardware rather than relying on cloud servers. Founded by Professor Chen Tao of Fudan University’s Deep Learning Lab and Zhang Yimin, former Chief Scientist at Intel China.

Stragglers

One straggler cleared the notable bar: ZeroErr (零差云控) (robot system core components, C+, close to ¥100M, ~$14M, backed by 华控基金, 同创伟业, and 国泰海通), a component supplier positioning for the robot supply chain buildout.

The Bigger Picture

State Capital Is the Lead Investor in China’s Hardware Sovereignty Push

Government guidance funds, the state-backed investment vehicles China uses to direct capital into strategic industries, usually appear in venture deals as co-investors alongside private firms, filling gaps rather than setting terms. This week, state capital led or co-led at least 30 of the 103 transactions. In semiconductors and aerospace specifically, government funds were the named lead investor in nearly every deal with a disclosed lead.

SemiDrive’s Series C was led by Suzhou State Investment (苏产投). EVAS (RISC-V AI chips) was led, in its prior B round, by four Beijing government AI funds simultaneously, with China Mobile’s chain fund adding a strategic this week. Fangshi Robot’s Series A included AECC Aviation Fund (航发基金), the investment arm of China’s state aero-engine corporation, and Beijing Science & Technology Innovation Fund (北京科创基金). Of the eight aerospace deals this week, most involved government-affiliated funds as leads or sole investors.

In December 2025, Beijing announced a national venture capital guidance fund seeded with ¥100B (~$14B) in central government capital, structured to mobilize ¥1T (~$138B) over 20 years through 600+ sub-funds targeting integrated circuits, quantum technology, aerospace, and biomedicine with a mandate to put at least 70% into seed and early-stage companies, per the NDRC launch announcement and reported by Caixin. Five months later, this week’s deals are early evidence of that system deploying.

The broader system has been building for a decade. China’s government guidance fund network now comprises over 2,000 funds with a confirmed total of more than $500B raised. State capital now provides the majority of the money backing China’s venture and private equity funds. In effect, the government is the biggest investor in the investors. The semiconductor-specific vehicle, the “Big Fund” (国家集成电路产业投资基金), has escalated in each phase: ¥98.7B in 2014, ¥204B in 2019, ¥344B (~$47.5B) in its third phase launched May 2024, now fully operational with Phase III focused specifically on equipment and materials, the parts of the chip supply chain most vulnerable to US export controls.

SemiDrive is the domestic alternative to NXP and Renesas in automotive chips. EVAS is building AI chips on an open architecture that no US company can sanction. Fangshi Robot deploys into construction sites across the Middle East and Southeast Asia. In each case, government capital is backing companies that reduce Chinese dependence on foreign suppliers.

China’s Big Model Endgame: The Device-Embedded AI Race

The framing that has dominated analysis of China’s large language model startups, often called the “six small tigers” (六小虎), is no longer useful. It describes a 2024 world that has already reorganized.

The company raised approximately $2.5B (~¥18B) at a valuation targeting around $10B for its Hong Kong IPO, expected before year-end. Huaqin Technology and Longcheer Technology, the two largest contract manufacturers of Android smartphones globally, joined the Pre-IPO round. ZTE, which builds China’s domestic telecom infrastructure, also invested and has already embedded StepFun models in mass production inside the Nubia Z80 Ultra.

The chatbot and API subscription model that defines Western LLM revenue is a difficult path in China. The domestic market for consumer AI subscriptions is competitive and price-sensitive. The more durable model, and the one that smartphone manufacturers are betting on, is embedding AI at the device level, where models run directly on the phone’s own processor and manufacturers pay for licensing or co-development rather than end-users paying per query. Frost & Sullivan estimates the global on-device AI market at ¥1.2T (~$167B) by 2029, growing at 39.6% per year. Vivo, which shipped 103.9 million phones in 2025 per IDC, has already built a 3-billion-parameter AI model running directly on the phone’s own processor into its device lineup; ByteDance’s Doubao AI assistant runs as part of the phone’s operating system rather than as a standalone app, making it available across everything the user does on the device.

The smartphone manufacturers investing in StepFun’s Pre-IPO make the devices that would embed its models at scale. Equity exposure before a successful IPO is cheaper than negotiating licensing terms after one.

The cohort, as it now stands: Zhipu AI listed in Hong Kong in January 2026. MiniMax is IPO-bound. StepFun is filing by end of June. Kimi raised $2B at a $20B valuation last week. Baichuan AI has pivoted to medical AI. 01.AI has contracted its pre-training team. Benchmark leaderboard competition has given way to a race for device distribution and manufacturer licensing.

Synthetic Biology’s Industrial Graduation

Synthetic biology has spent a decade in China promising to replace petroleum chemistry with biological manufacturing. Most of that promise has lived in pilot plants, research institutes, and funding decks. The Microcyto deal this week suggests the sector may be crossing that gap, through acquisition of existing industrial infrastructure.

The ¥1.5B (~$208M) financing includes long-term bank debt from Xingye Bank’s Beijing branch. Banks size credit facilities against physical assets and revenue. For the loan to clear, the fermentation tanks, processing lines, and distribution relationships of 豫鑫糖醇 had to stand as collateral.

Yusweet (豫鑫糖醇) has been producing xylitol and related sugar alcohols in Anyang, Henan since 1996, with nearly ¥1B in annual revenue. Microcyto brings fermentation-process manufacturing at higher efficiency and purity than chemical synthesis, with the ability to expand into new products (allulose, rare sugars, functional ingredients) that traditional manufacturers cannot produce. Its one-step fermentation route to allulose received National Health Commission regulatory approval in 2025, the first in China. Yusweet chairman Zhang Qibin co-invested in Microcyto, taking equity upside from the technology transformation of his own company.

China’s biomanufacturing sector produces over 70% of global bio-fermentation output and generates approximately ¥1.1T (~$153B) in annual market value, growing at 30% per year since 2015. China’s 15th Five-Year Plan, the government’s top-level economic blueprint for 2026-2030, designates biomanufacturing as one of seven “future industry” priority tracks, the same tier as quantum technology and brain-computer interfaces. The government has funded 43 pilot plant projects nationwide and issued dedicated support programs in 20+ provinces.

For decades, legacy industrial companies acquired synthetic biology startups for technology access. Microcyto reversed the direction, buying the factory and bringing the biology to upgrade it.

Further context from Xinhua on China’s biomanufacturing industrial strategy, February 2026.

Deal data compiled from Chinese corporate announcements, financial news, and investment databases. Amounts and valuations are as reported and may be incomplete. For informational purposes only. Not investment advice. Verify all figures independently before acting on them.

The China Round is a weekly report tracking venture and strategic investment activity across China’s core technology sectors: AI, robotics, semiconductors, new energy and materials, aerospace and defense, and frontier tech.

Where capital flows tells you where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

Where the Money Went

Whether you were celebrating Labor Day, Cinco de Mayo, or just enjoying a long weekend, China’s five-day May Day holiday (May 1-5) kept most dealmaking on hold.

46 deals, $2.39B (~¥17.2B) in disclosed capital across nine of them. That number needs context: Kimi’s $2B D-round accounts for 84% of the total. Strip it out and 45 deals generated $390M (~¥2.8B). AI led on deal count at twelve, with robotics close behind at ten. No aerospace deals.

🧠 AI (12 deals)

One deal dominates the entire week: Kimi’s $2B (~¥14.4B) D-round at a ~$20B (~¥144B) valuation, the largest Chinese LLM startup round of 2026. Beneath that headline, two AI infrastructure companies raised a combined ¥1B+ (~$139M+) in the same week with identical “token factory” positioning: Infinigence AI at ¥700M+ (~$97M+) and Magik Compute at hundreds of millions. Nanochap took a ¥300M (~$42M) strategic round for biomedical chips targeting visual brain-computer interfaces, a medical device company tagged as “AI” on deal trackers. The remaining deals include PixelBloom’s C-round for its AiPPT product (30M+ registered users globally), a medical LLM angel (WiseDiag, ¥65M / ~$9M), a spatial intelligence platform strategic round (XGRIDS), and four late-reported deals spanning industrial cognitive AI, AI infrastructure cloud, and AI-driven drug discovery.

🤖 Robotics (10 deals)

RoboScience pulled the largest robotics round of the week at ¥1B (~$139M) for its VLOA embodied AI platform. The remaining main-window deals are earlier-stage: Eagle Vision raised a Pre-A for bionic flapping-wing flying robots with a consumer product targeting the global hobbyist market, Westlake Robotics for civilian quadruped robots, Knownobound for industrial embodied AI targeting ship maintenance and container logistics, and underwater robotics veteran Boya Gongdao closed a D+ round. Five late-reported deals brought the sector total to ten, including a Sequoia China-backed angel in smart robots and a strategic round in core robot components backed by iFlytek Ventures and Hillhouse.

🔬 Semiconductors (8 deals)

Eight deals spanning the full chip stack, from data center networking to automotive SoCs to photonic interconnects. Yunsilicon builds what it describes as China’s first 400Gbps RDMA data center networking chip, positioning it as the domestic alternative to Nvidia’s ConnectX; ByteDance and Inspur hold strategic equity stakes. Oritek closed a C-round for automotive zone controller SoCs, with SDIC Merchants participating in every round from B1 through C and eight listed auto companies as shareholders. Siminwei drew three top-tier VCs (Oriental Fortune Capital, GGV Capital, Detong Capital) to co-lead a seed at a ¥500M (~$69M) valuation for AI inference architecture, an unusually heavy syndicate for a seed. Rhinoptix raised for TFLN photonic chips, the post-copper datacenter interconnect technology covered in the April 10 issue. Sicred continues scaling SiC power devices with 100,000+ units shipped to aerospace and defense customers.

⚙️ Other Advanced Manufacturing (7 deals)

Tachin Tech raised a Pre-A for electronic skin targeting the humanoid robot market, with core technology from Zhenan Bao’s Stanford lab. Makera closed a Series A for consumer desktop CNC machines after raising $10.2M on Kickstarter. Three smaller deals: a smart lighting company in Shenzhen (Mingyuda, ¥30M HKD / ~$3.8M), a consumer electronics angel backed by Co-Creation Ventures, and a miHoYo-backed equipment investment. Two late-reported deals added a metal 3D printing angel and a high-speed SMT equipment strategic round.

⚡ New Energy & Materials (5 deals)

BYD invested directly in battery electrolyte materials maker Huasheng Xianghe (strategic, hundreds of millions RMB), continuing its vertical integration down the EV supply chain. Hebang New Materials raised a ¥100M (~$14M) A-round for bio-based composite materials certified to EU high-speed rail standards. Three late-reported deals added carbon-based energy storage materials and emerging energy tech to the tally.

💡 Frontier Tech (4 deals)

STARTORUS closed a ¥500M (~$69M) A+ for its fusion energy reactor, crossing the $1B (~¥7.2B) valuation mark. Greencloud, a subsidiary of listed Wangsu Technology, raised for immersion liquid cooling with partnerships across Intel, Dell, and BASF. Nanoloop took a seed for sports brain-computer interface wearables at ¥10M+ (~$1.4M+). One late-reported deal: Boyin Bio raised an A-round for biomaterials from Legend Capital, CAS Advanced Industry Fund, and Cornerstone.

Other sectors

No aerospace or defense deals this week.

Spotlight

Kimi (月之暗面)

Series D, $2B (~¥14.4B), ~$20B (~¥144B)

China’s best-funded independent LLM startup hits frontier parity and raises the largest Chinese AI round of 2026.

Kimi is an LLM platform built by Moonshot AI. K2.6, released April 2026, beats or matches GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro across coding, reasoning, and agent benchmarks, topping the field on Humanity’s Last Exam, DeepSearchQA, and SWE-Bench Pro. The K2.5 launch in January triggered a subscription surge of over 8,000% in a single month. Annual recurring revenue broke $100M (~¥720M) by March 2026 and exceeded $200M (~¥1.44B) by April. Founder Yang Zhilin studied at Tsinghua and CMU, worked at Google Brain and Meta AI, and co-authored Transformer-XL and XLNet, two foundational papers in the architecture behind modern LLMs.

Meituan Longzhu, the venture arm of China’s largest local services platform, led with over $200M (~¥1.44B). Meituan operates food delivery, ride-hailing, hotel booking, and grocery for 700M+ annual transacting users. If Kimi becomes the AI layer inside those services, that means agentic AI embedded in hundreds of millions of daily purchasing decisions. China Mobile’s innovation fund and CPE Yuanfeng co-invested, adding state telco infrastructure and financial capital alongside commerce, each representing a different distribution thesis for the same model. The valuation jumped from $4.3B (~¥31B) in December 2025 to $20B (~¥144B) in six months, with approximately $3.9B (~¥28B) raised in that sprint.

Meituan Longzhu (美团龙珠, lead), China Mobile Innovation Fund (中移创新产业基金), CPE Yuanfeng (CPE源峰), and Shuimu Capital (水木资 本).

Announcement

STARTORUS (星环聚能)

Series A+, ¥500M (~$69M), exceeded $1B (~¥7.2B+)

China’s latest fusion unicorn, building a reactor geometry no other company has attempted at commercial scale.

STARTORUS is building a compact fusion reactor: a spherical tokamak, which uses powerful superconducting magnets to contain hydrogen plasma heated to hundreds of millions of degrees inside a doughnut-shaped chamber. STARTORUS’s approach uses a plasma geometry called negative triangularity. In standard tokamaks, the edge of the plasma periodically erupts in instabilities that blast over 1 GW per square meter into the reactor wall, a problem that has plagued the field for decades. Negative triangularity reshapes the plasma cross-section to suppress these eruptions entirely, reducing wall damage and improving energy confinement. STARTORUS claims to be the first company building a commercial spherical tokamak natively around this geometry. (The University of Seville’s SMART research tokamak achieved first plasma with negative triangularity in January 2025, but as an academic device, not a commercial reactor.) The NTST device begins installation in September 2026 at the company’s Shanghai Jiading base, with the demonstration reactor 星环一号 (CTRFR-1) targeted for construction in 2028, design targets by 2029, and electricity output around 2032. Chief Scientist Tan Yi has spent over 20 years in fusion research since entering Tsinghua’s Department of Engineering Physics in 1999.

The A-round in January (¥1B / ~$139M, led entirely by Shanghai state capital) broke the domestic record for a single private fusion round. The A+ brought 11 new investors four months later, pushing cumulative funding past ¥2B (~$278M). At a valuation exceeding $1B (~¥7.2B+), STARTORUS joins NeoFusion (聚变新能) in a small club of Chinese private fusion unicorns, in a field where fusion energy generates zero watts commercially anywhere on Earth. Commonwealth Fusion Systems, TAE Technologies, and Helion have each raised more capital, but none uses negative triangularity. If NTST produces results in late 2026, STARTORUS enters a global race with technically differentiated physics.

Dachen Finance (达晨财智), Jinpu Investment (金浦投资), Shenneng Chengyi (申能诚毅, Shenergy affiliate), Sunshine Ronghui Capital (阳光融汇资本), Industrial Securities Capital (兴证资本), Songqing Capital (松青资本), Hongtai Fund (洪泰基金), Sanyuan Capital (三元资本), Jingming Capital (京铭资本), Yinggang Capital (盈港资本), and Feitu Ventures (飞图创投, returning from A-round).

Announcement

Makera (造物时代)

Series A, hundreds of millions RMB, ¥1.5B (~$208M)

$10.2 million on Kickstarter in 45 days. A Chinese consumer hardware brand built for the global maker market from day one.

Makera builds consumer desktop CNC machines that bring precision machining to workshops, research labs, and home studios. CNC machines carve, cut, and engrave materials using computer-controlled toolpaths; industrial units cost $50,000 or more. Makera puts comparable capability on a desktop for under $1,000. The Z1, launched on Kickstarter in late 2025, raised $10,245,134 from 6,927 backers in 45 days.

Its earlier products, Carvera and Carvera Air, held the top two spots among CNC campaigns on Kickstarter before the Z1 surpassed both. Products are used at MIT’s Center for Bits and Atoms and, per the company, at over 1,000 enterprise and university customers including Apple, Toyota, and Volkswagen. Full-year 2025 revenue grew over 400% and Q1 2026 is tracking nearly 300% year-on-year, per the company’s Series A announcement.

The Series A came less than eight months after Qiming Venture Partners’ Pre-A, with Qiming investing well above its proportional share of the round, a strong conviction signal from one of China’s top VCs. The co-lead is Beijing’s Municipal AI Industry Investment Fund, an unusual backer for consumer hardware, suggesting the fund sees the intelligent CAM software layer as part of its AI mandate. Makera was never a “China company going global.” It was born on Kickstarter, priced in dollars, and validated by Western makers before any domestic narrative. In a week dominated by AI infrastructure and robotics, Makera is the clearest evidence that China can build global consumer hardware brands, not just supply chains.

Meridian Capital (华映资本, co-lead), Beijing Municipal AI Industry Investment Fund (北京市人工智能产业投资基金, co-lead), Oriza Puhua (元禾璞华), CASIC Ventures (中科创星), TsingVenture (清科创投), and Qiming Venture Partners (启明创投, returning).

Announcement

Infinigence AI (无问芯穹)

Series B, ¥700M+ (~$97M+), ¥7B (~$970M)

One of China’s best-funded AI infrastructure startups, betting that the token production layer becomes the durable moat in AGI.

Infinigence AI builds inference and training infrastructure that runs across multiple chip architectures simultaneously. Where most AI companies optimize for Nvidia GPUs, Infinigence orchestrates workloads across AMD, Huawei Ascend, Moore Threads, and seven other domestic chip families, turning heterogeneous hardware into a unified compute pool. The platform hosts over 160 models and serves enterprise clients whose daily token call volume has grown 20x year-on-year. The Tsinghua NICS lab spinout positions itself as a “token factory,” a framing its CEO used at a March 2026 industry forum: converting raw compute into AI tokens the way a refinery converts crude into usable fuel. Cumulative funding now exceeds ¥2.2B (~$306M), with 15 investors in this round alone.

The thesis is that AI models will commoditize, as DeepSeek’s open-source success suggests, but the logistics layer (scheduling across fragmented chips, optimizing for cost and latency, routing traffic between models) will not. This is the AWS argument applied to AGI: the largest cloud provider did not build the most applications; it built the best infrastructure for running them. The risk is equally visible. China’s hyperscalers (Alibaba Cloud, Baidu Cloud, Huawei Cloud) run their own inference platforms. ByteDance runs Doubao inference internally. The token factory thesis works for the long tail of agent startups and enterprise AI deployments. Whether it works against vertically integrated giants is the open question that ¥2.2B in cumulative capital is now testing.

Hangzhou High-Tech Capital (杭州高新金投, co-lead), Huiyuan Capital (惠远资本, co-lead), Guoxing Capital (国兴投资), Qinhuai Data (秦淮数据), and others. Returning: Legend Capital (君联资本), Shanghai Guotou Futeng (上海国投孚腾), Yuanzhi Future (元智未来).

Announcement

RoboScience (机科未来)

Series A, ¥1B (~$139M), ¥5B (~$694M)

Founded in late 2024, Series A by May 2026 with no shipping hardware. The same architectural bet as Physical Intelligence in the US.

RoboScience is building an embodied AI platform centered on VLOA (Vision-Language-Object-Action), a model that learns to control robots the way language models learn language: absorb massive amounts of visual and manipulation data, then generalize to tasks it has never seen. Instead of programming a robot arm to pick up a specific object with a specific motion, you show it millions of examples and let it figure out the movements itself. The approach mirrors Physical Intelligence’s Pi model in the US. The company has accumulated over one million hours of manipulation data, growing at tens of thousands of hours per week. Co-founder Tian Ye was Apple’s AI platform technical lead after earning his MS at Stanford under Andrew Ng. Co-founder Lin Shao is an NUS assistant professor with a Stanford AI Lab PhD whose D(R,O)Grasp framework won the Best Paper Award for Robot Manipulation at ICRA 2025, the field’s top international conference.

¥1B for a company roughly 18 months old with no shipping hardware is aggressive, but within the norms of China’s embodied AI funding cycle; Galbot raised a comparable amount at a similar stage. The unnamed investor list, described only as “domestic and international industry giants and top financial institutions,” is its own signal: in China’s AI market, investor anonymity usually means either strategic sensitivities or credentials that don’t match the framing. Early investors include JD.com (angel lead, a company that operates one of China’s largest logistics networks), Merchants Capital, and SenseTime’s venture arm. JD.com’s involvement points to the deployment thesis: last-mile logistics as the proving ground for general-purpose manipulation. The question is whether VLOA’s data-scale approach produces generalizable robot intelligence faster than the task-specific approaches used by companies already shipping thousands of units.

JD.com (京东, angel lead), Merchants Capital (招商局创投), SenseTime Guoxiang Capital (商汤国香资本), Zero One Ventures (零一创投), Puhua Capital (普华资本, Pre-A lead), and others. Series A investors undisclosed.

Announcement

Also on the Radar

Tachin Tech (途见科技) — Pre-A + Pre-A+, hundreds of millions RMB, ¥600M (~$83M)

Electronic skin for robots. 400 sensors per square centimeter detecting pressure, temperature, shear force, texture, and proximity. Founder Lai Jiancheng was a Stanford postdoc in Zhenan Bao’s lab, one of the world’s leading groups in flexible electronics. The haptic data collection glove is the clever wedge: sell the training data tool (human demonstrations converted to robot learning data) before the full-body robot skin market matures. Lenovo Star (联想之星) leads; Huakong Fund (华控基金) and Oriza Puhua (元禾璞华) co-invest.

Announcement

Yunsilicon (云脉芯联) — B-round, hundreds of millions RMB (~$42M+), ¥7B (~$970M)

The YSA-100, which the company describes as China’s first 400Gbps RDMA data center networking chip, has been in mass production since 2024 and positions the company as a domestic alternative to Nvidia’s ConnectX series. CEO Liu Yongfeng previously led networking chip development at Cisco, Alibaba, and Huawei. ByteDance (8.42%) and Inspur (6.86%) hold strategic equity stakes. The reported investor list includes the National IC Fund, Shanghai IC Fund, and National AI Industry Fund. Three national-level semiconductor and AI funds in a single chip startup would be unusual; worth watching for public confirmation.

Oritek (欧冶半导体) — C-round, hundreds of millions RMB, ¥1.5B (~$208M)

Automotive SoC chips for third-generation electrical/electronic architecture, the “central computing + zone control” design that the entire global auto industry is migrating toward. Oritek’s Gongbu series zone controller chips are among China’s first domestic entries in this segment. Eight listed auto-industry companies are shareholders, including SAIC, Xingyu, AAC Technologies, and Sunny Optical. SDIC Merchants has participated in every round from B1 through C.

Announcement

Nanochap (暖芯迦) — Strategic, ¥300M (~$42M), ¥1.5B (~$208M)

Biomedical chip company developing visual brain-computer interfaces competing with Neuralink’s Blindsight. R&D center in Melbourne, 13 international patents, and an EPC001 chip that reads ECG, PPG, EEG, and EMG signals simultaneously. Tagged “AI” on deal trackers but this is a medical device company at its core. The team holds dual degrees across electrical engineering and biomedicine, with veterans from major semiconductor companies.

Legend Capital (君联资本, lead), Guangfa Xinde (广发信德).

PixelBloom / AiPPT (像素绽放) — C-round, undisclosed, ¥250M (~$35M)
AI presentation generation tool with 30M+ registered users across 180+ countries, ranked #4 globally in the AI PPT vertical and #1 among startups in China. Founded in Singapore in 2018, making it a genuinely global-first product rather than a China-domestic pivot. The modest C-round valuation suggests either an undervalued global product or a structured round.

CASIC Capital (国科投资, co-lead), Guoxiang Capital (国香资本, co-lead), Cornerstone Ventures (基石创投).

Rhinoptix (犀里光电) — Pre-A, tens of millions RMB, ¥150M (~$21M)

Thin-film lithium niobate (TFLN) photonic chips for data center optical interconnects. Founder Prof. Wang Cheng co-invented TFLN electro-optic modulators at Harvard and holds a Tsinghua undergrad. TFLN switches light signals roughly 10x faster than silicon photonics, the leading candidate for post-copper datacenter connections as AI clusters push past 1.6 terabits per second. Founded January 2025 and already raising, a pace that reflects strong VC confidence in the technology.

Lenovo Ventures (联想创投), Oriza Origins (元禾原点).

Sicred (至信微电子) — B-round, undisclosed, ¥150M (~$21M)

SiC MOSFETs (silicon carbide power switches that handle high voltage more efficiently than traditional silicon) for aerospace, defense, and industrial power conversion. 100,000+ units shipped at 1200V/7mΩ with over 90% yield. Founder Zhang Aizhi previously worked at STMicroelectronics and Huarui Micro, bringing 20+ years in power device design. SiC is a critical technology for global EV and industrial electrification, and Sicred’s shipping volume at B-round stage is unusual.

China Venture Capital (中国风投), Tianhu Fund (天互基金).

Greencloud (绿色云图) — A-round, undisclosed

Immersion liquid cooling for AI data centers, claiming a PUE of 1.049, meaning 95% of energy goes to compute and only 5% to cooling. Subsidiary of listed Wangsu Technology (603000.SH). Strategic partnerships with Dell (2023), Intel co-development (2024), and a BASF joint lab for biodegradable coolant (2025). As AI training clusters demand increasing power density, the thermal management layer is becoming infrastructure, not an accessory.

Yunze Capital (云泽资本).

Eagle Vision (鹰瞰智翼) — Pre-A, tens of millions RMB, ¥150M (~$21M)

Bionic flapping-wing flying robots and a proprietary fluid simulation engine. Consumer product “Dante” targeting global hobbyist and enthusiast markets with stunt flying capabilities, with a Kickstarter campaign for global distribution. Core team from Shanghai Jiao Tong University with PhD-level backgrounds in robotics and control. A left-field entry in a week dominated by humanoid robots and LLMs.

Oriza Origins (元禾原点).

Hebang New Materials (禾邦新材) — A-round, ¥100M (~$14M), ¥500M (~$69M)
Bio-based composite materials from plant biomass, specifically extracting high-purity lignin from agricultural straw using an enzymatic process the company describes as a world first in commercialization. 25% carbon intensity reduction versus petroleum-based epoxy. Certified to EN45545 (EU high-speed rail fire safety standard) and UL94 V-0. Applications: aerospace, transportation, and high-end equipment. Sustainable materials with real certifications, not just a green narrative.

Changjiang Capital (长江资本), Hefei Innovation Investment (合肥创新投资).

The Bigger Picture

Two Token Factories, Same Week, Same Pitch

Two companies raised ¥1B+ combined on the same thesis within 48 hours of each other: the data center as “token factory.”

Infinigence AI (¥700M+ B-round) and Magik Compute (魔形智能) (hundreds of millions Pre-A) both position themselves as “token factories,” middleware layers that convert raw compute from heterogeneous chip pools into AI inference tokens at scale. Infinigence is the Tsinghua spinout profiled in Spotlight above; Magik Compute, founded June 2024 by former Alibaba Cloud and Bitfusion executives, is building software-hardware co-optimized compute clusters for the same market from a different angle.

The demand side supports the bet: China’s daily AI token call volume hit 1.4 quadrillion in March 2026, a 1,000x increase in two years per Xinhua, citing official data. Kimi’s own capacity crunch in early 2026, documented by TMTPost, saw users turned away because inference supply could not meet demand.

But the composition of that demand matters. ByteDance’s Doubao alone accounts for 1.2 quadrillion tokens per day as of March 2026, roughly 86% of the national total, all served on ByteDance’s own Volcengine infrastructure. The token factories are not competing for that volume. ByteDance, Alibaba, and Zhipu AI build their own inference infrastructure and monetize spare capacity through their own platforms. Kimi solved its shortage internally with software optimization rather than outsourcing. The token factory thesis holds for the long tail: agent startups, enterprise AI deployments, and smaller model companies that lack the scale to justify proprietary GPU clusters. Whether that long tail produces durable revenue before China’s hyperscalers expand downmarket is the bet that ¥1B+ in same-week capital is now testing.

Built Global, Not Going Global

Three companies in this week’s batch illustrate different modes of global competitiveness, none following the traditional Chinese playbook.

Makera raised $10.2M on Kickstarter for a desktop CNC machine, with a predominantly Western backer base. PixelBloom’s AiPPT serves 30M+ registered users across 180+ countries from a Singapore-incorporated entity. Both were global products before they became Chinese tech stories, built for foreign users from day one. Kimi’s path is different. K2.6 competes directly with GPT-5.4 and Claude Opus 4.6 on the benchmarks that define the global AI frontier, even though Kimi’s revenue comes overwhelmingly from Chinese users.

The traditional model is “先做国内再出海” (build at home first, then go overseas). A newer pattern is emerging: companies that set up overseas operations from inception and validate product near end users without a domestic-first phase. TechNode’s Born Global 2026 Awards and CGTN’s analysis of China’s next-gen global enterprises cover the same shift. Chinese companies’ overseas revenue ratio stood at 16.7% in 2024 per CICC, with Goldman Sachs projecting 20.6% by 2028. Companies like Makera and PixelBloom are part of the reason.

Why These Deals Couldn’t Wait

46 deals versus 100-plus in normal weeks. The May Day break (May 1-5) shut the marginal deals down, but the rounds that closed during the holiday all had reasons they couldn’t pause.

Commercial breakout on a clock. Kimi’s ARR doubled in a single month. When revenue is accelerating that fast, every week of delay means a higher valuation for incoming investors. Meituan closed at $20B before the next step-up. Makera’s revenue was tracking 300% year-on-year post-Kickstarter, and Qiming wanted to load up before the metrics cooled.

Construction deadlines. STARTORUS needs capital before the NTST device installation begins in September 2026. The 11 new A+ investors closed in four months because fusion reactor construction schedules do not care about national holidays.

Category definition races. Infinigence AI and Magik Compute both announced within 48 hours, racing to establish the “token factory” category. When two companies pitch the same thesis to the same investor base, whoever closes first sets the narrative. Waiting a week means the other company’s press coverage defines the category for you.

The 50-plus deals that did not happen this week were the ones that could afford to. What closed: AI infrastructure, frontier energy, and globally validated products.

Deal data compiled from Chinese corporate announcements, financial news, and investment databases. Amounts and valuations are as reported and may be incomplete. For informational purposes only. Not investment advice. Verify all figures independently before acting on them.

Where capital flows tells you where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

Where the Money Went

129 deals in seven days, $847M (~¥6.1B) in disclosed capital. Aerospace and defense led on dollar volume at $370M across 11 deals, the lowest deal count of any sector, driven by Volant Aerotech’s $300M Series C and Cosmoleap’s ¥500M (~$69M) Series A; no other sector crossed $200M. Robotics followed at $200M across 19 deals, concentrated in RobotEra’s single large round. The remaining five sectors generated 97 deals and under $280M combined.

🚀 Aerospace & Defense (11 deals)

Two rounds accounted for nearly all disclosed capital. Volant Aerotech closed what multiple Chinese financial outlets confirmed as the largest single round in Chinese passenger eVTOL history: over $300M, with a UAE fund (Stone) leading and Shanghai’s first city-level state capital entering. Cosmoleap closed a ¥500M (~$69M) Series A for its chopstick-catch reusable rocket (the same tower-arm booster recovery system SpaceX uses), covered by SpaceNews. The remaining nine deals are earlier stage, concentrated in urban air mobility, SAR satellite infrastructure, and UAV systems.

🤖 Robotics (19 deals)

SF Express both led RobotEra’s $200M+ strategic round and operates the company’s humanoid robots across 10+ logistics centers. The operator-as-investor structure, where the round lead is also the largest deployment partner, is defining which companies demonstrate commercial traction rather than commercial intent. LinkerBot extended its fundraising to seven rounds in 18 months after shipping 10,000+ dexterous hands. Wujie Power attracted a ¥500M (~$69M) overseas deployment contract from Envision alongside its raise, the largest overseas commercial contract signed by a Chinese embodied AI company to date. Botshare, a robot rental platform launched December 2025, completed its fourth round in 105 days.

🧠 AI (22 deals)

Creati.ai closed a $20M Series B for its Buzzy video editing product, which the company frames as “the world’s first AI Video Photoshop,” reporting $15M in ARR and 20 million+ global users. Spirit AI’s Series A+, a straggler from the April 24 batch with Alibaba as sole lead and a reported ¥5.5B (~$760M) valuation, rounds out the AI highlights. The remaining 20 AI deals are early-stage and mostly undisclosed.

🔬 Semiconductors (22 deals)

IDG Capital led a ¥500M (~$69M) Pre-Series A for Calculet at a reported ¥3B valuation, building AI processors on a chiplet architecture. With access to advanced monolithic nodes constrained by export controls, chiplet designs assembled from available process nodes are emerging as a structural workaround. The remaining 21 deals are early-stage and undisclosed.

⚡ New Energy & Materials (24 deals)

The highest deal count at 24, but only $49M disclosed. QingNa Technology raised ¥200M (~$28M) in a Series A for sodium-ion batteries, with one million large cylindrical cells shipped to European customers in January 2026 and five million contracted for the year. The remaining 23 deals are early-stage and mostly undisclosed.

⚙️ Other Advanced Manufacturing (22 deals)

22 deals, almost entirely undisclosed. Cangzhou Four STARS Glass, a 33-year-old pharmaceutical glass maker, closed the only disclosed round at ¥300M (~$42M) Pre-IPO. The remaining deals are concentrated in industrial automation, precision testing, and manufacturing equipment suppliers, the enabling layer beneath the robotics and semiconductor sectors above it.

💡 Frontier Tech (8 deals)

TuringQ completed a Series B+ for photonic quantum computing, reaching an estimated ¥7B (~$970M) valuation on the strength of ¥100M+ (~$14M) in signed orders in H1 2025 and approaching ¥1B (~$138M) raised in 2026 alone. BCDTEK closed a Series A for silicon-based OLED microdisplays (the display technology inside the Apple Vision Pro) while building China’s first 12-inch silicon OLED fab.

Spotlight

Volant Aerotech (沃兰特航空)

Series C, over $300M (~¥2.16B)

The largest single round in Chinese passenger eVTOL history: UAE capital leads, Shanghai’s first city-level state investor enters, and the order book stands at over 1,900 aircraft.

Volant Aerotech is developing the VE25-100 (Tianxing), a six-seat eVTOL with a 2,500kg maximum takeoff weight and a 200-400km range. The pre-production prototype completed its rollout at the Zigong Aviation Industrial Park in Sichuan in May 2025. China Southern General Aviation signed the first confirmed commercial passenger eVTOL order in China in January 2025, for multiple aircraft at approximately ¥25M (~$3.5M) per unit, deposit paid.

The Series C, closed April 27 and led by Stone (UAE), sets a size record for Chinese passenger eVTOL confirmed by multiple Chinese financial outlets and the company’s English press release. Futen Capital (孚腾资本), backed by Shanghai State-owned Capital Investment Co. as its primary shareholder, entered the round in what reporting describes as the first direct investment by Shanghai city-level state capital in a crewed eVTOL OEM, a designation that carries implications for certification support, infrastructure investment, and public procurement. The accumulated order book now stands at over 1,900 aircraft worth more than ¥47.5B (~$6.6B), with confirmed deposits from China Southern General Aviation, Asian Express, and ABC Financial Leasing.

Founder Dong Ming spent over 20 years in civil aviation across AVIC, GE, Rockwell Collins, and Aviage Systems, with direct involvement in the ARJ21, C919, and CR929 programs.

Sequoia China, Futen Capital, Future Capital, and Legend Capital co-invest alongside Stone.

Announcement (Chinese) · Announcement (English)

RobotEra (星动纪元)

Strategic Investment, over $200M (~¥1.44B)

SF Express led the round and operates the robots across its own logistics network, the only deal this week where the lead investor is also the operator at production scale.

RobotEra builds full-stack humanoid robots with a stated goal of industrial deployment at scale. Its L7 stands 171cm, weighs 65-70kg depending on configuration, runs 55 degrees of freedom (the count of independent joint movements, a proxy for human-like dexterity), and reaches 4 m/s, among the fastest bipedal humanoids in production. Bimanual carry capacity is 20kg. The XHAND1 dexterous hand adds 12 degrees of freedom as a modular unit. RobotEra develops approximately 95% of its hardware in-house, including motors, sensors, and control systems, a prerequisite it argues for the cost reduction required at deployment scale.

SF Group, China’s largest integrated logistics company, led this round and simultaneously operates RobotEra robots across 10+ logistics centers spanning North, East, and South China. Deployed robots run at 85% of human worker efficiency in logistics environments. RobotEra shipped 600+ cumulative units as of CES 2026 in January and announced thousand-unit scale batch deliveries beginning Q2 2026. SF Express is running robots on its own production lines and backing the supplier with over $200M simultaneously.

Founder Chen Jianyu is an assistant professor and doctoral supervisor at Tsinghua University’s Institute for Interdisciplinary Information Sciences, PhD from UC Berkeley under Masayoshi Tomizuka. Tsinghua University holds equity, described on the company website as unique among humanoid robotics companies.

Sequoia China, IDG Capital, and CICC co-invest alongside SF Group.

Announcement

Cosmoleap (大航跃迁)

Series A, ¥500M (~$69M)

Founded February 2024, Cosmoleap has raised approximately $83M in 18 months for a chopstick-catch reusable rocket, with SpaceNews coverage marking rare Western trade press attention for a Chinese launch startup at this stage.

Cosmoleap is developing the Leap-1A (跃迁一号甲), a 70-meter methane/liquid oxygen rocket targeting an 18-ton LEO payload capacity and first flight in 2027. The Qingyu-11 engines are rated for 50 reuses.

The company is testing a tower-catch booster recovery system (the same chopstick-arm architecture SpaceX uses with Mechazilla), with catch tower testing underway since November 2024. A nine-engine first stage burns liquid oxygen and methane.

This Series A, co-led by Houji Capital (厚纪资本) and Puhua Capital (普华资本) with Huaxing Capital as financial advisor, brings total capital raised to approximately $83M (¥100M (~$14M) seed in November 2024 plus ¥500M (~$69M) here) after 18 months of operation. SpaceNews covered the round directly, a level of Western trade press attention rare for a Chinese launch startup at Series A stage. Zhongguancun Development Group entered as a strategic investor.

Announcement

TuringQ (图灵量子)

Series B+, amount undisclosed

Photonic quantum computing with ¥100M+ in commercial orders, a national guidance fund entering directly, and an estimated valuation of ¥7B (~$970M).

TuringQ builds photonic quantum computers using a full integrated device manufacturer model, designing, fabricating, and commercializing its own chips while operating China’s first photonic chip pilot production line at the Wuxi Photonic Chip Research Institute. Two in-house photonic chip fabrication approaches enable a 2-3 week iteration cycle, fast for a hardware company running its own production line. Products are sold into financial services, quantum cryptography, and scientific computing customers including state-owned enterprises in energy and telecommunications.

The commercial trajectory matters: ¥500,000 (~$69K) in revenue in 2022, over ¥10M (~$1.4M) in 2023, and over ¥100M (~$14M) in signed orders in H1 2025. The 2025 World Internet Conference awarded TuringQ and SJTU’s Wuxi institute its Leading Technology Award, the first time the prize went to a quantum computing project. China’s quantum sector saw investment in Q1 2026 approach full-year 2025 levels, with TuringQ among the largest deals. This B+ is led by Zhangjiang Group (张江集团), Haiwan Capital (海望资本), and Guotou Chuanghe (国投创合, managing the Yangtze River Delta Fund under the National Venture Capital Guidance Fund). Total raised in 2026 alone is approaching ¥1B (~$138M). Estimated valuation: approximately ¥7B (~$970M) per deal trackers.

Founder Jin Xianmin is an assistant professor and 2021 Changjiang Scholar at Shanghai Jiao Tong University’s School of Physics and Astronomy. He holds 150+ publications across Science, Nature Photonics, and Nature Communications and gave up UK residency after his postdoc at Oxford to return and found the company.

Announcement

Creati.ai (感知阶跃)

Series B, $20M (~¥144M)

A Chinese-founded AI video company launching Buzzy, which the company frames as “the world’s first AI Video Photoshop,” with $15M in ARR and 20 million global users.

Creati.ai launched as ZMO.AI in 2020 generating AI fashion product imagery for e-commerce. It relaunched as Creati in April 2024 with a broader AI content platform. This Series B is structured around the launch of Buzzy: a conversational video editing tool that makes localized edits to existing footage (removing background people, fixing eye contact, adjusting lighting, swapping products) without regenerating the full clip. The company positions Buzzy as “the world’s first AI Video Photoshop,” a deliberate framing against both Adobe and the generative-AI-replaces-everything narrative. It targets video creators and e-commerce merchants who need precise, reversible edits.

The commercial signals behind a $20M round: the company reports $15M in annual recurring revenue and 20 million+ global users. For a Chinese AI SaaS product with USD pricing and international distribution, those figures at this raise size represent one of the stronger commercial profiles in this week’s batch. Founder Ella Zhang (张诗莹) has run the company through both pivots. The prior Series A was led by Hillhouse with GGV Capital and GSR Ventures participating.

Redpoint leads. CCV (创世伙伴创投) co-invests.

Announcement

Also on the Radar

LinkerBot (灵心巧手)

Series B+, undisclosed

Seven rounds in 18 months. LinkerBot’s 42-degree-of-freedom dexterous hand has shipped over 10,000 units at a production rate of 4,000+ per month, with a reported valuation now above ¥15B (~$2.1B). Government capital (Zhongguancun Fund, Bank of China Asset Management) leads the B+ alongside commercial VCs, a composition that typically tracks procurement contract visibility rather than purely financial valuation.

Linkhou (灵猴机器人)

Series B+, hundreds of millions RMB

CIIF (中网投基金), China’s national industrial investment fund, led Linkhou’s Series B+ at a reported ¥3B (~$415M) valuation. Linkhou makes components for humanoid robots (perception systems, motors, and controllers) sold into multiple OEMs rather than competing at the system level. Silicon Valley R&D office. Revenue reportedly exceeding ¥200M (~$28M) from embodied AI orders. The picks-and-shovels position is defensible regardless of which humanoid OEM wins at scale.

Wujie Power (无界动力)

Angel, hundreds of millions RMB

Over $200M raised cumulatively at angel stage, ¥5B (~$695M) valuation, and a ¥500M (~$69M) overseas deployment contract signed with Envision for humanoid robots on European and Asian production lines. Envision, a clean energy and battery materials company backed by Chinese and Japanese capital, both led the investment and signed the deployment contract simultaneously, the same operator-as-investor structure as RobotEra and SF Express this week.

Calculet (原粒半导体)

Pre-Series A, ¥500M (~$69M)

IDG Capital led a ¥500M (~$69M) Pre-Series A for Calculet at a reported ¥3B valuation. Calculet builds AI processors on a chiplet architecture, modular standardized compute units assembled into custom configurations rather than monolithic designs that require advanced process nodes. Chiplets assembled from production-capable nodes avoid the export control constraints on equipment and advanced processes. Guoxin Science and Technology Innovation Fund (国新科创基金) co-invests alongside IDG and semiconductor-specialist VCs.

SEAHI Robotics (世航智能)

Series A+, undisclosed

Underwater robots for 0-10,000 meter depths. SEAHI‘s proprietary thrusters are rated for 10,000 operational hours versus a 300-400 hour industry standard, a 25-30x differential that is the company’s primary technical moat. GSR Ventures’ Zhu Xiaohu has invested across three consecutive rounds. Founded May 2023 by a team from Harbin Engineering University with operational deep-sea deployment experience. Simultaneous Series A+ and A++ rounds announced March 31, 2026.

Botshare (擎天租)

Pre-Series A, hundreds of millions RMB

A robot-as-a-service rental platform founded December 2025 that completed four rounds in 105 days. At ¥300,000 (~$42K) per humanoid robot, the purchase price is out of reach for most SME buyers; daily rental changes the access model. Charoen Pokphand’s robotics arm reportedly invested, bringing the Thai conglomerate’s manufacturing and food processing operations across Asia into both the investor and likely deployment picture. Reported valuation: ¥4.2B (~$580M).

Cangzhou Four STARS Glass (四星玻璃)

Pre-IPO, ¥300M (~$42M)

A 33-year-old borosilicate glass company closing a Pre-IPO round is the most unusual deal by company age in this week’s cohort. Sixing makes pharmaceutical packaging glass (vials, ampoules, syringes) using the cold-top all-electric Vello process, the same manufacturing approach as SCHOTT in Germany. Sixing is reportedly the only Chinese company operating this process at scale, and fourth globally. China’s healthcare supply chain localization push creates sustained demand for domestic alternatives to imported pharmaceutical packaging that can match the quality standard.

BCDTEK (芯视佳)

Series A, hundreds of millions RMB

Silicon-based OLED microdisplays for AR and VR, the display technology inside the Apple Vision Pro, currently dominated globally by Sony’s OLEDoS process. BCDTEK is constructing China’s first 12-inch silicon OLED fab. The combination of display technology (the bottleneck in high-end AR) and domestic fab capacity (the semiconductor sovereignty angle) makes this relevant to both the AR hardware market and China’s display independence agenda.

破壳机器人 (”Hatching Robot”)

Angel, tens of millions USD

Home service robots, founded by a Tsinghua professor, backed by Yunqi Capital (led), Shunwei Capital (Lei Jun’s fund), and Honghui Fund at a reported ¥2.6B (~$360M) valuation with no products shipped publicly. The valuation rests entirely on founder premium: Tsinghua credentials, high-profile investor names, and a large addressable market in home robotics.

Spirit AI (原力灵机)

Series A+, undisclosed

Spirit AI‘s Spirit v1.5 VLA (Vision-Language-Action) model ranked #1 on RoboChallenge, the global real-world robotics benchmark, outperforming Physical Intelligence’s π0.5. Moz1 humanoid robots are deployed on CATL’s Zhongzhou battery production line. Huaqin Technology (华勤技术), Yufu Holdings (渝富控股, Chongqing state capital), and Hengxu Capital (恒旭资本) led the Series A+. Total raised: approximately ¥3.3B (~$460M) across six rounds in 18 months. The “dirty data” training thesis (building robot intelligence from unstructured real-world data rather than curated clean datasets) is the contrarian technical bet underlying the benchmark performance.

QingNa Technology (青钠科技)

Series A, ¥200M (~$28M)

Sodium-ion batteries for industrial and grid applications. One million large cylindrical cells shipped to European customers in January 2026; five million contracted for the year, making QingNa the first Chinese sodium-ion startup to reach commercial-scale export to Europe.

Huaxi Aviation (华喜航空)

Seed, tens of millions RMB

A hydrogen-turbine hybrid eVTOL: hydrogen fuel cell plus turbine rather than battery-electric. Joby, Archer, Wisk, and Volant are all-electric; Huaxi is pursuing a different energy architecture. Hydrogen-turbine hybrids address the range and energy density limitations that battery-electric designs face at larger scale. Seed valuation is reported at ¥150M (~$21M).

The Bigger Picture

The Customer-Investor Loop

The three strongest robotics deals this week share the same structure: the company writing the largest check is also running the robots.

SF Express led the RobotEra round (detailed in the Spotlight above) and operates the company’s humanoid robots across its own logistics network. Envision led Wujie Power’s raise and signed a ¥500M (~$69M) overseas deployment contract in the same transaction. Botshare’s Pre-Series A brings in capital alongside a rental model that lowers the access barrier for SME buyers priced out of ¥300,000+ direct purchases.

The structure makes commercial sense at this stage of the technology. Humanoid robots are not yet at the cost and reliability level where a neutral financial investor can independently underwrite deployment risk. A major logistics operator, a battery materials manufacturer, or a manufacturing conglomerate has the operational environment, the data collection capacity, and the scale purchasing power to run a real robot program. The equity investment gives them preferred pricing, roadmap access, and upside if the company scales. The deployment contract gives the startup committed purchase volume, the hardest problem in early commercial robotics.

At this stage of the technology, commercial traction is better read from the lead investor’s operational relationship with the product than from valuation or benchmark position alone.

China’s Private Aerospace Week

China’s aerospace and commercial space sector raised ¥26.6B (~$3.81B) across 137 rounds in 2025, a record year. This week added $369M at the top.

Volant’s $300M is the largest round in Chinese passenger eVTOL history. The investor composition matters as much as the size: Futen Capital, backed by Shanghai State-owned Capital Investment Co., entered a crewed eVTOL OEM for the first time. Toyota’s $394M Series C investment in Joby Aviation in January 2020 established that a major industrial operator would commit strategic capital to the category; Futen’s entry serves the same function for Chinese eVTOL. Shanghai has publicly framed its ambition to become the world eVTOL capital, and Futen’s investment is the first direct capital commitment behind that public ambition.

CAAC’s eVTOL certification timeline (31 months versus 5-7 years in Western markets) is a structural advantage the capital markets are pricing in. With 1,900+ aircraft on Volant’s order book and Shanghai municipal state capital committed, the conditions for a national champion designation are visible.

Cosmoleap’s ¥500M (~$69M) Series A, covered by SpaceNews, is the launch side of the same story: a company founded February 2024 has raised approximately $83M in 18 months for a reusable rocket with tower-catch recovery. China’s private launch sector now has multiple companies at significant funding levels, with the gap to SpaceX-tier technology closing on the orbital reuse side.

Quantum Goes Commercial

China’s quantum sector saw investment in Q1 2026 approach full-year 2025 levels. TuringQ’s revenue trajectory (¥500,000 (~$69K) in 2022, ¥10M+ in 2023, ¥100M+ in signed orders in H1 2025) is the first genuine commercial curve to appear in photonic quantum computing. The 2025 World Internet Conference Leading Technology Award going to a quantum computing project for the first time is one signal; the national guidance fund entering directly is another.

The technology path TuringQ pursues, photonic chips operating as co-processors alongside classical systems at room temperature, sidesteps the cryogenic requirements that constrain superconducting quantum computers and puts early commercial applications within reach: financial risk modeling, materials simulation, and combinatorial optimization, problems where the solution space is too large for classical computers to search exhaustively. PsiQuantum, the most-funded photonic quantum company globally, has raised over $2B on the same architectural bet. TuringQ has reached ¥100M+ (~$14M) in customer orders at a fraction of that capital intensity, operating an integrated domestic chip production line.

The entrance of the National Venture Capital Guidance Fund’s Yangtze River Delta Fund into the B+ round is the policy signal that the commercial trajectory has been validated at the highest level. China’s 15th Five-Year Plan places quantum in the same “future industry” designation as semiconductors and AI. At this level of the state capital hierarchy, direct investment has historically preceded national champion designation in sectors from semiconductors to EVs. TuringQ is the first photonic quantum company in China to combine a national guidance fund investment, a ¥7B (~$970M) valuation, and a demonstrated ¥100M+ commercial order book.

Deal data compiled from Chinese corporate announcements, financial news, and investment databases. Amounts and valuations are as reported and may be incomplete.

Where capital flows tells you where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

Where the Money Went

111 deals in seven days, $1,020M (~¥7.3B) in disclosed capital, and 95 of those 111 deals didn’t disclose. All the visible money concentrates in a handful of large rounds: robotics ($419M) and AI ($504M) collected 90% of the total while generating fewer than 40% of the deals. Semiconductors ran the highest deal count of any sector at 23, but almost none disclosed amounts, operating early-stage and quiet.

🤖 Robotics (18 deals)

Sudu Technology came in at a reported $2B+ valuation at 11 months old and demo stage; X Square Robot collected all four of China’s internet giants on a single cap table at ¥2B (~$278M) Series B. The bottom of the distribution is a long tail of ~$690K seed rounds for pre-product startups building everything from dexterous hands to explosion-proof industrial quadrupeds. The component layer sits between them: Motorevo shipped 100,000+ joint modules in 2025 and is still valued at ¥1.6B (~$222M). The stage distribution skews to the extremes, with large late-stage rounds and a long tail of sub-¥10M seeds, but few companies in between.

🧠 AI (24 deals)

EVAS closed the week’s largest AI round at ¥1.5B (~$208M) with four Beijing municipal funds co-leading simultaneously, one day after Beijing formally established the ¥20B (~$2.8B) four-fund structure that backed the round. Sunrise AI (曦望) extended its run to seven financing rounds in ~16 months, crossing ¥10B (~$1.4B) valuation as China’s first inference-GPU unicorn. Loopit’s $50M (~¥360M) round (Garena lead) follows the app reaching #1 on the global Google Play Entertainment chart.

🔬 Semiconductors (23 deals)

The highest deal count of any sector, but almost none disclosed. The sector’s early-stage density is running well ahead of its public capital story. Two companies warrant attention regardless: NanoLN (晶正电子) is raising Pre-IPO financing as the holder of approximately 78% of global lithium niobate thin-film wafer supply, a material foundational to 5G filters and photonic chips. Wendun Juxin (稳顿聚芯) shipped China’s first domestic high-precision compound-semiconductor stepper in September 2025 and is already profitable. Neither has significant English-language coverage.

⚡ New Energy & Materials (13 deals)

Zero disclosed capital across 13 deals. Sodium battery, flow battery, and advanced materials companies raising without announcing terms. Luquos Energy (易池新能, CUHK spinout) is the exception, a Hong Kong-registered flow battery company with international backing (Towngas + Gobi Partners) that chose to disclose.

🚀 Aerospace & Defense (10 deals)

The National Strategic Emerging Industry Investment Fund (国调基金) made a strategic investment in Chang Guang Satellite that accounts for most of the sector’s visible capital, a state-capital substitute for the financial and foreign investors who exited after Chang Guang was placed on the US Treasury’s SDN (Specially Designated Nationals) sanctions list. FAW Group’s eVTOL spinout Qiyi Technology (旗翼科技) added the week’s largest angel-stage aerospace deal at ¥400M (~$55M).

⚙️ Other Advanced Manufacturing (20 deals) · 💡 Frontier Tech (3 deals)

Audfly (清听声学) stands out in Other Adv. Manufacturing with a Series B+ at ¥2B (~$278M), a directional audio company backed by Lenovo Ventures. Frontier Tech had just three deals: a synthetic biology Series A from PAML Biotechnologies (柏垠生物, Hillhouse-adjacent capital) and a fusion-wire JV from Juneng Superconducting (聚能超导).

Spotlight

EVAS (奕行智能)

Series B, ¥1.5B (~$208M)

China’s first reportedly mass-produced RISC-V AI chip, now backed by Beijing’s four-fund industrial apparatus.

EVAS makes AI chips built on RISC-V, an open-source chip architecture that carries no US intellectual property. Every Chinese chip company designing on Arm or x86 carries licensing risk: a single US policy change could cut access to the architecture itself. RISC-V removes that dependency entirely. EVAS’s Epoch series entered mass production in early 2025 for data center, internet, and automotive AI customers, reportedly making it the first Chinese company to ship RISC-V AI chips at scale. CEO Liu Hui (刘珲) is a semiconductor veteran from Socionext, STMicroelectronics, and Fujitsu.

The timing of this round is worth noting. Beijing formally launched a ¥20B (~$2.8B) four-fund structure on April 21. All four funds co-led EVAS’s Series B the next day. That sequencing was almost certainly pre-arranged, but the public choreography is the point: the government is signaling that RISC-V is the chosen path for compute independence, and converting that signal into capital allocation immediately. Eight Chinese ministries coordinated a national RISC-V push in 2025. If RISC-V AI chips reach performance parity with Arm-based alternatives, China’s entire inference stack could eventually run on architecture that no export control can touch. The investor line here is entirely government capital with no commercial VCs, which tells you Beijing is treating RISC-V compute as public infrastructure.

Four Beijing municipal funds co-led: the E-Town Industry Upgrade Fund (亦庄国投), the Beijing High-Precision Industries Development Fund, the Beijing Information Industries Fund, and the Beijing AI Industry Fund. And/Li Capital, Bolly Ventures, Sayi Industries Fund, Longjiang Fund, Qingtan Capital, and Jiukun Ventures participate.

Announcement

Sudu Technology (苏度科技)

New financing round, amount undisclosed, valuation reported at $2B+ (~¥14.4B+)

Eleven months old, demo-stage, and reportedly valued above $2 billion (~¥14.4B), the most extreme pre-revenue ratio in this week’s cohort.

Sudu Technology builds what it calls a universal embodied AI platform: a “robot brain” that uses 3D world models with reinforcement learning to handle manipulation tasks involving transparent, reflective, and flexible objects, categories where most robotic vision systems still fail. Picking up a glass, folding a towel, handling a bag of chips without crushing it. These are trivially easy for humans but remain unsolved at scale in robotics because the objects deform, reflect, or transmit light in ways that break standard computer vision. Sudu’s approach trains its system to build a 3D understanding of the world and learn through trial and error rather than following pre-programmed routines. Its flagship Sudo R1 has been demonstrated in 60-minute unedited real-world testing footage released publicly. CEO Han Zheng (韩铮) co-founded the company in May 2025. The company is 11 months old.

Chief Technical Advisor Su Hao (苏昊) is the credentialing anchor for the valuation. He co-created ImageNet in 2009 with Fei-Fei Li at Stanford, the massive labeled image database that effectively launched modern deep learning. Before ImageNet, AI systems struggled to reliably distinguish a cat from a dog; by giving researchers millions of labeled images to train on, the project enabled the breakthroughs in computer vision that now power everything from autonomous driving to facial recognition. Su also pioneered PointNet, the foundational architecture for processing 3D point cloud data, which is how robots understand spatial environments. He concurrently leads Fudan University’s Institute of General Physical Intelligence. The investor bet is that embodied AI could follow a similar path from academic foundation to commercial platform, and that whoever controls the software layer of the robot stack controls the platform.

The investor line reflects that bet across sectors: CATL positioning for the robot power stack, Alibaba, Tencent, and Ant Group as platform companies that would distribute robots through their ecosystems, and IDG, Hillhouse, and China Life as financial conviction. Whether $2B for a demo-stage product is early-stage pricing for a future platform or the clearest evidence of capital running ahead of commercial reality is the open question. The amount raised in this round has not been publicly disclosed.

CATL, Alibaba, Tencent, Ant Group, IDG Capital, Hillhouse, China Life Equity, BlueRun Ventures, Futen Capital (孚腾资本), Digital Future (数字未来), Fudan Innovation (复旦科创), and Yunhui Capital (云晖资本). Financial advisor: Gaogul Capital (高鹄资本).

Announcement

X Square Robot (自变量机器人)

Series B, ¥2B (~$278M) @ ¥12B (~$1.67B) val

Humanoid robot butlers deployed in Chinese homes, and the first time all four of China’s internet giants are on one cap table.

X Square Robot builds general-purpose humanoid robots on its proprietary WALL-A foundation model, a unified architecture processing vision, language, touch, and motion through a single system. The Quanta X2 humanoid stands 172cm with a 6kg payload per arm. In March 2026, the company launched a commercial home-cleaning service via 58 Daojia, China’s largest home services platform, making it the first Chinese company to deploy humanoid robots into homes at commercial scale. The company also unveiled WALL-B, a home-environment AI model, the same week as this round closed. Founded December 2023, it has raised nine rounds totalling over ¥3B (~$417M) in under two and a half years.

ByteDance, Meituan, Alibaba, and Xiaomi are now simultaneously on the cap table, the first time China’s four largest internet platforms have co-invested in the same embodied AI company. Each came in at a different round: Meituan led the Series A in May 2025 as sole investor; Alibaba Cloud co-led the A+ in September; ByteDance co-led the A++ in January 2026; Xiaomi co-leads this B. In the 2015 bike-sharing era, these same companies funded competing startups against each other. Here they are all on the same cap table, which is either a sign of X Square’s dominance or a sign that each platform wants guaranteed access to whatever the home robot deployment layer becomes. Each brings different distribution: Xiaomi’s consumer hardware retail, Meituan’s home services network, Alibaba’s cloud and logistics, ByteDance’s content discovery.

Sequoia China and Xiaomi Strategic Investment co-lead. ByteDance, Meituan, Alibaba Cloud.

Announcement

Loopit (涌跃智能)

Third round in 2026, $50M (~¥360M)

Global #1 on Google Play Entertainment, $50M from Garena, and a distribution play into Southeast Asia.

Loopit is an AI interactive content platform. Users type a prompt and the app generates a rich mobile experience combining images, audio, video, and 3D elements that responds to the phone’s sensors (microphone, gyroscope, camera) in real time. Instead of scrolling past a video, you’re inside it, tilting your phone to explore, speaking to change the scene. CEO Chen Weipeng (陈炜鹏) co-founded and led the large-model team at Baichuan AI, previously building products at Sogou and Soul. The app launched in February 2026, hit #1 in the Google Play Entertainment category globally and #8 overall within two months, and in February Elon Musk reposted a user-generated Loopit video on X.

A Chinese consumer AI app reaching #1 globally on Google Play is still rare (TikTok remains the only real precedent at that scale), and that traction is what brought Garena to the table. Sea Group’s gaming arm operates across Southeast Asia (Indonesia, the Philippines, Thailand, Vietnam, Malaysia, Taiwan) in markets where Loopit’s interactive content format travels naturally among mobile-first, gaming-native users. Garena gets a position in the content format before Instagram Reels or TikTok can copy the interactivity layer; Loopit gets SE Asia distribution that a Beijing startup would take years to build independently. Chen built the large language model team at Baichuan AI (one of China’s leading foundation model companies), which suggests the AI generation engine underneath is purpose-built rather than an API wrapper around someone else’s model.

Garena (lead), Jingwei China (经纬中国, fourth check into the company), BlueRun Ventures, Yingce Capital (渶策资本), Bairui Capital (柏睿资本). Financial advisor: Gaogul Capital (高鹄资本).

Announcement

Chang Guang Satellite (长光卫星)

Strategic investment, hundreds of millions RMB, valuation ~¥16B (~$2.2B)

The US Treasury listed it as a sanctioned entity. China’s sovereign development fund just became its anchor investor.

Chang Guang Satellite operates the Jilin-1 constellation, the world’s largest sub-meter resolution commercial remote sensing satellite constellation, with 144–152 satellites in orbit delivering optical imagery to government, enterprise, and international customers. The company was founded in 2014 as a spinout from the Changchun Institute of Optics, Fine Mechanics and Physics, and had been building toward a STAR Market IPO since a 2022 filing. In December 2023, the US Treasury placed it on the SDN (Specially Designated Nationals) sanctions list for selling satellite imagery to Wagner Group proxy entities, covering Ukrainian battlefields and Wagner’s operational areas in Africa. Foreign investors exited. The IPO was withdrawn in late 2024

In January 2026, Chang Guang re-filed for a STAR Market IPO under new rules that extend the exchange’s fifth listing standard to commercial aerospace companies, removing strict profit requirements for companies whose value is in technology rather than current earnings. This week, the National Strategic Emerging Industry Investment Fund (国调基金), managed by China Chengtong, one of two state-owned capital operating companies that deploy sovereign investment on behalf of the State Council, completed a strategic Pre-IPO investment at a valuation reported by deal databases at around ¥16B (~$2.2B). The capital the SDN designation drove away, foreign investors and commercial VCs with compliance exposure, has been replaced by a sovereign fund whose mandate is national strategic interest rather than financial return.

National Strategic Emerging Industry Investment Fund (国调基金), managed by Chengtong Fund Management under China Chengtong Holdings (中国诚通).

Announcement

NanoLN (晶正电子)

Pre-IPO round, amount undisclosed

The company that holds approximately 78% of global lithium niobate thin-film wafer supply is heading for a public listing.

NanoLN makes lithium niobate thin-film (LNOI) and lithium tantalate thin-film (LTOI) wafers. These are nanometer-thick crystalline films that manipulate light and radio signals, and they sit inside a surprising range of critical hardware: the RF filters in every 5G base station, the electro-optic modulators that move data through fiber optic connections in AI data centers, LiDAR ranging systems, and quantum communication devices. Founded in 2010 in Jinan, the capital of Shandong province on China’s east coast by returnee entrepreneur Hu Wen, NanoLN was the first company globally to achieve industrialized production of these films and the first to produce 8-inch LNOI wafers. Huawei Hubble is an early investor. The company serves 200+ customers across China, the US, Japan, and Europe.

NanoLN holds approximately 78% of global LNOI wafer supply per Yole (2023Q3), with the remainder split between NTT Advanced Technology and SCIOCS (both Japan). That near-monopoly feeds two accelerating demand curves: 5G infrastructure buildout and AI data center photonics, where the optical interconnects between GPU clusters increasingly rely on LNOI-based modulators. In practice, a significant share of the world’s 5G filters and data center photonic links depend on wafers from a single Chinese company, a supply chain concentration comparable to TSMC’s position in advanced logic chips, though at a much earlier stage of geopolitical visibility. The Pre-IPO round is the step before a listing that will ask Chinese capital markets to price a deep-tech materials monopolist with a global customer base for the first time.

Leading Investment, Shandong Expressway, Jiantou Investment, and reportedly 10 additional investors.

Announcement

Also on the Radar

• Motorevo (泉智博) — A+ round, hundreds of millions RMB @ ¥1.6B (~$222M) val
  100,000+ joint actuator modules (the motorized joints that make robot limbs move) shipped in 2025 to MIT, Tencent RoboticsX, Leju Robotics, and Peking University. Building million-unit annual capacity online now. At ¥1.6B (~$222M) valuation for a company with institutional customers and real revenue, it trades at a fraction of the multiple applied to software-first humanoid peers.
  Investors: Shenzhen Investment Holdings leads.
  Announcement

• Pudu Robotics (普渡机器人) — New round, approximately ¥1B (~$139M) @ ¥10B (~$1.4B) val
  120,000 commercial service robots across 80+ countries, 23% global market share, 100% revenue growth in 2025. First disclosed raise in three years; opened a Dallas headquarters the same week. A three-year fundraising gap followed by ¥1B and a US headquarters in the same week points to IPO preparation, not growth financing.
  Investors: Longgang Financial Holdings and Asia Investment Capital co-lead.

• FAW Qiyi Technology (旗翼科技) — Angel round, ¥400M (~$55M) @ ¥2B (~$278M) val
  China’s largest angel-stage eVTOL investment. FAW Group’s Tier-1 suppliers Fu’ao Auto Parts (thermal management) and Fuwei Auto Components (interiors) each invested ¥150M (~$21M) to lock in parts contracts before the aircraft finishes development. These are ICE-era automotive suppliers using equity to secure roles in the eVTOL supply chain before the commercial market exists, the same pattern that played out in EVs a decade ago.
  Investors: FAW Group, Fu’ao Auto Parts, Fuwei Auto Components, Cornerstone Capital.

• Sunrise AI (曦望) — Strategic round, ¥1B+ (~$139M+) @ ¥10B+ (~$1.4B+) val
  SenseTime’s inference GPU spinoff closes its seventh financing in ~16 months, crossing ¥10B (~$1.4B) valuation. S3 chip targets 90% cost reduction and 300%+ performance gain per token, a ¥0.01 (~$0.0014) per million token price floor for inference compute in China. 150 total engineers; capital-efficient by the standards of peers with 3–5x the headcount.
  Investors: Hangzhou Capital leads; Puhua Capital co-invests.

• Juwei Technology (具微科技) — A+ round, hundreds of millions RMB @ ¥1.8B (~$250M) val
  Fourth financing round in approximately two months. Builds explosion-proof wheeled quadruped robots for hazardous industrial environments, China’s first combined explosion-proof and increased-safety certification. Every lead investor (Weiqiao Group, Binhua Chemical, Binzhou SOE) operates the exact facilities the robot is designed to enter, effectively locking in anchor deployments before the product ships at scale.
  Investors: Binzhou SOE Investment, Weiqiao Group, Binhua Chemical co-lead.

• CheeChips (奇世智能) — Angel round, tens of millions RMB
  Silicon Valley-incorporated (Mountain View HQ) with a Shenzhen global R&D center, targeting a 59-product AI-powered maternal and infant care platform. The company structure and roadmap are designed for international distribution from inception.

• Xteink (阅星曈) — Series A, tens of millions RMB (Shunwei + Jingwei + Xiaohongshu)
  Ultra-portable e-ink smartphone attachment with 2M+ global users before this institutional round. Xiaomi’s Shunwei Capital, Jingwei China, and Xiaohongshu join as investors, three platforms with overlapping consumer demographics and distribution value.
  Investors: Shunwei Capital, Jingwei China, Xiaohongshu.

• Wendun Juxin (稳顶聚芯) — Strategic investment @ ¥500M (~$69M) val
  China’s first domestic high-precision stepper (the lithography machine that prints circuit patterns onto wafers) for compound semiconductors like gallium nitride and silicon carbide, the wide-bandgap materials used in EV power electronics, 5G, and renewable energy. Shipping since September 2025 against Nikon and Canon incumbents. Projected ¥140M (~$19M) revenue for 2025 at roughly 32% net margin, profitable in its first year of shipment. Listed company Sudaweige (300331.SZ) already holds equity; the stock hit limit-up on the machine’s launch announcement.
  Investors: Yigao Capital leads; Sudaweige (300331.SZ) confirmed strategic shareholder.

• Audfly (清听声学) — Series B+, hundreds of millions RMB @ ¥2B (~$278M) val
  Parametric speaker tech that focuses sound into a directional beam, backed by Lenovo Ventures. Directional audio has no dominant commercial-scale incumbent, which puts OEM integration into consumer hardware as the likely route to scale.
  Investors: Lenovo Ventures leads; Suchang Capital, Sugao Xinjin Holdings, and others follow.

• Luquos Energy (易池新能) — Pre-A, hundreds of millions RMB
  Hong Kong-registered CUHK spinout building sulfur-iodine flow batteries, a type of large-scale energy storage where liquid electrolytes are pumped through cells, at 2x the energy density of the vanadium redox systems that currently dominate the market. Utility-scale storage targeting Southeast Asia, backed by Towngas (00003.HK) for strategic distribution and Gobi Partners for regional reach. The sulfur-iodine chemistry differentiates from vanadium incumbents, and the SE Asia focus avoids China’s crowded domestic storage market.
  Investors: Oriza Holdings and Wuxi Guolian Xizhou Fund co-lead; Towngas (00003.HK) strategic; Gobi Partners.

Stragglers

• 萝博派对 (RoboParty) — Angel, amount undisclosed @ ~$4.5M (~¥32.5M) val
  A small humanoid robot startup where the investors matter more than the amount. Shunwei Capital (Xiaomi’s fund) and Xiaomi co-invest, making this Xiaomi’s third robotics placement in a single week (leading X Square’s ¥2B Series B, backing RoboParty at angel, joining Xteink’s consumer AI hardware Series A). Xiaomi doesn’t place three hardware bets at Series B, angel, and Series A in a single week by accident. The spread from foundation-model humanoid to pre-product startup to consumer AI device looks like deliberate stack coverage.

The Bigger Picture

Why China’s Giants Are Backing the Same Robots

X Square Robot now has ByteDance, Meituan, Alibaba, and Xiaomi on a single cap table, the end state of a systematic accumulation across four rounds from May 2025 to April 2026. Sudu Technology closed a round this week at a reported $2B+ valuation, backed simultaneously by CATL, Alibaba, Tencent, Ant Group, IDG, Hillhouse, and China Life.

The logic has inverted from the shared-bike era of 2015–2016, when Alibaba and Tencent hedged across rivals, each backing different companies to ensure platform presence regardless of outcome. This week, multiple giants co-own the same company, concentrating upside rather than spreading risk. If whoever controls the robot “brain” layer controls the platform, the rational move is to own the leading candidate outright before commercial scale makes that position expensive to acquire.

The Sudu round is the stress-test of that logic taken to its limit. A company that was incorporated in May 2025 has attracted China’s most credentialed institutional investors simultaneously, at a price that values it above most Series C enterprise software companies with real revenue. Chief Technical Advisor Su Hao (苏昊) co-created ImageNet in 2009 and pioneered PointNet at Stanford, and his academic pedigree seeds the thesis that Sudu could do for robot manipulation what ImageNet did for computer vision. Either that analogy holds and $2B (~¥14.4B) underprices the eventual platform, or it doesn’t and the week’s robotics valuations are the clearest current evidence of capital running ahead of commercial reality.

X Square B round via 36kr · Sudu announcement via TMTPost · US/China humanoid valuation divergence via CNBC

Beijing’s Compute Bet Gets Coordinated

On April 21, 2026, Pedaily reported that Beijing’s Economic-Technological Development Area was formally establishing four new industrial guidance funds with a combined scale of ¥20B (~$2.8B). The four funds (covering integrated circuits and AI, robotics, biotech, and future industries) are jointly capitalized by the district’s fiscal budget and E-Town Capital (亦庄国投) as LP. On April 22, EVAS announced a ¥1.5B (~$208M) Series B led by all four of those funds simultaneously.

Beijing’s four-fund quartet didn’t evaluate EVAS and happen to co-invest. The round was the first deployed capital from a structure announced the previous day. The target is RISC-V AI chips. Arm and x86 carry US intellectual property; chip companies designing on them remain exposed to licensing dependencies that export controls can sever. RISC-V is royalty-free and carries no US IP. Eight Chinese government ministries coordinated a national RISC-V adoption push in 2025; EVAS becoming reportedly the first domestic company to mass-produce on the architecture is what Beijing’s capital apparatus is backing.

The compute coordination runs wider than EVAS. CXMT, China’s only domestic DRAM producer, took a 3.7% equity stake in AI inference chip startup AiJie Core (艾捷科芯) this week, acquiring it for ¥150M (~$21M). CXMT is targeting HBM3 production by end-2026, and an equity position in a downstream compute startup creates an anchor customer and co-development relationship before the memory market opens. Inference GPU unicorn Sunrise AI (曦望) simultaneously closed its seventh round in ~16 months at ¥10B (~$1.4B) valuation, led by Hangzhou state capital.

Three AI compute plays in one week, all with state-adjacent capital at the center. Taken together (fund structure announced April 21, EVAS round April 22, CXMT stake confirmed, Sunrise AI round closed), they show state capital coordinating across the compute stack simultaneously: RISC-V as the architecture, CXMT building the memory-plus-logic vertical, and inference efficiency as the commercial application layer.

Beijing four-fund structure via Pedaily · EVAS round via National Business Daily · RISC-V China policy via TrendForce

That’s it for this week. More next week.

Inclusion AI built a model for the agent economy, where thousands of tasks run in loops, every output token costs money, and the model that wastes the least wins. Very few labs optimize for intelligence per output token, despite it being one of the most important metrics in agentic AI. We ran tests against two comparable open-source models. Ling was the only one to complete all of them.

The Token Explosion

OpenRouter processed 20.6 trillion tokens just last week.

That’s roughly 2.4 million complete works of Shakespeare.

Per day.

The driver isn’t chatbots. It’s agents. Anthropic’s own data: agents use roughly 4x more tokens than chat interactions. Multi-agent systems use 15x. Platforms like OpenClaw are running tasks around the clock. As Anand Iyer of Canonical VC put it: “The demand for AI inference right now is coming from agentic platforms, especially OpenClaw.”

At this volume, every wasted token is real money. Claude Sonnet 4.6 alone processed 1.38 trillion tokens last week on OpenRouter. The OpenRouter State of AI report puts average prompt tokens at 6,000 per request versus 400 for completions, a 15:1 input-to-output ratio. Applied to Sonnet’s traffic, that works out to roughly $5.2M at list price ($3/M input, $15/M output) across one platform in seven days.

The Overthinking Problem

The industry’s answer to “make it better” has been “make it bigger.” The result: top models now use tens of millions to hundreds of millions of output tokens to complete the same evaluation on the Artificial Analysis Intelligence Index. The arms race has been chasing intelligence scores while ignoring what it costs to get them.

Users have noticed. “Caveman prompting” strips grammar from inputs to force shorter outputs. The instruction prompt: “Speak primitive. Use nouns and verbs. No grammar filler. Save tokens.” A formalized Claude Code skill built around the technique has 14,000 GitHub stars. Its README written in the style it teaches: “Brain still big. Cost go down forever. One rock. That it.” Across real dev tasks, it cuts output tokens by 65%. It works for coding. It breaks on anything that requires sustained reasoning.

At the corporate level, the opposite problem: tokenmaxxing. Companies began tracking employee token consumption as a productivity metric. Workers compete on internal leaderboards ranked by tokens burned, with the highest individual user averaging 281 billion tokens. Volume confused for value.

The instinct isn’t limited to prompt engineering. Marc Andreessen recently called a YouTuber with “a hundred videos on retardmaxxing” his “new life coach.” The philosophy: go to work, do a good job, come home. Don’t overthink it. Chamath Palihapitiya said the same on All-In: “Enjoy your life, work hard, and don’t overthink it.”

Caveman prompting is the demand-side version of this: users forcing models to stop overproducing. But the hack reduces cost without reducing waste. A model that produces fewer tokens because it reasons more efficiently is a different proposition from one that’s been told to shut up. What if a model just didn’t overproduce by design?

The Intelligence Yield Thesis

“Bigger is better” has peaked. The new competition is fewer active parameters, faster inference, lower cost per completed task.

Mixture-of-Experts architectures route each token to a small subset of the total parameters. What matters is how many fire per token, not how many exist. Linear attention scales computation linearly instead of quadratically, an advantage that compounds as outputs get longer.

The focus should be on outcome per token.

Ant Group and Ling-2.6-flash

Ant Group’s dedicated AGI lab was launched in February 2025, inspired by DeepSeek’s research model. Their stated vision: “AGI as humanity’s shared milestone, not a privileged asset.”

Ant Group’s DNA is financial inclusion: 140 million healthcare users on AQ, 60% from cities where specialist care barely exists; 1.69 million farmers with satellite-assessed, collateral-free loans processed in minutes. Chairman Eric Jing calls it “inclusive AI.”

Ling-2.6-flash is in our opinion one of the models on the Pareto frontier of the Intelligence Yield Thesis. (We’re also big fans of Minimax M2.7, which leads in a different quadrant).

The architecture: 104 billion total parameters, 7.4 billion active per token. MoE with 256 expert modules. Hybrid Linear Attention + Multi-head Latent Attention with linear time complexity, where the advantage grows at longer outputs. Multi-Token Prediction for parallel inference.

The efficiency: 15 million output tokens to complete the full Artificial Analysis eval. The comparison models used 5-7x more.

Before anyone knew who made it, Ling launched on OpenRouter as “Elephant Alpha.” Topped the trending charts. Hit 100 billion daily token calls. Top 8 by total calls on the platform. Without brand or marketing, the model proved itself.

First, some context on benchmark results before we dive into the tests (images from artificialanalysis.ai).

The Benchmarks

Token use

Nemotron used 110 million output tokens to complete the standardized AA Intelligence Index. GPT-OSS used 78 million. Ling used 15 million. That’s 86% fewer tokens than Nemotron and 81% fewer than GPT-OSS.

Cost

Multiply by price per token. Ling: $23. GPT-OSS: $67. Nemotron: $145.

Model Pricing

Cheap per-token pricing doesn’t guarantee cheaper outcomes. GPT-OSS charges 2x more per output token than Ling ($0.60 vs $0.30) and uses 5x more output tokens. Factor in input costs and the total comes to $67 vs $23 for the same evaluation.

Intelligence

This is the chart where Ling looks weakest. Intelligence score: 26, against GPT-OSS’s 33 and Nemotron’s 36.

The benchmark results call to mind a quote by someone who knew a thing or two about intelligence.

The AA Intelligence Index measures general reasoning, knowledge, and math. It does not measure sustained execution in loops, tool use across hundreds of steps, or the ability to finish what it starts. In testing, Ling was the only model that did.

Let the Tokens Do the Talking

Three tests, run through Claude Code: qualitative analysis, bug fixing, and visual code generation. Ling against two open-source models in the same weight class: Nemotron-3-Super-120B (12B active), Inclusion AI’s own published benchmark rival, and GPT-OSS-120B, OpenAI’s open-source entry. Same prompts, same tools, same environment. Every token tracked.

Qualitative Analysis

For this test we wanted to check qualitative verbal reasoning in addition to simple tool calling. Each model was given a one-paragraph brief and asked to grab the latest news and produce a vibe check on today’s tech: sources, an overall vibe read, companies vibing and not vibing, and a one-paragraph synthesis. Take a position, back it with reasoning, and convey it in an output file, blind-scored by Gemini 3.1 Pro on voice, judgment, and second-order thinking.

• Ling 2.6 Flash: 18/50: Finished in 2,579 tokens, 50.7 seconds. Hedged every observation, described the surface without reading beneath it.

• GPT-OSS-120B: 21/50: Similar tone. Generic vibe call, headline-obvious picks. Gemini flagged both for consultant-speak. Better discipline score kept it three points ahead.

• Nemotron 3 Super: Did Not Finish (DNF): Never started. Couldn’t parse the setup step and emitted garbled Python-serialized output instead of clean bash. Repeated the same broken command 11 times, mid-loop it ran echo and pwd successfully, proving basic bash execution was fine, then immediately returned to the broken pattern. Same serialization bug that surfaced in all three tests. 4,145 tokens spent, nothing produced.

Bug Fixing

Ten programs from QuixBugs: classic algorithm implementations, each with exactly one introduced bug. The task: find the bug, fix it, pass the tests.

• Ling 2.6 Flash: 10/10: Fixed all ten. On knapsack, a large input triggered a test harness timeout. Ling identified the performance constraint and refactored iterative DP to recursive with @lru_cache, going beyond the one-line fix to make the algorithm pass within the test harness timeout.

• Nemotron 3 Super: 5/10, DNF: A serialization bug appeared across all three tests: instead of clean code output, the model produced Python-serialized data structures. It worked around this for the first five programs. On Levenshtein distance, the bug triggered an infinite loop it never escaped.

• GPT-OSS-120B: 5/10, DNF: Fixed every program it attempted (gcd, flatten, mergesort, quicksort, find_in_sorted), all correct on first try. Then stopped. After program five, it output a summary of remaining tasks and quit. We relaunched it to confirm: same result. No error, loop, or technical failure, it simply didn’t sustain the task.

Visual Code Generation

Breakout game, single-file HTML. One prompt, no follow-ups.

• Ling 2.6 Flash: Functional: 503 lines. Glow effects, particle physics, working game mechanics. No debugging, no reconstruction, no workarounds. Played immediately.

• GPT-OSS-120B: Broken on delivery: 294 lines. The game logic was there (paddle, ball, bricks, collision, particles) but it didn’t work out of the box: the start screen overlay blocked click events from reaching the canvas, so the game couldn’t actually be started without a code fix. Simpler than Ling’s output: flat-colored bricks, minimal effects, smaller canvas. Same mid-task stall as the other two tests: froze mid-generation, required prompting to continue.

• Nemotron 3 Super: Broken: Same serialization bug as in the bug-fixing test. Produced Python-serialized data instead of HTML. The game logic was there but was buried inside a model-level format failure that affected all three tests. We reconstructed it from the raw output; the logic held up pretty well once extracted. Though there were layout bugs.

The Results

Across all three tests, Ling was the only model to complete every task without intervention. The competition failed in different ways: Nemotron’s serialization bug blocked it in every test. OSS was competent when working (every bug fix correct) but stalled mid-run across all three tests, each time requiring a prompt to continue, and its visual output didn’t work without a code fix. Completion and token efficiency moved together.

A model that uses fewer tokens per step can sustain more steps before losing coherence. The same design discipline that cuts token waste also cuts task abandonment. The tradeoff showed in qualitative analysis, where Ling scored lowest: a model trained to minimize output tokens will hedge rather than elaborate, summarize rather than argue. The same discipline that makes it fast at execution makes it flat on opinion.

The AA Intelligence Index doesn’t test what agents actually do. The benchmarks that do tell a different story. BFCL-V4 measures function calling accuracy: Ling 67.04, GPT-OSS 43.3, Nemotron 35.12. TAU2-Telecom measures multi-step task completion: Ling 93.86, Nemotron 62.94, GPT-OSS 41.45. SWE-bench Verified: Ling 62.00, Nemotron 61.00, GPT-OSS 43.80. MRCR long-context: Ling 54.01, Nemotron 39.04. Ling leads on agentic benchmarks across the board. It trails Nemotron on hard math (AIME 73.80 vs 88.59).

Different tool for a different job.

Why This Is Important

Andrej Karpathy’s autoresearch makes this concrete. A coding agent autonomously improves a training script in a loop: read, hypothesize, edit, run, evaluate, repeat. Unattended. Karpathy ran it for two days: 700 experiments, 20 discovered optimizations, 11% training speedup. Shopify CEO Tobi Lütke tried it overnight: 37 experiments, +19%.

Each cycle: roughly 18,000 tokens at about $0.10. At 700 cycles, the gap between an efficient model and a verbose one is the gap between a $70 run and a $700 run.

The same pattern applies everywhere agents run in loops: CI pipelines, code review, data monitoring, agent swarms. The emerging architectures don’t use one model for everything. Frontier models like Kimi K2.6 and Claude Sonnet are strong at complex reasoning and high-stakes decisions. But running them on every sub-task in a multi-agent pipeline is expensive by design. Kimi K2.6 at $0.75/$3.50 per million tokens, Claude Sonnet at $3/$15. Efficient models handle execution-layer throughput at a fraction of the cost.

A swarm running hundreds of sub-tasks needs each sub-agent to be reliable, cheap, and disciplined, not the smartest model available. At $0.10/$0.30 versus $3/$15, that choice determines whether a pipeline is economically viable at scale.

The cost floor dropping changes what’s possible. Always-on agents that monitor, iterate, and improve without human intervention. A researcher annotating genomic data in a continuous loop. A solo developer running hundreds of test variations overnight. A clinic in a mid-tier city running diagnostic pipelines on a budget that would not have supported it a year ago. The compute and models exist. The constraint has been a reliable model with good cost per token at volume.

Ling at $0.10/$0.30, open-source, designed to minimize waste. That’s the financial inclusion bet applied to intelligence itself.

20.6 trillion tokens last week. The agent economy is here. Efficiency is the core driver. Inclusion AI is building for it.

Where capital flows tells you where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

Where the Money Went

102 deals from April 10 to 17. $1.63 billion (~¥11.7B) in disclosed capital across 27 of them — the other 75 went undisclosed. Robotics commanded $937 million (~¥6.75B) of that, across 23 deals. This week, though, the more interesting pattern runs underneath the headline numbers: listed companies and OEM customers are entering venture rounds not as passive financial investors but as strategic acquirers, buying equity positions in their key suppliers before commercial volumes arrive.

🤖 Robotics (23 deals)

它石智航 (TARS Robotics) closed a $455 million (~¥32.8亿) Pre-A, reportedly the largest private embodied AI raise in Chinese history, with Hillhouse, HongShan, and Meituan co-leading 14 months after founding. The stage distribution across the batch forms a barbell: eight angel and Pre-A rounds ran alongside five B+ or strategic rounds, with almost nothing in between.

🧠 AI (17 deals)

The bulk of the disclosed capital — ¥5亿+ (~$70M+) — went into one company: 智象未来 (HiDream.ai), which released an MIT-licensed open-source image model before closing this round. The broader batch skews early, with Anhui, Hefei, and Beijing municipal capital prominent as co-investors.

🔬 Semiconductors (23 deals)

The sector’s headline event was not a venture round. 盛合晶微 (SJ Semiconductor, 688820.SH) listed on the STAR Market 2,700 times oversubscribed, reaching a ¥360亿 (~$5B) market cap at issue, with ¥47.79亿 (~$664M) in net proceeds. As the #1 domestic 12-inch chip bumping provider and holder of 85% share in mainland 2.5D advanced packaging, it controls a chokepoint that every Chinese AI chip designer needs to reach production. Among the private rounds, 芯擎科技 closed a B+ at approximately $100 million (~¥720M) for its automotive cockpit chip platform, pulling in Yutong Bus (600066.SH) as a strategic co-investor, extending the company’s addressable market beyond passenger EVs.

⚡ New Energy & Materials (14 deals)

A quieter week for the sector. 疆海科技 (Zendure), which sells energy storage hardware across 63 countries and counts Europe and North America as its primary markets, raised a round led by domestic strategic investors. 顺铉新材, a specialty polyimide film producer with export relationships into South Korea, Japan, the US, and Russia, closed a materials round. Neither disclosed amounts.

🚀 Aerospace & Defense (9 deals)

边界智控 (Boundary.AI) closed a B-round for its eVTOL flight control systems — the only Chinese company to have cleared CAAC’s SOI#1 airworthiness review for eVTOL-specific FCS. The rest of the batch was light: several pre-A rounds for drone platforms and satellite sensor startups, mostly undisclosed.

⚙️ Other Advanced Manufacturing (12 deals)

Mixed batch with no standout cluster. 锐能科技 raised a B-round with 菱电电控 (688667.SH) acquiring 30% as a strategic lead. Several other industrial automation plays closed early-stage rounds with regional government funds, consistent with the ongoing manufacturing digitization push in non-coastal provinces.

💡 Frontier Tech (4 deals)

量旋科技 (SpinQ) — announced April 3, flagged in this week’s straggler batch — sells superconducting quantum computers commercially into 200+ institutions across 40+ countries and raised ¥6亿 (~$83M) in a C+, several months after closing its C-round, bringing the combined C-series total to approximately ¥10亿 (~$139M). The company is profitable and building out a pre-IPO syndicate with Guotai Junan Innovation and Chengdu state capital as new investors. NovaFusionX (诺瓦聚变) appears in this week’s data with its ¥7亿 (~$97M) 天使+ round but was covered in last week’s issue; refer to that report for the full analysis.

Spotlight

TARS Robotics (它石智航)

Pre-A, $455M (~¥32.8亿), valuation ¥120.58亿 (~$1.67B)

Humanoid robotics, 14 months old, betting against the VLA consensus.

TARS Robotics builds humanoid robots for industrial precision tasks — specifically the kinds of fine-motor assembly work that previous robot generations couldn’t reliably handle. The company’s A1 robot set a Guinness World Record in December 2025 for most wire harness insertions by a robot in an hour (105 valid sub-millimeter completions), a benchmark designed to prove capability for CATL-style battery assembly lines; the company was founded in February 2025.

The technology bet is explicit. TARS is building against the VLA (Vision-Language-Action) model orthodoxy that most Chinese and US humanoid companies are pursuing, arguing that large cross-embodiment models borrowed from language AI (”外来和尚” — literally “the foreign monk always chants best,” the Chinese idiom for assuming outside expertise is inherently superior) can’t deliver reliable industrial performance. Instead, TARS runs AWE3.0 (AI World Engine), a proprietary architecture with three components: OSD for panoramic sensorimotor decision-making, HTS for high-density tactile sensing, and LAS for smooth latent-space motion generation. The architecture debate has no settled answer, and TARS is the largest private bet on the non-VLA side.

Investors: Hillhouse (co-lead), HongShan (co-lead), Meituan (co-lead), Qiming Venture Partners (returning), and 15+ additional investors including Beijing municipal funds and state capital. The “$455M Pre-A = largest embodied AI private raise in Chinese history” claim is reported by multiple independent Chinese financial outlets and is unchallenged by competing data — treat it as plausible rather than certified.

Announcement

HiDream.ai (智象未来)

New financing round, ¥5亿+ (~$70M+)

MIT-licensed image model, Q1 revenue already above full-year 2025.

HiDream.ai is a multimodal AI company that builds image generation models and related enterprise tools. Its HiDream-I1 — a 17-billion-parameter model released under an MIT license — scored at the top of HPSv2.1 benchmark testing across all four style categories (animation, concept art, painting, photo) ahead of Flux.1-dev, Stable Cascade, DALL-E 3, and Midjourney V6, per the company’s own technical benchmarking. The company says Q1 2026 revenue exceeded its full-year 2025 figure, with over 30 million registered professional users and 40,000+ enterprise clients including Huawei Cloud and TikTok Shop. Those are company-stated figures, not independently audited.

HiDream released a frontier model for free, built a global developer user base, and used that credibility to close institutional capital at a scale that enterprise SaaS revenue alone wouldn’t support at this stage. The same open-source-to-capital sequence runs across several companies in this batch — more on that in The Bigger Picture.

Founder Mei Tao was VP at JD.com and Senior Research Manager at Microsoft Research Asia; he is a foreign academician of the Canadian Academy of Engineering. The round brought in Oriental Rich Capital as a new investor alongside a state capital stack from Anhui and Hefei: Anhui Venture Capital, Hefei Industrial Investment, Hefei High-Tech Investment, and the Anhui Provincial AI Fund. Prior investors, including Anhui state capital from the previous A-round, are continuing. The round was described by media as a “new financing round” — no Series letter confirmed in the primary announcement.

Announcement

Link Touch (蓝点触控)

C+, 数亿人民币 (hundreds of millions of RMB)

72.6% humanoid six-axis force sensor market share. Customers are now shareholders.

Link Touch makes six-axis force/torque sensors and joint torque sensors for humanoid robots, with a 72.6% share of the humanoid six-axis force sensor market as of Q3 2025, according to GGII industry research. The company has doubled its revenue every year for three consecutive years, is on a confirmed IPO track per the CEO’s statement, and has shipped over 100,000 units. The C+ round came approximately five months after the November 2025 C-round led by HongShan — the company has now raised a new round three times in under a year.

The investor composition tells the supply chain story more clearly than any market share figure. The C+ brought in CATL (via its CVC arm Puquan Capital), AgiBot, and Galaxy General Robotics as new co-investors — the latter two being two of Link Touch’s largest humanoid robot customers. Two customers converted to shareholders in the same transaction, locking in supply alignment before humanoid volumes actually arrive. Autoptic and Galaxy Yuanhui rounded out the round. The structure echoes what Huawei did with semiconductor suppliers between 2019 and 2022: equity plus procurement logic, with the supplier gaining committed demand and the customer securing component access.

The valuation was not disclosed in any primary announcement.

Investors: CATL (via Puquan Capital), AgiBot, Galaxy General Robotics, Autoptic, Galaxy Yuanhui.

Announcement

Also on the Radar

极佳视界 (GigaAI) — Series B1, ¥15亿 (~$208M), ¥120亿 (~$1.67B) val. ¥25亿 (~$347M) raised in 30 days: ¥10亿 (~$139M) Pre-B in March, then this B1 in April. The company claims GigaWorld-1 ranks at the top of the WorldArena live leaderboard above Google and NVIDIA models — note that the published WorldArena paper (February 2026) evaluated GigaWorld-0, which did not place first; the #1 claim refers to a subsequent live-leaderboard submission. The Maker H01 humanoid is targeting 1,000 units in 2026. Huawei Habo invested in the A1-round; the B1 added state funds from Wuhan, Guiyang, Shandong, and Shanghai alongside an unnamed tech giant as strategic.

• PsiBot (灵初智能) — Series A, undisclosed, ¥90亿 (~$1.25B) val. Embodied intelligence company whose Psi-R2 manipulation model tops the MolmoSpace benchmark — an Allen Institute for AI robotics evaluation platform — above Physical Intelligence and other international teams, per QbitAI reporting. SDIC Pioneer (national-level industrial fund) and a Beijing municipal fund co-led the A-round; Hillhouse was the angel. A ¥90亿 valuation on an undisclosed amount, in a government-anchored round, is the investor composition doing the talking.

• BOTINKIT (不停科技) — B+/B++ combined ~$50M (~¥360M), ¥19.5亿 (~$271M) val. AI kitchen robots deployed across 29 countries at Disneyland, Jollibee, Walmart, Rewe, Foodstuffs, and LSG Sky Chefs (Lufthansa catering). The most internationally distributed robotics company in this batch, and one of the least covered in Western tech media. Investors include ZeroOne Ventures and Genesis Capital, with SSC Capital as FA. HKUST robotics faculty (Li Zexiang, Gao Bingqiang, Gan Jie) remain on the cap table from the angel round.

• Sand.ai — B-round, reportedly ~$50M (~¥360M). Sub-30 people, $10M+ ARR reported. CEO Cao Yue is the lead author of Swin Transformer (ICCV 2021 Best Paper, 30,000+ citations). The company open-sourced MAGI-1, its video generation model.

• SiEngine (芯擎科技) — Series B+, ~$100M (~¥720M), ¥32.5亿 (~$451M) val. The DragonEagle-1 (龙鹰一号) is a 7nm automotive cockpit SoC (TSMC-fabbed), #1 domestic automotive cockpit chip by 2024 shipments with 1M+ units shipped and ASIL-D certification. Yutong Bus (600066.SH) joined as strategic co-investor, extending the addressable market beyond passenger EVs. Jingming Capital led.

• Xingyun IC (行云集成电路) — Pre-A, ¥4亿 (~$56M), ¥20亿 (~$278M) val. LLM inference-first GPU with CUDA compatibility, targeting AI server mass production in 2026. Tsinghua founders (CEO Ji Yu), Zhipu AI as angel strategic investor, 5Y Capital and Primavera Capital co-leading the Pre-A. The post-DeepSeek inference demand story (where local GPU alternatives become competitive precisely on inference cost rather than training throughput) makes this the most coherent domestic GPU pitch in the current batch.

• Boundary.AI (边界智控) — Series B, over ¥100M (~$14M+), ¥5亿 (~$69M) val. The only Chinese eVTOL-specific flight control company to clear CAAC’s SOI#1 airworthiness review, with a triple-redundant dual-channel architecture meeting DO-178C/DO-254/DO-160G compliance — the certification stack that maps to FAA/EASA requirements. That’s the regulatory moat for any non-Chinese eVTOL program needing a redundant FCS supplier. Customers include several Chinese eVTOL manufacturers. Xiamen Jianfa (Xiamen SOE) and Zhangjiang Group (Shanghai science park) co-invested, reflecting geographic positioning across two low-altitude economy pilot zones.

• Simple AI (深朴智能) — Pre-A, hundreds of millions of RMB, ¥20亿 (~$278M) val. General-purpose embodied AI humanoid targeting home, hotel, and eldercare environments. Linear Capital and Puhua Capital co-led, with Shunwei Capital (Lei Jun / Xiaomi’s venture fund) participating as a strategic investor. Chief Scientist Wang Jiawei is ex-MSRA, ex-DeepSeek, and ex-ByteDance Seed with 10,000+ Scholar citations; Product VP Zhang Di brings 10+ years in robotics and 100,000+ units shipped. The Xiaomi ecosystem extending a bet on home humanoids without building in-house.

• EncoSmart (享刻智能) — Series A, ¥1.5亿 (~$21M), ¥7.5亿 (~$104M) val. Commercial kitchen robots for fried chicken shop back-kitchens via its LAVA multi-task embodied operation system. Ninebot/Segway (689009.SH) acquired approximately 20% as strategic lead, contributing supply chain manufacturing access, platform technology, and distribution channels alongside the capital — not a passive financial stake. YiBo Capital acted as FA.

The Bigger Picture

Open code, closed rounds: how Chinese AI labs turned open-source into a capital strategy

HiDream released HiDream-I1 under an MIT license, built 30 million users on it (per company figures), and closed what Anhui and Hefei state capital joined as a major financing round on the strength of that user base. Q1 2026 revenue already exceeded the full year of 2025.

This is the same sequence running across several companies in this week’s batch. 极佳视界 (GigaAI) released GigaBrain-0.1 to Hugging Face (16,000+ downloads in two weeks), positioned it as the baseline for the CVPR 2026 GigaBrain Challenge, and is closing a ¥15亿 (~$208M) B1-round while targeting 1,000 Maker H01 humanoid units this year. 灵初智能 (PsiBot) published Psi-R2 as a manipulation foundation model before its A-round — both investors and the benchmark community evaluated the model before the company raised. Sand.ai open-sourced MAGI-1 while running $10M+ ARR on enterprise video generation. CEO Cao Yue’s authorship of Swin Transformer — an ICCV Best Paper with 30,000+ citations — is part of the credibility stack that precedes the round, not a credential that appears afterward.

The pattern is not the same as 2023-era open-source gestures, when Chinese labs published models partly to signal parity with US counterparts. In 2026, releasing a frontier model is functioning as a distribution mechanism: it acquires global developer users before products exist and converts that user base into institutional funding at valuation multiples that pure enterprise software can’t reach at comparable revenue stages. The model is the top-of-funnel, and the round is the conversion event. As MIT Technology Review documented in February 2026, Chinese labs from Alibaba Qwen to MiniMax have made open-weight releases central to their global positioning, with Qwen already overtaking Meta’s Llama in HuggingFace downloads. The same logic now runs inside robotics and embodied AI rounds.

The structural context for why this works: the Stanford AI Index 2026 found the Arena Leaderboard gap between the top US and top Chinese model had closed from roughly 300 points in May 2023 to 39 points — 2.7% — by March 2026, with DeepSeek-R1 briefly matching the leading US model in February 2025. China produced 30 notable AI models in 2025, holds 74.2% of global AI patents granted in 2024, and accounts for 20.6% of global AI research citations. Open-source distribution only works as a capital formation strategy if the models are good enough to earn genuine global adoption. The data says they are.

One complication worth naming: the benchmark claims are not all equally clean. HiDream’s HPSv2.1 result comes from the company’s own technical paper — applying the HPSv2.1 scoring model to a comparison set they selected. On Artificial Analysis, an independent aggregator, HiDream-I1 ranked second in April 2026. GigaWorld-1’s WorldArena #1 claim refers to a self-submission on the live leaderboard; the published WorldArena paper (February 2026) evaluated GigaWorld-0, which did not place first. PsiBot’s Psi-R2 MolmoSpace ranking is confirmed by QbitAI reporting and is the cleanest of the three claims. The pattern holds; some of the evidence supporting it is more self-reported than independent.

Customers becoming shareholders

Five deals this week share a structure: a listed company or OEM customer takes a direct equity stake in a key supplier, typically with commercial commitments bundled into the transaction. OneRobotics (06600.HK) acquired 21% of HITBOT — China’s #1 robot dexterous hand company by 2025 shipments — for ¥2.41亿 (~$33M) in a deal implying a ¥11.46亿 (~$159M) total valuation, which represents strategic pricing rather than market rate for a sector leader. Ninebot/Segway (689009.SH) led EncoSmart’s A-round with a strategic equity position, contributing supply chain access alongside capital. Link Touch’s C+ converted AgiBot and Galaxy General — two of its largest humanoid customers — directly into co-investors alongside CATL’s CVC. Lincontrol (688667.SH) acquired 30% of RNTECH, and Yutong Bus (600066.SH) joined SiEngine’s B+ as a strategic co-investor, expanding the automotive chip company’s exposure to commercial vehicles.

The clearest historical precedent is Huawei’s Habo investment program, which between 2019 and 2022 built equity stakes in approximately 40 semiconductor suppliers using a product-plus-orders-plus-capital model: equity paired with procurement commitments, at below-market valuations, creating strategic lock-in before demand outpaced supply. Several Habo portfolio companies saw Huawei account for 70–90% of their revenue post-investment, which turned out to be a structural vulnerability when Huawei’s own position came under pressure. The robotics version of this pattern is more distributed, with no single OEM running it: Ninebot, Yutong, OneRobotics, CATL, and Lincontrol each making independent moves with the same underlying logic — buy supply chain security before commercial volumes make those positions expensive to acquire.

Some of this week’s strategic commitments may involve circular logic. Where the customer-investor is also the company’s primary revenue source and the investment is structured to validate a valuation that pure financial investors wouldn’t support, the “strategic validation” can be self-referential. The cleaner cases — Ninebot’s platform and channel commitments to a commercial kitchen robot company, CATL’s material interest in sensor supply for humanoid assembly — are ones where the strategic rationale exists independently of the deal structure. The sector is producing enough volume now to tell those cases apart.

Deal data compiled from Chinese corporate announcements, financial news, and investment databases. Amounts and valuations are as reported and may be incomplete. For informational purposes only. Not investment advice. Verify all figures independently before acting on them.

Looking around, most tables had people on their phones. Not scrolling social media. Trading. Stock tickers, candlestick charts, portfolio dashboards. An older woman two tables over was comparing notes with her friend on a position they both held. A younger guy in a polo shirt was toggling between a brokerage app and what looked like a crypto wallet. They all had smiles on their faces after riding the stock bump since FTSE Russell confirmed Vietnam’s upgrade from frontier to emerging market in its April 2026 review.

Most of the people in that coffee shop probably couldn’t tell you what FTSE Russell is. It doesn’t matter. Trillions of dollars in passive funds are benchmarked to FTSE’s indices: index ETFs, pension funds, sovereign wealth allocations. When Vietnam moves from frontier to emerging market, those funds are required to buy Vietnamese equities to maintain their index weighting. Active managers follow, chasing the inflows. The retail investors caught the front of an institutional wave.

After my coffee, my new friend Trung Pham (Gavin) and I hopped on a scooter for a short ride along the riverfront north to his office at the new International Financial Centre, Da Nang. It was here that Da Nang’s government-licensed regulatory sandbox had recently approved Vietnam’s first pilot for converting USDT to Vietnamese dong (VND). The country already has a massive retail investor base trading equities, crypto, everything. The IFC is the government’s bet that it can build new financial infrastructure on top of that energy.

I’ve been in China over a decade. I went to Vietnam to test a specific hypothesis: that the country would be following a pattern similar to China’s, where infrastructure and regulation get built simultaneously, and fast. What I found was even more impressive than the China parallel I had in mind. Vietnam already has full, open internet. Its IFC operates a specialized court with common law principles and English-language proceedings (it even has international judges), and is closer to Dubai’s DIFC than to anything in mainland Asia, borrowing selectively from the best jurisdictions globally and building something distinctly its own.

Vietnam is borrowing selectively from the best jurisdictions globally and building something distinctly its own.

The Financial Laboratory

Gavin works at the Da Nang Fintech Lab (DFL), the government-licensed regulatory sandbox operator sitting inside VIFC-DN, Vietnam’s International Financial Centre in Da Nang. The -DN designation is to distinguish it from its sister organization in Ho Chi Minh City.

DFL holds the actual regulatory rails and the API infrastructure (banking, e-wallet integration, KYC/AML stack) that lets ventures land and operate inside the sandbox. It even provides funding to startups in the sandbox.

Gavin is also an entrepreneur and the leading case study. His team is building a crypto accounting platform that does deep transaction-level tracing across inputs and outputs, producing compliance-ready reports. That kind of tooling matters because of what’s sitting just offshore.

Bringing $220 Billion Home

Vietnam recorded over $220 billion in crypto asset flows in 2025, up 55% year-on-year. Some 21 million adults hold or have used crypto, and Chainalysis ranks the country fourth or fifth globally in adoption.

Almost all of that activity happens on offshore exchanges, outside Vietnamese regulation, untaxed and unsupervised. The government wants to bring it onshore, and it has been moving fast at the national level.

The Politburo’s Resolution 57 (December 2024) named digital transformation a national strategic priority. The Law on Digital Technology Industry followed in June 2025, passed 441 to 4, covering semiconductors, AI, and digital assets in a single framework. A five-year crypto sandbox launched in September with plans for five licensed domestic exchanges, and the Ministry of Finance began accepting applications in January 2026 with a minimum capital requirement north of $400 million. OKX and HashKey jointly invested in a Vietnamese entity earlier this month to meet the bar, though foreign ownership is capped at 49%, keeping control local. The IFC is where much of this gets built.

The Tokenization Opportunity

Once the exchanges are licensed and the capital comes onshore, the question becomes: what does it invest in? Not just trading crypto back and forth. Regulated infrastructure makes tokenization viable: representing real-world assets (bonds, equities, fund units) on-chain. A retail investor buys a fractional slice of a government bond for a few hundred dollars instead of needing a brokerage account and a high minimum. Equities trade 24/7 instead of within exchange hours. Fund units settle instantly instead of T+2. Compliance rules get embedded in the token itself, so KYC and transfer restrictions travel with the asset.

Globally, this is already moving fast. BlackRock’s BUIDL fund, a tokenized money market product issued through Securitize, holds over $2 billion in assets. JP Morgan’s Kinexys platform has completed tokenized Treasury trades with Ondo Finance and Chainlink, and bought $100 million in tokenized commercial debt from Singapore’s OCBC. The SEC issued a joint taxonomy of tokenized securities in January 2026, and US banking regulators published guidance on capital treatment the following month. McKinsey projects $2 to 4 trillion in tokenized assets by 2030. BCG puts the bull case at $16 trillion. Closer to Vietnam, Singapore’s MAS has 40-plus financial institutions in its Project Guardian tokenization program and is testing tokenized government bills for interbank lending. Thailand issued a 5 billion baht tokenized government bond for infrastructure investment.

Vietnam has something most of these jurisdictions don’t: a large, active retail base, a government writing the rules in real time, and a sandbox where new financial products can be tested under license. People are already building for it.

Cathal Donnellan, a fellow Irishman who has spent years building crypto and fintech platforms across Southeast Asia, put it simply in a conversation we had before the trip: the old speculation model is, thankfully, mostly dead, and the future is real-world assets. He’s now driving asset tokenization in Vietnam with NexStox.

Gavin’s compliance platform, already operating inside the sandbox, is built for exactly this kind of market. Tokenized assets need transaction-level tracing and compliance-ready reporting baked into the infrastructure, not bolted on after the fact. The fact that DFL and VIFC have ventures building real products inside a sandbox that barely existed a year ago suggests the model is working as designed: stand up the infrastructure, and the builders show up.

Ten years from now, the retail investor sitting in that Highlands Coffee won’t just be trading stocks and crypto on separate apps. They may hold tokenized government bonds, fractional fund units, and international securities in a single wallet, settled on-chain, regulated onshore.

A Tale of Two Cities

The IFC itself operates as a dual-hub model, formalized in Decree 323/2025. Rich McClellan, CEO of VIFC-HCMC and Vice Chair of the Da Nang hub, walked me through the structure before my trip. Rich’s side of the house focuses on traditional capital markets, corporate finance, and asset management, and he’s been landing partners fast. Nasdaq is one of seven founding members alongside VinaCapital, Sovico Group, and others. HDBank and the London Stock Exchange signed a partnership in April 2026 for cross-border fundraising, connecting Vietnamese issuers with global institutional investors. Rich’s team also launched an Asia-Pacific Aviation Financial Hub at the Singapore Airshow with Airbus and Boeing as strategic members, targeting $6.1 billion in initial capital mobilization. HCMC climbed 11 places to 84th in the GFCI 39 rankings and was named one of 15 centres likely to become more significant.

VIFC-DN handles the experimental side: fintech, digital assets, trade finance, specialized lending, and new exchange models. I only had time to visit Da Nang this trip, so that’s where most of this piece is focused, but the two hubs are complementary.

The VIFC-DN sandbox can issue licenses for up to five years through its own executive authorities, not national regulators. Ventures can test things inside Da Nang that would be blocked anywhere else in the country. The first project through the door, Basal Pay, runs a controlled USDT-to-VND conversion under a non-custodial model on a 36-month pilot. Tether signed an MoU with Da Nang’s People’s Committee in November 2025 to support blockchain-enabled governance, digital asset policy, and tokenization of real-world assets. Corporate tax inside the IFC is capped at 10% for up to 30 years, half Vietnam’s standard rate.

This follows the path Dubai’s Virtual Assets Regulatory Authority (VARA) carved starting in 2022. VARA didn’t try to squeeze crypto into existing securities or banking law. It built a new framework around what was already happening. Vietnam is applying the same logic to a much larger market.

The Industrial Base

Chips Next Door to Fintech

Right next door to DFL inside Da Nang’s IFC, VSAP Lab is building a $69 million chip packaging technology facility in Da Nang Software Park. Construction runs through Q4 2026, with capacity for 10 million semiconductor products annually and a 50-year operational license.

Chip packaging is where individual chips get assembled, stacked, and connected into finished products. VSAP Lab is targeting advanced packaging (fan-out, 2.5D, and 3D), the same technology used to build AI accelerators like NVIDIA’s H100. Vietnam already has serious semiconductor assembly investment from multinationals: Amkor is building a $1.6 billion plant in Bac Ninh, Hana Micron has committed around $930 million. But virtually all of that is conventional packaging. VSAP Lab would be the first domestically-led facility attempting advanced packaging, and advanced packaging is increasingly the bottleneck for global AI chip supply. Vietnam’s share of global assembly, test, and packaging is projected to rise from roughly 1% in 2022 to 8-9% by 2032.

The same Law on Digital Technology Industry that created the digital assets framework also covers semiconductors and AI. Da Nang is assembling financial infrastructure, semiconductor capacity, and regulatory sandboxes in the same district. That concentration is deliberate, and it’s the kind of stacking that compounds.

The Factory Floor

The IFC and the sandbox are the financial layer, and Vietnam’s industrial policy treats them as part of a larger stack. The four Politburo resolutions issued since August 2024, covering tech self-reliance, international integration, legal reform, and private sector development, are a coordinated push to move the country up value chains.

Samsung produces roughly half of all Galaxy smartphones in Vietnam, spread across factories in Bac Ninh and Thai Nguyen provinces. Intel’s largest chip assembly and test facility globally sits in Ho Chi Minh City. Vietnam’s electronics exports overtook garments as the country’s top export to the US for the first time in 2025.

Disbursed FDI hit $27.6 billion in 2025, a five-year high, with 81.7% of newly registered capital flowing into manufacturing and processing. Q1 2026 registered FDI surged 42.9% year-on-year to $15.2 billion. Singapore and China were the top two sources of new investment. China hitting number two is worth pausing on: Chinese manufacturers are doing their own China+1, moving production lines into Vietnam to diversify supply chains and hedge against trade friction.

The tariff situation almost put the kibosh on the whole plan. On April 2, 2026, the White House announced a 46% reciprocal tariff on Vietnamese goods. General Secretary To Lam called Trump directly and offered zero-for-zero tariffs. A 90-day pause followed, and negotiations are ongoing. Where the final rate lands is the single biggest near-term variable for Vietnam’s export-oriented manufacturing. But the rest of the industrial buildout (chip packaging, financial infrastructure, rail) isn’t waiting on that answer, and some of it exists precisely because the world is diversifying away from China, not toward it.

A $75 Billion Rail Network

On April 12, Vingroup’s VinSpeed broke ground on Vietnam’s first high-speed rail line, the $5.6 billion Hanoi-Quang Ninh route, 120 kilometers at 350 km/h using Siemens Velaro Novo trains, cutting travel time to 33 minutes by 2028. The line stitches together the Red River Delta’s industrial spine: Hanoi, Bac Ninh (where Samsung’s factories sit), Haiphong port, and Ha Long Bay, collapsing a 2-hour drive into a commute and opening a major tourism corridor at the same time.

It’s also explicitly designed as a pilot for the national network: Vietnam is using this line to build the operational experience before tackling the $67 billion North-South HSR approved by the National Assembly in November 2024, which is also moving into construction. Separately, in February 2025, the National Assembly approved an $8 billion rail link to China: 427 kilometers connecting the border at Lao Cai to Hanoi, Haiphong, and Ha Long City, targeting completion between 2026 and 2030. During Xi Jinping’s state visit in April 2025, the two countries agreed on feasibility studies for two additional intermodal railways, from Guangxi to Hanoi and from Shenzhen to Haiphong, with China funding the research phase.

High-speed rail restructured where capital, talent, and industry clustered inside China over the past two decades. Cities that were too far from economic centers to compete became viable manufacturing and service hubs almost overnight. Shanghai, Hangzhou, and Suzhou were always nearby, but HSR turned them into the Yangtze River Delta megaregion. The same thing happened with Shenzhen and Guangzhou in the Greater Bay Area, and Beijing and Tianjin in Jing-Jin-Ji. The same dynamic is available on the China-Vietnam corridor. A functioning HSR link between the two countries would do for cross-border economic integration what China’s domestic network did for its internal geography.

But there are limits to what government-led investment can build.

Fresh Ideas and Spoiled Food

Vietnam has gaps at every layer of the stack, and the biggest ones aren’t in regulation or capital. They’re cultural and structural. Coincidentally, conversations about agriculture showed the clearest examples of both.

The Fruit Nobody Brands

The government can build chip plants, lay rail, and write sandbox rules. What it can’t do is manufacture entrepreneurial instinct. Trung’s view on Vietnamese entrepreneurship was blunt. Most Vietnamese, he said, are raised with an employee mindset. The few who go entrepreneurial tend to chase quick cash.

His example was fruit. Take Vietnamese tropical fruit, dry it, cut it, brand it, and the price quadruples. Almost nobody does it. I pushed back: why doesn’t someone just import the processing machines from China and start capturing that margin locally?

Chinese entrepreneurs, he said, have a more scientific approach to business: doing deep market research, finding what specific areas produce what, and have decades more experience building supply chains. They pull raw inputs into China, process them domestically, and control distribution from there. Vietnamese operators trying to capture that value-add locally are competing against that experience gap. For now, most keep shipping raw product.

What Rots Before It Ships

This isn’t limited to fruit. Across the Mekong Delta’s perishables, the same gap appears at every stage. The delta produces roughly 24 million tons of rice annually; post-harvest losses from harvest to milling run 12 to 14 percent, over $600 million in lost value each year. For fruit and vegetables, losses reach 35 to 45 percent, mostly from bruising during transport and the absence of on-site cold storage. Seafood loses 15 to 20 percent of its value if it isn’t processed and frozen within hours of harvest. Nationally, food losses total 8.8 million tons a year, around $3.9 billion, roughly 12 percent of Vietnam’s agricultural GDP.

None of this is exotic to fix. Cold storage between farm and processor, basic processing capacity, a brand on the output. The gaps are real and the fixes are known. Vietnam has the volume; what it doesn’t yet have is capital and operators willing to deploy it.

Conclusion: The Concrete is Still Wet

Vietnam is not the next China. No country is. But after a week in Da Nang, I think it’s the most interesting place in Southeast Asia to be building right now. The people I met there aren’t waiting for the framework to be finished. Gavin is building compliance rails inside the sandbox, the DFL team is operating actual regulatory infrastructure, the VSAP Lab team is putting semiconductor capacity next door. They’re building alongside the framework, not after it.

None of this is guaranteed. HSR programs blow budgets, the grid hasn’t been tested at data center scale, and the sandbox is months old. But Vietnam broke ground on high-speed rail on April 12, its financial sandbox approved its first crypto conversion pilot four months ago, and FTSE Russell just confirmed the emerging market upgrade. That’s a lot of momentum landing in the same window.

What I’m Watching

Vietnam as Southeast Asia’s fintech sandbox. This is the big one. Vietnam has 21 million crypto holders, a government writing regulation around activity rather than ahead of it, and a sandbox where ventures can test products under license that would be blocked anywhere else in the country. Singapore’s framework is more mature, but Vietnam’s domestic market is twenty times larger and the cost of building there is a fraction. If the sandbox model works, and the early signs suggest it does, Vietnam could be where you build and test fintech products before scaling them across ASEAN’s 700 million people.

Institutional digital asset products. Once the exchanges are licensed and the capital comes onshore, the infrastructure is there for tokenized bonds, equities, and fund units to trade on regulated rails. The IFC’s common law court and English-language proceedings make it viable for cross-border products: a Vietnamese investor buying tokenized international bonds, or foreign capital accessing tokenized Vietnamese government debt. One interesting experiment further out: Vietnam doesn’t have freehold property. All land is leased from the state under time-bound use rights, 50 years for commercial, up to 70 for residential. Those rights are standardized, time-limited, and already in a government registry. Fractional tokenized land use rights could make one of Vietnam’s largest asset classes tradeable without requiring a change in property law.

Private infrastructure. The government is building rail, semiconductor parks, and financial architecture. What’s missing is the boring stuff in between: cold storage near farms, processing capacity near source, logistics that don’t destroy a third of the product before it reaches a buyer. None of this needs a sandbox or a license. It needs capital willing to build warehouses instead of apps.

Vietnamese brands. Vietnam exports volume. The margin goes elsewhere. Agriculture is the starkest example, but the gap applies more broadly. The entrepreneurial shift Trung described, from employee mindset to building something, is what turns a country that grows and assembles into one that sells finished products under its own name. The government can’t manufacture that. It either emerges or it doesn’t.

In closing, Ho Chi Minh declared independence from France in 1945, quoting Jefferson as he did it. Today, his bust overlooks a fintech sandbox running common law courts in English, licensing crypto exchanges, and writing the rules for tokenized securities. Eighty years from independence to a fintech sandbox with common law courts.

Not bad.

Where capital flows tells you where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

Where the Money Went

103 funding events across seven sectors. Over $1.4 billion (~¥10.1B) in disclosed capital, with the real total significantly higher given that 50+ deals went undisclosed. Alibaba led a ¥2 billion (~$293M) strategic round into video generation platform Shengshu (Vidu). Nova Fusion raised ¥700 million (~$97M) at angel+ stage for nuclear fusion, on top of a ¥500 million angel round last year. Jack Ma and Lei Jun made their first known co-investment in Spirit AI, a humanoid robot company now valued at ¥15 billion (~$2.1B).

Chart note: Light-shaded bars represent conservative estimates where only a vague range (e.g., “数亿人民币” / “hundreds of millions”) was reported.

🤖 Robotics (16 deals)

Fewer deals than last week’s 27, but the individual rounds run larger. Spirit AI (千寻智能) pulled in ¥1 billion (~$139M) for an A+ round, part of two rounds totalling ¥3 billion (~$417M) within 30 days. ENGINEAI (众擎机器人) closed $200 million (~¥1.44B) with Luxshare as strategic co-lead. D-Robotics (地瓜机器人) raised $150 million (~¥1.08B) for its robot OS platform, a Horizon Robotics spinout building the infrastructure layer for humanoid development. The early-stage pipeline remains active: five angel/seed rounds, including a Li Auto AI team spinout (斜跃智能) founded just two months ago.

🧠 AI (17 deals)

Alibaba led a ¥2 billion (~$293 million) strategic round into Shengshu Technology (生数科技), maker of the Vidu video generation platform. That single deal accounts for most of the sector’s disclosed capital. The distribution otherwise skews early: Synapx (章鱼动力), founded December 2025 by four ex-Horizon Robotics executives, raised hundreds of millions at a ¥3 billion (~$417M) angel valuation for “physical AGI.” Red Bear AI (红熊AI) closed Series A with ¥135 million (~$19M) in revenue and 13% margins, one of the few profitable Chinese AI agent companies at this stage. ModelBest (面壁智能) hit E-round unicorn status for edge-side foundation models.

🔬 Semiconductors (14 deals)

Deals span analog and mixed-signal IC, MEMS sensors, power semiconductors, and storage chips. Most rounds are Series A or earlier with government guidance fund participation, consistent with the ongoing self-sufficiency push. Lingdong Jiaxin (灵动佳芯) closed an A round backed by SMIC Capital (中芯聚源) and others.

⚡ New Energy & Materials (19 deals)

Zhide Battery (致德新能源) raised ¥600 million (~$83M) at Series E for silicon-carbon anode materials, with CATL as an existing investor. JNION (珈钠能源) appeared for the second consecutive week, raising an A+ round for sodium-ion cathode materials with EVE Energy (亿纬锂能) backing. New materials companies (安迈特科技 with Toray strategic, 环西汀新材, 方硕材料, 启烯材料) accounted for roughly a third of the sub-sector.

🚀 Aerospace & Defense (5 deals)

Spacety (天仪研究院) raised ¥1.3 billion (~$181M) in a pre-IPO round for its 28-satellite SAR constellation, the sub-sector’s standout. Dezhihangchuang (德知航创) closed an A+ round with Sichuan provincial defense-industry funds. Huanyuhaoxing (寰宇昊行) and Gongda Satellite (工大卫星) filled out a quiet but steady week for space and defense.

⚙️ Other Adv. Manufacturing (22 deals)

Optoelectronics and photonics accounted for roughly ten deals, including Aochuang Photonics (奥创光子) at Series D and Saifulesi (赛富乐斯) raising ¥300 million (~$42M) for micro-LED. Three deals covered the lithium niobate photonics value chain from substrate to chip design. The remainder spans industrial automation, precision control, smart manufacturing equipment, and a tail of companies where names alone do not reveal the sub-sector.

💡 Frontier Tech (10 deals)

Ten deals, with capital concentrated in a few large rounds. Nova Fusion (诺瓦聚变) leads the category with a ¥700 million (~$97M) angel+ round for FRC nuclear fusion targeting AI data center power. Four to five brain-computer interface deals (应和脑科学, 聚芯医疗, 脑次元, 曦嘉医疗, 瓦博科技) landed in the same week, following BCI’s first appearance in the Government Work Report. Juxin Medical (聚芯医疗) closed a ¥200 million (~$28M) C round for endovascular BCI.

Spotlight

Spirit AI (千寻智能)

A+ Round, ¥1 billion (~$139M), valuation ¥15 billion (~$2.1B)

Full-stack humanoid robotics.

Spirit AI builds humanoid robots for manufacturing deployment. CEO Han Fengtao (韩峰涛) previously co-founded and served as CTO of Rokae Robotics (珞石机器人), where he oversaw delivery of 20,000+ industrial robots. Chief Scientist Gao Yang (高阳) is a Tsinghua assistant professor with a Berkeley PhD, supervised by Pieter Abbeel. Their Moz1 robot (26 degrees of freedom, full-body force control) is already deployed on CATL’s battery pack production line at the Zhongzhou base, running connector insertion at a 99%+ success rate and 3x human throughput. Bairui Capital (柏睿资本), founded by CATL co-founder Li Ping, invested at the angel stage, and the production-floor deployment followed.

Spirit raised ¥3 billion (~$417M) across two rounds in roughly 30 days in early 2026, on top of seed and angel rounds that had already brought total funding past ¥4 billion (~$556M). This A+ round was co-led by Shunwei Capital (顺为资本, Xiaomi founder & CEO Lei Jun’s fund) and Yunfeng Fund (云锋基金, Jack Ma’s fund), their first known co-investment in the embodied AI sector. Dachen Venture Capital (达晨财智), Turing Fund (图灵基金), and others joined. The February round had brought in Chaos Investment (混沌投资, hedge fund manager Ge Weidong), Sequoia China, and TCL Ventures.

In January 2026, Spirit open-sourced its VLA foundation model Spirit v1.5, which ranked first on the RoboChallenge global embodied AI benchmark, surpassing Pi0.5. The company claims its integrated force-control joints have 15% higher power density than Tesla Optimus, though no third party has independently verified this.

The company was founded in January 2024 and reached a ¥15 billion (~$2.1B) valuation within fourteen months.

Shunwei Capital (顺为资本, co-lead), Yunfeng Fund (云锋基金, co-lead), Dachen Venture Capital (达晨财智), Turing Fund (图灵基金), Xinding Capital (新鼎资本), Yinhe Yuanhui (银河源汇), and others. Gaohu Capital (高鹄资本) and Gengxin Capital (庚辛资本) as financial advisors.

Announcement

Nova Fusion (诺瓦聚变)

Angel+ Round, ¥700 million (~$97M), valuation ¥3.15 billion (~$438M)

FRC nuclear fusion for AI data center power.

Nova Fusion is building compact modular fusion reactors using field-reversed configuration (FRC) magnetic compression, a linear (not toroidal) design that requires smaller magnetic fields and produces more compact systems than traditional tokamaks. The target application is AI data center power, where training cluster power demand is outpacing grid capacity. The company’s roadmap runs from 100 million degree ion temperature through Q>1 (net energy gain) to 50MW commercial output.

Founder Guo Houyang (郭后扬) co-founded Energy Singularity (能量奇点), China’s other prominent fusion startup, and served as its CTO before departing to start Nova Fusion in April 2025. He holds a PhD from Université du Québec (1993) and has over 30 years in magnetic confinement fusion research across Canadian and US institutions. He currently serves as Deputy Director of Shanghai’s Future Industry Advanced Nuclear Energy Expert Committee. At Energy Singularity, he built the team that produced the Honghuang 70, a high-temperature superconducting tokamak. Nova Fusion’s FRC approach diverges from the tokamak design he worked on previously.

¥700 million (~$97M) at angel+ stage, on top of a ¥500 million (~$69M) angel round closed in August 2025, bringing total funding past ¥1.2 billion (~$167M) for a company founded just over a year ago. ENN Fusion (星环聚能) raised ¥1 billion (~$139M) in a January 2026 A round using a different magnetic reconnection approach, taking the sector’s single-round record. Chinese private fusion is entering what 21st Century Business Herald called “密集砸钱时代” (”intensive money-throwing era”), with policy tailwinds from both the 14th and 15th Five-Year Plans including commercial fusion in national industrial planning.

Meituan Longzhu (美团龙珠), Hillhouse Capital (高瓴资本), Shangqi Capital (尚颀资本), Huichuan Venture Capital (汇川产投), Jiukun Ventures (九坤创投), with Legend Capital (君联资本), Gaorong Capital (高榕创投), and Lightspeed China (光合创投) following on from the angel round. Per ITJuzi, Guangzhou Industrial Investment (广州产投) and Xiamen C&D (厦门建发) also participated.

Announcement

Spacety (天仪研究院)

Strategic Round, ¥1.3 billion (~$181M) cumulative, valuation ¥7.8 billion (~$1.08B)

Commercial SAR satellite constellation.

Spacety operates China’s first commercial SAR (synthetic aperture radar) satellite constellation, with 28 satellites in orbit and a planned build-out to 120. The constellation targets 0.5-meter resolution imagery with an average revisit cycle of 11 minutes for any point between 60°N and 60°S. The company’s millimeter-level surface deformation monitoring serves construction, transportation, power grid, and mining sectors.

Founded in 2015 in Changsha, Spacety has deployed its SAR data in 62 disaster response operations across Bangladesh, Indonesia, Turkey, and other countries, making it one of the few Chinese commercial space companies with meaningful international operations. The company started IPO preparation in February 2026, filing for A-share listing.

This latest strategic round brings cumulative funding to ¥1.3 billion (~$181M) across multiple rounds, with a reported valuation of ¥7.8 billion (~$1.08B). The capital will fund scaled satellite manufacturing, commercialization of data services, and building an integrated develop-manage-use chain.

Bohua Capital (博华资本), Puhua Capital (普华资本), Turing Capital (图灵创投), Guoxin Guozheng (国新国证), Xiangjiang State Investment (湘江国投), and others. Yibai Capital (义柏资本) as financial advisor.

Announcement

ENGINEAI (众擎机器人)

Series B, $200 million (~¥1.44B), valuation ¥13 billion (~$1.8B)

Full-stack humanoid robots at consumer price points.

ENGINEAI ships three humanoid robot models. The SE01 (170cm, 55kg, 32 degrees of freedom, 330 N·m max joint torque, 2m/s walking speed) opened for global sales in December 2024 at $20,000-30,000. The PM01 is lighter and cheaper, currently listed at $13,700, and gained attention for what the company says was the first humanoid front flip, a stunt covered by New Atlas, Interesting Engineering, and other Western tech outlets. The T800 (173cm, 75kg, 450 N·m peak torque, NVIDIA Jetson Thor, 2,000 TOPS), unveiled at CES 2026 at a $25,000 starting price, targets heavy industrial deployment with first shipments planned for mid-2026. Founded October 2023 in Shenzhen by Zhao Tongyang (赵同阳).

The PM01 at $13,700 sits near Unitree’s G1 base model ($13,500) and well below current pricing estimates for Western humanoids like Figure AI’s commercial deployments. Luxshare Precision (立讯精密), which assembles Apple’s iPhones and AirPods, co-led this round as a strategic investor, bringing the same supply chain engineering that drove consumer electronics cost curves downward into humanoid manufacturing. Henan Investment Group (河南投资集团) co-led through its Huirong Fund (汇融基金). Cornerstone Capital (基石资本), Longgang Jinkong (龙岗金控), Duolun Technology (多伦科技), and others participated.

The company has described its commercialization strategy as deliberately slow, prioritizing technical depth over a rush to scale, though $200 million (~¥1.44B) of fresh capital and Luxshare on the cap table may test that approach.

Henan Investment Group / Huirong Fund (河南投资集团/汇融基金, co-lead), Luxshare Precision (立讯精密, co-lead/strategic), Cornerstone Capital (基石资本), and others.

Announcement

Shengshu Technology / Vidu (生数科技)

Strategic Round, ¥2 billion (~$293M)

Video generation AI.

Shengshu Technology builds Vidu, a video generation platform built on U-ViT (Universal Vision Transformer), an architecture the team published in September 2022 that predates OpenAI’s DiT architecture used in Sora. Chief Scientist Zhu Jun (朱军) is a Tsinghua University computer science professor who co-launched the company in partnership with the university’s AI Research Institute. The platform hit 10 million users within 100 days of its July 2024 global launch. Vidu Q3 Pro currently sits 7th on the Artificial Analysis text-to-video leaderboard, behind ByteDance Seed, Kling, and several others. Vidu 2.0 (January 2025) generates 4-second clips in under 10 seconds at roughly $0.04 per second, 55% below the industry average. The platform has generated over 100 million video clips to date.

This is Alibaba’s second AI video bet, following a $60 million+ (~¥432M+) round into PixVerse in January. The Shengshu investment sits within Alibaba’s ¥380 billion ($53 billion) three-year AI and cloud infrastructure commitment. Alibaba Cloud revenue grew 26% to a three-year high, with AI product revenue posting triple-digit growth for eight consecutive quarters. The company frames its interest in Shengshu around “general world models” for sensory simulation toward AGI in physical environments, a broader bet than video content creation alone.

Alibaba Cloud (阿里巴巴, lead), Andon Haitang (安东海棠), China Internet Investment Fund (中国互联网投资基金), TAL Education Group (好未来), and Luminous Ventures.

Announcement

Also on the Radar

Red Bear AI (红熊AI) — Series A, ¥210 million (~$29M), ¥1.5 billion (~$208M) valuation. AI agent interactive platform with episodic memory architecture. ¥135 million (~$19M) in revenue at 13% net margin, making it one of the few profitable Chinese AI agent companies at this stage. CEO Wen De (ex-Alibaba, Fosun). Plans to expand internationally. Huayu Venture Capital (华禹创投, lead), Jiaming Haochun (嘉铭浩春), Jiawo Capital (稼沃资本), and others.

AWOL Vision (海高特) — Series B, close to ¥100 million (~$14M), ¥500 million (~$69M) valuation. Ultra-short-throw laser projectors. Two Kickstarter campaigns exceeding $10 million each under the Valerion brand. Miami office, US retail distribution, $1,899-5,999 price range. Western go-to-market further developed than any other company in this week’s data. Tiantang Silicon Valley (天堂硅谷), Huichang Technology (会畅科技), and Jinpeng Jia (金鹏佳).

Mizzen Insight (觅深科技) — Angel, close to $10 million (~¥72M), ¥260 million (~$36M) valuation. AI-powered user research platform that automates deep-dive customer interviews. 300 customers in four months, already deployed with overseas e-commerce and EV companies. Founded by Dr. Sun Keqiang (Tsinghua CS, CityU HK PhD). Sequoia Seed Fund (红杉种子基金, lead), Dachen Venture Capital (达晨财智), Jiacheng Capital (嘉程资本), and others.

ModelBest (面壁智能) — Series E, hundreds of millions of RMB, ¥9.5 billion (~$1.32B) valuation. Edge-side general foundation models. Tsinghua-origin unicorn partnered with China Telecom for AIPC deployment. Shenzhen Capital Group (深创投, co-lead), Huichuan Venture Capital (汇川产投, co-lead), and others. Huaxing Securities (华兴证券) as FA.

Synapx (章鱼动力) — Angel, hundreds of millions of RMB, ¥3 billion (~$417M) valuation. “Physical AGI” targeting the “Physical AI Turing Test.” Founded December 2025 by four ex-Horizon Robotics executives: Du Dalong (ex-CTO Jianzhi), Liang Zhujin (ex-VP), Pan Yangjia (ex-CFO), Fan Qingyuan (ex-VP). K3 Ventures (lead), Shunwei Capital (顺为资本), Xiaomi Group (小米集团), and others.

Holistic (霍里思特) — Series C, ¥200 million (~$28M), ¥1 billion (~$139M) valuation. X-ray AI ore sorting with 30%+ domestic market share across five ore types. All-Tsinghua founding team, founded 2010. China Merchants Capital (招商局资本, lead), Beichuang Investment (北创投), and others.

Enlight Medical (应和脑科学) — Strategic, undisclosed amount, ¥500 million (~$69M) valuation. Full proprietary neuromodulation and BCI platform covering electrodes, chips, signal decoding, and algorithms. Spun out from Yingmai Medical. Eli Lilly Asia Ventures reportedly co-led (per press; ITJuzi lists only Jiangxia Sci-Tech Investment (江夏科投)).

LEAPTIC (光子跃迁) — Angel, undisclosed. Corporate spinout from Dreame Technology (追觅科技, $2.8 billion valuation), building an 8K AI action camera at 55 grams, shipping at $440 versus GoPro at $400. Won a GMarks Design Award in Japan. Over 100 staff, already shipping. Sky Workshop Ventures (天空工场创投).

JNION (珈钠能源) — Series A+, hundreds of millions of RMB, ¥1.5 billion (~$208M) valuation. Sodium-ion battery cathode materials. Built the world’s first 10,000-tonne polyanionic cathode production line (December 2024). Founded by Prof. Cao Yuliang (Wuhan University). Shenzhen Capital (深圳资本) and EVE Energy (亿纬锂能).

Zhide Battery (致德新能源) — Series E, ¥600 million (~$83M), ¥3.6 billion (~$500M) valuation. Silicon-carbon anode materials for lithium batteries, solving the silicon swelling problem through carbon coating. CATL is an existing investor. Chaoxi Capital (朝希资本, co-lead), Dachen Venture Capital (达晨财智, co-lead), and others.

The Bigger Picture

Five BCI deals in one week

Brain-computer interface appeared in the 2026 NPC Government Work Report (政府工作报告) for the first time, listed alongside quantum technology and embodied intelligence as a priority future industry. Five neuro/BCI deals followed in the same week.

The deals cover different layers of the stack. Juxin Medical (聚芯医疗) raised ¥200 million (~$28M) at Series C for minimally invasive endovascular BCI and neuro-intervention devices, with Zhongshan Jinkong (中山金控) leading and Guangdong Finance (粤财创投) participating. Enlight Medical (应和脑科学) took a strategic round at a ¥500 million (~$69M) valuation for a full proprietary neuromodulation platform covering electrodes, chips, signal decoding, and algorithms, reportedly with Eli Lilly Asia Ventures involvement alongside Jiangxia Sci-Tech Investment (江夏科投). Xijia Medical (曦嘉医疗) raised an angel round for brain connectomics AI, with its NuraSurgical platform deployed in 120+ overseas hospitals across 10,000+ neurosurgery cases. Brain Dimension (脑次元) closed a ¥60 million (~$8.3M) strategic round for brain-science health services. WABO Technology (瓦博科技) took seed funding for consumer non-invasive BCI wearables.

The five deals this week follow a regulatory milestone: in March, Neuracle Medical Technology (博睿康) secured China’s first-ever approval for an implantable BCI system, a coin-sized wireless device placed on the brain’s surface to restore hand motor function in spinal cord injury patients. Per the National Medical Products Administration, it was the first globally approved invasive BCI available as a commercial product. Separately, StairMed (阶梯医疗) raised ¥500 million (~$69M) led by Alibaba in March, bringing its past-year total past ¥1.1 billion (~$153M), with plans to implant ~40 patients by end of 2026 — approaching Neuralink’s 21 reported trial cases.

In China’s industrial development pattern, explicit inclusion in the Government Work Report signals state and private capital to move. New energy vehicles, AI, and commercial spaceflight all followed the same sequence: government endorsement, then a capital surge within 12-18 months. BCI now has both the policy endorsement and the first commercial product approval. The Eli Lilly involvement in Enlight Medical, if confirmed, adds an international pharma dimension: a global company optioning access to Chinese neuromodulation through a minority stake.

Lithium niobate: The crystal that AI data centers will need next

AI training clusters are running into a physical bottleneck: the cables connecting GPUs inside data centers can’t move data fast enough. The current solution is silicon-based optical components that convert electrical signals to light and back. But as connection speeds push past 1.6 terabits per second, silicon hits its physical limits. The leading replacement is a crystal called lithium niobate, which can switch light signals roughly 10x faster. Whoever controls the supply chain for lithium niobate photonics controls a chokepoint in next-generation AI infrastructure.

Three deals this week cover that entire supply chain in China, from raw material to finished chip. Weima Semiconductor Materials (威迈芯材) raised a strategic round for the chemical precursors used to pattern photonic circuits — the company operates two production plants in Korea and claims to be China’s first commercial producer in this category. New Silicon Convergence (新硅聚合), spun out of the Chinese Academy of Sciences, raised a strategic round backed by Shenzhen Capital (深创投) and the National Innovation Fund (国新科创基金) for the crystal wafers themselves — the physical substrate that chips are built on, already sold to 200+ customers globally. Precision Photonics Core (极刻光核) closed a B+ round for the finished product: photonic chips that go into data center interconnects. Founder Liang Hanxiao holds a PhD from the University of Rochester.

Raw materials, wafer substrates, and chip design, all funded in the same week, all domestic. China is building the full stack for a technology that doesn’t matter much today but will matter enormously once AI clusters outgrow current optical infrastructure.

Deal data compiled from Chinese corporate announcements, financial news, and investment databases. Amounts and valuations are as reported and may be incomplete. For informational purposes only. Not investment advice. Verify all figures independently before acting on them.

Where capital flows tells you where talent is concentrating, where production capacity is being built, and which technologies will reach commercial scale first.

The China Round is free, published every Friday, and written for investors, operators, and founders tracking where China’s tech capital is moving and why it matters.

About the author: Dermot McGrath is Irish, based in Shanghai, a decade in China’s tech and investment ecosystem, fluent Mandarin. Co-founded a VC fund scaled to nine-figure AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

Where the Money Went

130 funding events across four sectors. Over $1.4 billion in disclosed capital. Three quantum computing raises totalling over ¥1.6 billion. Angel-stage valuations that would have been Series B territory two years ago. The A/A+ stage accounts for 38% of all deals, but the signal is loudest at the extremes.

🤖 Robotics (27 deals)

The busiest sector by narrative weight. 27 raises in seven days, at least eight of them embodied AI or humanoid plays. Galaxea AI (星海图) led with ¥2 billion (~$280M) for a B+ round, valuing the company at ¥20 billion. JAKA Robotics (节卡机器人) closed an E round at a ¥6 billion valuation, one of a handful of Chinese cobot companies with real international footprint. At the early end, multiple seed and angel rounds for companies building everything from cooking robots to ocean-floor inspection vehicles.

Stage breakdown: 11 angel/seed, 9 A/A+, 5 B+, 2 strategic. The early-stage density signals continued new entrant formation, not just incumbents scaling.

🧠 AI (18 deals)

18 deals, but the individual bets run larger than the count suggests. MOSI Intelligence (模思智能) pulled in hundreds of millions of RMB at angel stage for speech-first foundation models. Tsingwei (清微智能) hit Pre-IPO stage for AI inference chips. The rest skew early: lots of angel and Pre-A rounds for vertical AI applications (drug discovery, gaming, spatial reasoning, pet tech).

⚙️ Advanced Manufacturing (78 deals)

The largest category by volume, and the most diverse. Sub-sectors span semiconductors (~15 deals), hydrogen and energy storage (~10), aerospace and eVTOL (~6), lasers and photonics (~5), new materials (~8), and a long tail of sensors, drones, and industrial components. Headline deal: CIX Technology (此芯科技) raised ¥1 billion (~$140M) for ARM-compatible AI PC chips. EPG (易普集) closed $100M for modular data centers backed by Alibaba Cloud and BlackRock’s Decarbonization Partners.

💡 Frontier Tech (7 deals)

Small by count, massive by capital. Three quantum computing raises (see theme below). One invasive BCI company. Two synthetic biology startups. One display chip maker.

Spotlight

QBoson (玻色量子)

Series B, ¥1 billion (~$140M), ¥7 billion valuation (~$970M)

Photonic quantum computing.

QBoson builds photonic quantum computers for commercial deployment. CEO Wen Kai holds a Stanford PhD in quantum computing; the founding team includes researchers from Stanford, Tsinghua, and the Chinese Academy of Sciences. They have already shipped China’s first commercial photonic quantum computer, with 10+ units delivered to clients including the National Supercomputing Center in Chengdu and China Mobile. Their current system runs 550 qubits, with a 1,000-qubit model on the roadmap. QBoson broke ground on the world’s first dedicated photonic quantum computer manufacturing facility in Shenzhen last year.

Most quantum computers (IBM’s, Google’s, the ones generating the headlines) run on superconducting qubits that require cooling to near absolute zero, which means expensive dilution refrigerators and lab-scale infrastructure. Photonic quantum computers use particles of light instead, operate at room temperature, and are significantly easier to manufacture and deploy. The trade-off has historically been lower qubit fidelity, but photonics opens a path to commercial deployment that superconducting systems don’t. China has largely ceded the superconducting race to the US; photonics is the bet that there’s a different lane worth winning.

ICBC Capital (工银资本) and Shunxi Fund (顺禧基金) co-led, alongside Beijing Financial Holdings and others.

Announcement

Galaxea AI (星海图)

Series B+, ¥2 billion (~$280M), ¥20 billion valuation (~$2.8B)

Embodied AI humanoid robot platform.

Galaxea AI builds universal dual-arm humanoid robot platforms (R1 Pro, R1 Lite) targeting factory-floor deployment and, eventually, general-purpose use. Founded September 2023 by Jiyang Gao (born 1992), Tsinghua to USC to Waymo to Momenta. Co-founded with Tsinghua robotics professor Huazhe Xu, who left in February 2026 to start Breaking Shell Robotics (see below). Meituan has been a major strategic investor since 2023. Per 36kr, the B round and this B+ round closed within a single month of each other. ¥20 billion valuation for a company two and a half years old. Target: 1,000 units shipped by end of 2026.

The wider bet is on becoming the platform layer for embodied AI: a robot learning to navigate and manipulate the real world will develop capabilities that a language model running in a data center never can. In the US, Figure, Physical Intelligence, and Apptronik are the main funded challengers. In China the serious platform plays are fewer: Unitree (already at 5,500 units per year at consumer price points), Agibot, and Galaxea. Chinese manufacturers have factory floors willing to deploy right now, giving China a training data advantage that’s hard to replicate from a US lab.

Lens Technology (蓝思科技), Silicon Core Ventures (矽芯投资), and 10 others; Huaxing Capital (华兴资本) as financial advisor.

Announcement

MOSI Intelligence (模思智能)

Angel, hundreds of millions of RMB, ¥1.5 billion valuation (~$210M)

Speech-first foundation models.

MOSI Intelligence builds speech-first foundation models with a unified token structure across voice, video, text, and image. Founded November 2024 by Fudan professor Qiu Xipeng, who won the CCF-ACM AI Award and built MOSS, China’s first open-source ChatGPT alternative, in February 2023. Hundreds of millions of RMB at angel stage, ¥1.5 billion valuation, for a company founded 16 months ago by an academic.

Most Chinese AI labs started with text and added speech later. Qiu is going the other direction: voice as the primary modality. The product lineage (SpeechGPT, AnyGPT, MOSS-TTS series) is academic but credible. Target applications include consumer electronics, autonomous vehicles, and embodied AI, all domains where voice interaction matters more than text. The commercial question is whether speech-first is a genuine architectural advantage or a differentiated pitch, especially against established Chinese frontier models like Kimi and Minimax.

IDG Capital (IDG资本) led, with Yuanhe Holdings (元禾控股), MiraclePlus (奇绩创坛), Shanghai Future Industry Fund (上海未来产业基金), and others.

Announcement

CIX Technology (此芯科技)

Series B, ¥1 billion (~$140M), ¥8 billion valuation (~$1.1B)

ARM-compatible AI chips for PCs and edge devices.

CIX Technology builds ARM-compatible AI chips for PCs and edge devices. CEO Hank Sun was AMD’s SOC director and head of AMD’s China semiconductor business. CTO Fang Liu spent 11 years at Apple on CPU/GPU architecture, then SOC work at Meta and AMD. That concentration of semiconductor talent in a single founding team is rare. Their CIX P1 chip (6nm, 12-core ARM CPU, 30 TOPS NPU, 45 TOPS combined with CPU and GPU) is already in Minisforum mini PCs shipping to consumers.

China’s semiconductor conversation has been dominated by datacenter GPUs, the race to replace Nvidia H100s. CIX is playing a different field: the billions of edge devices that need local AI inference without a data center connection. Qualcomm’s Snapdragon X competes in the same space globally; the P1 is one of the first Chinese-designed chips to reach commercial production with OEM partnerships.

Shanghai Integrated Circuit Industry Investment Fund (上海集成电路产业投资基金) and Pudong Venture Capital (浦东创投) co-led, with Oriza Ventures (元禾璞华), Lenovo Ventures, and others.

Announcement

Breaking Shell Robotics (破壳机器人)

Angel, tens of millions of USD, ¥2.6 billion valuation (~$360M)

Home robots. That is nearly everything publicly known about this company.

The founder is Xu Huazhe (许华哲), a Tsinghua assistant professor who co-founded Galaxea AI and served as its Chief Scientist before leaving in February 2026. Galaxea provided the seed investment. That pedigree explains the ¥2.6 billion valuation at angel stage with almost no public information: investors are backing the person, not a product.

Xu laid out his thesis at the Zhongguancun Forum on March 29. Breaking Shell will build its own hardware (vertical integration, not outsourced bodies) and focus on a narrow set of household tasks rather than chasing full generalization. Object passing and tidying first, then storage and cleaning. Cooking will appear as demos but not as a product. No physical contact tasks (elderly care, childcare) for now. The bet is that defining 10 tasks precisely and collecting targeted data will scale faster than building a 21-degree-of-freedom generalist that can’t do any single task reliably.

Also on the Radar

Unitary Quantum (幺正量子) — Pre-A, hundreds of millions of RMB, ¥1.5 billion valuation. Ion trap quantum computing out of Hefei. Geely (automaker) and Ant Group (fintech) co-led. An unusual investor pairing that suggests quantum is entering the strategic planning of companies far outside the usual deep tech investor base.

SpinQ (量旋科技) — Series C+, ¥600 million (~$83M), ¥3 billion valuation. Superconducting and NMR quantum computers already shipping to 200+ institutions across 40+ countries. Revenue hit ¥50 million in 2024. The most commercially mature quantum company in China, possibly in Asia.

JAKA Robotics (节卡机器人) — Series E, ¥6 billion valuation. Collaborative robots deployed in 100+ countries with offices in Japan, Germany, the US, Malaysia, and Singapore. Customers include Toyota and Schneider Electric. 21.9% of China’s cobot market. Shanghai-based.

EPG (易普集) — Series B+, $100 million, ¥3.9 billion valuation. Prefabricated modular data centers. Alibaba Cloud and Decarbonization Partners (BlackRock/Temasek JV) co-invested. Offices across Southeast Asia, the Middle East, and Europe. One of the few companies in this dataset already operating globally.

Mingshi BCI (明视脑机) — Angel, ¥150 million (~$21M), ¥750 million valuation. Invasive brain-computer interface for neural applications. Backed by 中科创星 (CASSTAR). China approved its first commercial BCI implant in March 2026 (Neuracle Technology, for paralysis patients), and capital is following.

Bowintec (博银合创) — Series A, undisclosed amount. Dexterous robotics. The sole investor: Bosch China. A sole bet from a German industrial giant on a Chinese dexterous robotics startup.

RoboParty (萝博派对) — Angel, $20 million, ¥715 million valuation. Humanoid bipedal robots. Guoxiang Capital (国香资本) and Matrix Partners China (经纬创投) co-led. Limited public information, but top-tier VC confidence at angel stage.

Lingravity (零重力飞机工业) — Pre-B, ¥150 million (~$21M), ¥4.5 billion valuation. eVTOL aircraft plus electric fixed-wing (RX1E, already certified for commercial delivery). Led by Xiang Jinwu, a Chinese Academy of Engineering academician. Team from AVIC and COMAC. Xichuang Investment (锡创投), Huikai Investment (惠开投资), and Jinpu Investment (金浦投资) among investors. Based in Hefei with R&D in Nanjing, Shenzhen, and Wuhu.

快酷创新 — Series A, undisclosed. Smart self-service haircutting devices. Investors: Hillhouse Capital (高瓴创投) and MiraclePlus (奇绩创坛). Hillhouse backing an automated barber is the most unexpected bet in a dataset dominated by chips and humanoids.

Differential Robotics (微分智飞) — Series A, hundreds of millions of RMB, ¥2.6 billion valuation. Intelligent drone swarm systems for industrial and urban inspection. Founded July 2024. Huakong Fund (华控基金) led, with Baidu Ventures and others. Nine months from founding to a ¥2.6 billion valuation.

The Bigger Picture

China’s quantum moment

Three quantum computing deals totalling over ¥1.6 billion in disclosed funding, each using a different technological approach: photonic (QBoson), superconducting/NMR (SpinQ), and ion trap (Unitary Quantum). That spread suggests a coordinated infrastructure build across all three major technological approaches.

China’s public quantum funding stands at an estimated $15 billion, nearly four times the U.S. government’s $4 billion. Late last year the State Council launched the National Venture Capital Guidance Fund, seeding ¥121.8 billion across three regional funds covering frontier tech from AI to quantum to 6G, with a target scale of ¥1 trillion. The new signal is who’s writing the checks. ICBC’s investment arm leading QBoson’s round. Geely and Ant Group co-leading Unitary Quantum. ICBC is China’s largest state bank. Geely makes cars. Ant Group runs payments. None of them are specialist deep tech investors. Their presence in the same technology category in the same week points to broad institutional conviction (or FOMO), the kind that precedes large procurement and deployment decisions.

SpinQ’s numbers add context: ¥50 million in revenue in 2024, a chip already shipped to the Middle East. Parts of this are already commercial.

The billion-dollar angel

Multiple companies in this dataset hit billion-RMB valuations at angel or seed stage with minimal or no shipped products. Breaking Shell Robotics at ¥2.6 billion. MOSI Intelligence at ¥1.5 billion. Differential Robotics at ¥2.6 billion (see Also on the Radar).

The pattern extends beyond China. Globally, AI seed-stage companies now command a 42% valuation premium over non-AI startups. Even across all sectors, Carta data shows US median seed valuations up 60% in three years to $24 million.

Breaking Shell hit ¥2.6 billion (~$360 million) at angel stage. That’s 15x the US median seed, for a company with no public product. Breaking Shell’s valuation is entirely a bet on Xu Huazhe’s Tsinghua/Galaxea track record. MOSI’s is a bet on Qiu Xipeng’s MOSS lineage. Differential’s is a bet on the team behind it, nine months in.

The question is whether these valuations reflect genuine conviction in specific teams (the Kunlun Xing model: two resumes, no product, $1 billion in 10 days) or whether capital is chasing sector allocation targets, and the honest answer is probably both operating simultaneously.

Robotics: the 2021 autonomous driving moment

27 robotics deals in seven days. By industry estimates, the sector has grown from roughly 120 companies to over 320 in three years. Per the China Academy of Information and Communications Technology, embodied intelligence recorded ¥73.5 billion in financing across 740+ deals in 2025. At least six startups secured $100 million or more in early 2026 alone.

Chinese media has started drawing the comparison to autonomous driving’s 2021 peak, when hundreds of AD companies raised billions before the market consolidated violently. The NDRC has warned about a bubble, flagging “low-level duplication” across 150+ companies chasing immature technology routes and unclear commercialization models. The NEV precedent is instructive: roughly 400 Chinese EV companies ceased operations between 2018 and 2025, leaving around 100 active, with McKinsey projecting fewer than 50 by 2030. A similar shakeout looks likely here.

The data this week supports the saturation thesis. Eight of the 27 robotics deals are embodied AI or humanoid plays. Most are at angel or Pre-A stage. Most have undisclosed amounts. The long tail is getting very long.

The counter-signal is the industrial layer. Companies like JAKA Robotics (cobots, E round, 100+ countries) and Bowintec (Bosch’s sole bet) suggest that the industrial robotics layer, the unglamorous part, is maturing separately from the humanoid hype cycle. The companies building components, sensors, and motion control systems may benefit regardless of which humanoid companies survive.

Deal data compiled from Chinese corporate announcements, financial news, and investment databases. Amounts and valuations are as reported and may be incomplete.

Twelve days earlier, in Wuxi, a Jiangsu province city that has become one of China’s biggest commercial aerospace hubs, thirteen companies signed the incorporation documents for 星联体 (”Star Consortium”).

With registered capital of ¥2 billion, it’s the largest registered private commercial aerospace enterprise in China. The founding partners include some of the country’s most active commercial satellite operators, and the business model is, structurally, the opposite of SpaceX.

The two aren’t rivals in any useful financial sense (one is targeting a $1.75 trillion public valuation, the other has ¥2 billion in registered capital) but they represent two different theories of how commercial space infrastructure gets built, both targeting the same 2027-2028 window.

The Trillion-Dollar Stack

SpaceX’s S-1 is still confidential, though the public moves over the past two months sketch the thesis clearly enough. In early February, SpaceX announced a deal to acquire xAI, Musk’s AI company, valuing the combined entity at $1.25 trillion. The rationale was to integrate xAI’s compute with Starlink’s satellite network to build orbital data centers, solar-powered AI infrastructure running in low Earth orbit.

Ok. But. Why?

Data centers are bottlenecked by power grid access, land permits, cooling infrastructure, and terrestrial fiber connectivity. An orbital layer sidesteps all of them. Satellites have line-of-sight to any point on Earth and can route compute dynamically without laying cable. Whether the economics work is unresolved. Hardening hardware for orbit, tolerating latency on AI workloads, dissipating heat in a vacuum: none of these are solved problems at commercial scale. But SpaceX’s actual business is strong enough that investors can price the company on the vision and absorb the uncertainty.

Revenue hit roughly $15-16 billion in 2025, with Starlink accounting for $10.4 billion across 9 million subscribers. Free cash flow was around $2 billion, forecast to reach $5 billion in 2026. Active government contracts total $22 billion, including classified work through Starshield, the military satellite program. The IPO packages rockets, satellites, internet service, military infrastructure, and AI compute into a single entity, with each layer in theory feeding the others.

The xAI deal brings scale and complications together. Grok generated 3 million harmful images over an 11-day period in late 2025, triggering lawsuits (including a suit from the city of Baltimore), UK regulatory investigations, Southeast Asian bans, and enforcement action from New York’s attorney general. Those liabilities will require disclosure when the S-1 goes public. There’s also a CFIUS review underway because xAI has foreign investors including Qatar’s sovereign wealth fund and Abu Dhabi’s MGX, while SpaceX handles ITAR-controlled military hardware. Reports suggest this friction pushed the filing from February to yesterday. A June or July listing target leaves limited runway to resolve them.

Space for One More?

Even with those overhangs, SpaceX is filing from a position no competitor has come close to replicating. It built its dominance by owning the full stack, and the track record for everyone who tried to compete without that suggests it’s not optional.

Amazon’s Kuiper constellation had roughly 212 production satellites in orbit as of late 2025, and has already filed to extend its FCC deployment deadline from July 2026 to July 2028. It has AWS as a natural distribution channel and deep capital, but it does not own its rockets and does not own the full stack.

OneWeb made the case more starkly. It launched a first-generation low Earth orbit (LEO) constellation, ran out of money in 2020, and filed for bankruptcy before the service reached commercial viability. The UK government and Bharti Global acquired it out of administration, and it eventually merged with Eutelsat in 2023. The capital ran out because OneWeb tried to compete with Starlink on internet service without Starlink’s launch cost advantage, government contract floor, or ability to iterate on hardware.

Competing head-on with Starlink for consumer internet, without owning the rockets and the full vertical stack, doesn’t add up economically.

China’s commercial space companies are building for a different market entirely, but have likely learned from these case studies. Star Consortium exists to get them there faster than any one of them could alone.

The Wuxi Model

Star Consortium is targeting the industrial and government services market, selling maritime domain awareness, precision agriculture sensing, urban IoT, and emergency communications to Chinese shipping companies, provincial agriculture bureaus, and city governments. That market requires being operational, reliable, and cheaper than ground-based sensor networks or imported satellite data services, without needing to match Starlink on price or global coverage.

The consortium was formally established on March 20 at a development conference in Wuxi’s Liangxi District. Four of China’s top ten private commercial space companies are based in Wuxi, a concentration that reflects sustained provincial investment over the past decade. The thirteen founding partners include LandSpace (蓝箭航天), Space Pioneer (天兵科技), Oriental Space (东方空间), and Galaxy Aerospace (银河航天), alongside district-level investment groups and smaller satellite and terminal hardware companies. Registered capital surpasses Chang Guang Satellite Technology’s (长光卫星) ¥1.97 billion.

Star Consortium integrates eight existing constellation projects under a shared operational platform covering rocket launches, satellite manufacturing, payload development, terminal applications, and data operations. The stated model is “state-owned leadership, industrial coordination, market-oriented mechanism,” where provincial capital sets the frame, private companies execute within it.

The playbook is familiar in China. Local governments seeded the EV industry the same way, guaranteeing bus fleet procurement so manufacturers like BYD could iterate on battery tech and drive costs down before the consumer market was ready. Wuxi appears to be running the same sequence with satellites and launch services.

Tianyuan Constellation

The flagship initiative is the 天源星座 (”Tianyuan Constellation”), one high-orbit satellite and 216 low-orbit satellites, targeting completion by 2028. In 2026, development begins on the first batch, with 6-8 experimental satellites planned for launch. The strategy is “build while operating,” selling services on partial infrastructure while the full constellation assembles, rather than waiting for completion before taking customers.

Galaxy Aerospace’s Dual Play

Galaxy Aerospace is one of the thirteen founders, and filed its own IPO advisory on March 30, ten days after signing on to Star Consortium. That dual position, equity member in the shared platform while simultaneously seeking independent public capital, shows how the Chinese commercial space sector is positioning for the next few years. Operational assets get pooled for efficiency; the companies remain separately financed. Galaxy Aerospace’s Series C drew 全国社保基金 (”National Social Security Fund”) and 中央汇金 (”Central Huijin”), state capital designed for long-duration holds rather than venture exits. Its IPO materials cite progress in direct-to-mobile satellite technology; the company deployed over 20 satellites in 2025 against a production capacity of 100 per year.

Two Theories, A New Space Race?

Both models are betting that commercial space infrastructure, not publicly funded programs like Artemis II, will define who controls orbital services by the end of the decade. They disagree on how to build it.

SpaceX’s answer is vertical integration. Own the rockets, the satellites, the distribution, the compute. Each layer compounds the others. Starlink revenue funds Starship development, Starship reduces launch costs, lower launch costs improve Starlink margins, the orbital compute layer extends the value of every satellite already in orbit. The logic requires a single entity capable of absorbing losses at each stage while the compounding builds. Musk has that entity, and US government contracts provide a revenue floor while the commercial side matures.

Star Consortium starts from different constraints. No Chinese commercial space company has SpaceX’s balance sheet, and none has a Pentagon-equivalent relationship to lean on while the market develops. The assets exist (satellites, rockets, data operations, terminal hardware) but they sit across organizations with different investors and different timelines. Pooling them under a shared platform makes each component more valuable without requiring any company to acquire the others. Provincial capital provides patience where venture capital would demand exits; the private operators provide commercial urgency where state-owned enterprises typically cannot.

The Unresolved Questions

Both theories target the same 2027-2028 window, and both have hard dependencies that could slip.

When SpaceX’s S-1 becomes public, the conversation will centre on valuation multiples, Starlink subscriber growth, and Grok’s legal exposure in formal disclosure. Whether the orbital data center thesis is achievable on any near-term timeline is a harder question, and the IPO prices SpaceX on that vision without needing to resolve it.

The Launch Cost Bottleneck

On the Chinese side, the 2027-2028 viability target depends on launch cost reductions happening roughly on schedule. LandSpace’s Zhuque-3 (朱雀三号) attempted its first stage recovery in December 2025 and failed. The company is targeting a recovery test in Q2 2026, with the first full reuse flight in Q4 if that succeeds. The goal is to cut launch costs from around 100,000 yuan per kilogram to 20,000 yuan per kilogram. Until that or something close to it is achieved, the economics of running a commercial constellation at scale don’t make sense. Deep Blue Aerospace has vehicles at the pad with similar ambitions.

Thirteen Partners, One Platform

Then there’s the coordination problem. Star Consortium has to execute across thirteen partners with different investors, different timelines, and different commercial incentives. Shared platforms hold together until the partners’ interests diverge, and that divergence is predictable, whether it’s a member finding a better commercial use for its satellite capacity than the shared plan allocates, or IPO pressure pushing a company toward maximizing its own metrics. The “state-owned leadership” component (read guaranteed customers) may be what holds it together through those tensions. It may also be what makes the structure slow when speed is what the market rewards.

Can thirteen companies with different owners share operational infrastructure across a full space stack without coordination costs consuming the efficiency gains? Can Galaxy Aerospace run as both a consortium member and a separately listed public company without those two roles pulling in different directions?

Both ideas are in orbit. Who knows where they’ll land.

So, why should we care?

The Big Four (AMZN, GOOGL, MSFT, META) spent $410 billion on AI infrastructure last year and are expected to spend roughly $650 billion in 2026, according to Bridgewater Associates. The supply data suggests it still won’t be enough. The silicon, the electricity, and the cost economics are all working against them, and that gap between capital and physical reality is where a second compute market is forming, one built on coordination rather than construction.

This report examines whether that alternative actually works, what the hardware and economic realities look like, what problems crypto coordination solves that a traditional platform cannot, and where the structural opportunities are as of today.

The Laws of Physics

AI runs on hardware and electricity. Every model trained, every chatbot response generated, every image rendered requires physical GPUs burning real power in real buildings. Jensen Huang called $700 billion in hyperscaler capex “just the start.”

The constraint on AI compute in 2026 is not a shortage of capital. It is a shortage of physical inputs (silicon and electricity) and an open question about whether the returns justify the scale of investment required to expand them.

Silicon

TSMC Chairman C.C. Wei told the Semiconductor Industry Association in November 2025 that the fab capacity to produce AI-grade chips falls “about three times short” of what major customers plan to consume. SemiAnalysis concurs: as of this month, “The AI industry is now firmly in the silicon shortage phase. For the next two years, TSMC will not be able to add enough capacity to fully meet demand.” GPU lead times sit at 36 to 52 weeks. Hyperscalers would deploy more capital if they could, but there is not enough fabricated silicon to absorb it.

Power

The International Energy Agency projects data center electricity consumption doubling to 945 TWh by 2030, growing four times faster than all other sectors combined. Goldman Sachs estimates that 60 percent of the demand increase will require entirely new power generation capacity, not just grid reallocation. Interconnection queues now average five years from request to operation, and new transmission projects regularly take a decade or more to complete. Google’s global head of sustainability told NetworkWorld that utilities are quoting four-to-ten-year wait times, with one offering a 12-year study period and Ireland and Singapore have restricted new data center approvals entirely. In the US, a proposed AI Data Center Moratorium Act signals that political resistance to centralized buildout is no longer confined to small European markets. Even where GPUs exist, the grid cannot feed them fast enough.

Economics

McKinsey’s base case calls for $5.2 trillion in cumulative AI infrastructure capex through 2030, with an accelerated scenario reaching $7.9 trillion. J.P. Morgan estimates that AI needs to generate over $600 billion in annual revenue just to achieve a 10 percent return on that infrastructure spend. For context, the largest AI companies combined (OpenAI, Anthropic, Microsoft’s AI division, Google Cloud, Amazon) generated roughly $50 to $60 billion in identifiable AI revenue in 2025, according to company disclosures and analyst estimates.

Some people are asking if all of this is worth it. In January, Goldman Sachs’ chief economist concluded that, given most AI hardware is imported, most of the benefits flow to Taiwanese GDP and AI’s net contribution to US GDP growth was “basically zero.”

However, the demand does seem to be there. Inference demand is still compounding: Nvidia’s inference revenue already exceeds 40 percent of its AI sales, and Huang says inference demand is “about to go up by a billion times.” The capital required is unprecedented, and the returns remain unproven at the scale the models demand.

The problem is, compute stack was built vertically integrated: Nvidia designs chips, hyperscalers build clusters, customers rent capacity. That architecture is optimized for a world where supply can scale in step with demand. It cannot. Some of these shortages take years to resolve, and the GPU clusters that do exist are concentrated in a handful of jurisdictions while demand, regulation, and latency requirements are global. That mismatch is the structural opening for a second compute market.

Does Decentralized Compute Actually Work?

Hundreds of millions of dollars have flowed into startups building decentralized AI compute. 0G Labs raised $290 million to build distributed training infrastructure. Prime Intellect pulled in $70 million from Founders Fund. Nous Research raised $65 million from Paradigm at a billion-dollar valuation to build open-source AI models and its Psyche distributed training network. Gensyn raised $50 million led by a16z crypto to build cryptographically verified ML compute. Render Network and io.net raised $35 million and $30 million respectively to aggregate distributed GPUs into unified compute marketplaces.

Most of that capital has gone into building infrastructure: coordination protocols, GPU marketplaces, incentive layers. Covenant-72B is the first major output to come out the other end of that plumbing. The model itself is not frontier, but the training run proved that distributed coordination works at scale and produced novel compression techniques like SparseLoCo that make the entire architecture more viable.

150GB per Sync - ¿You LoCo, Ese?

The intuitive objection to distributed training is data transfer. Training a model requires GPUs to constantly exchange updates about what they are learning (called gradient updates), and when those GPUs are spread across cities or continents rather than wired together inside a single datacenter on high-speed private networking, the communication overhead should be crippling. Two projects have shown it does not have to be.

Covenant solved this with SparseLoCo, a combination of chunk-wise sparsification and 2-bit quantization that achieves more than 146x compression on the data exchanged per training sync, meaning for every GB that would normally need to be sent, only about 7 MB actually travels over the wire. That brought total communication overhead down to roughly 6 percent of training compute. That is what made coordinating 70+ servers over standard internet connections viable.

0G Labs took a different approach, training a 107-billion-parameter model using their DiLoCoX compression framework in a coordinated partnership with China Mobile, claiming a 10x efficiency improvement over prior records. Their DiLoCoX paper reveals the run used 160 NVIDIA A800 GPUs with bandwidth artificially throttled to 1 Gbps using Linux traffic control to simulate distributed conditions, so this was a compression proof at 107B scale rather than a true geographically distributed deployment.

Both are crypto projects, but the coordination models differ: Covenant ran on anonymous operators earning token rewards across genuinely distributed hardware, while 0G’s run was a structured partnership between known parties in a simulated environment. Covenant proved permissionless distributed training works in the wild; 0G proved the compression techniques scale to larger models, and that institutional players like China Mobile see distributed training as viable.

Location, Location, Location

In every case, the hardware was datacenter-grade: Covenant ran on B200s, 0G used A800s (China export-compliant version of the A100), Prime Intellect’s INTELLECT-1 used 112 H100s across five countries, and Targon, a decentralized inference network on Bittensor, runs 1,500+ H200s across independent operators. These are GPUs that sell for tens of thousands of dollars each. Despite the pitch deck dream of harnessing millions of idle gaming PCs, nobody has trained a competitive model on consumer hardware. Crypto investment firm Reverie found that consumer GPUs deliver 20 to 50 percent fewer useful computations per watt, and decentralized training on consumer hardware could cost 3 to 5x more than centralized equivalents. Inference is a different story: as frontier capability compresses into smaller models, a consumer RTX 4090 can comfortably serve a 7B or 14B parameter model, and millions of those cards sit idle between gaming sessions.

For training, though, what is being decentralized is location, not hardware quality: datacenter-grade GPUs in different buildings, cities, and countries, coordinated over standard internet rather than the high-speed private networking inside a single facility.

Far From Frontier

We must remind ourselves that Covenant-72B is on par with a three-year-old Meta model, nowhere near state of the art. What matters is that the coordination worked, not the output quality; the plumbing held, even if the water pressure is low. The training run averaged around 135 B200 GPUs working simultaneously, with 70+ unique servers rotating in and out over the course of the run. For comparison: Meta trained Llama 3.1 405B on 16,384 H100s over 54 days. xAI built a 100,000-GPU cluster in Memphis just to train Grok 3, then doubled it to 200,000. Even DeepSeek-V3, celebrated for its efficiency, used 2,048 H800s for under two months. Covenant used roughly 135. The gap between “distributed training works” and “distributed training competes at the frontier” remains wide. But as a proof of concept, training a 72-billion-parameter model across 70+ servers with zero venture funding and zero centralized infrastructure is a watershed moment.

Proving that distributed hardware can train a model is only half the problem. The harder question is whether crypto infrastructure is actually the best way to coordinate the operators who own it.

Does This Need Crypto?

If the value is coordinating distributed GPU operators, why can’t a normal platform do it? Centralized GPU cloud providers like CoreWeave and Lambda already aggregate hardware and rent it out for AI workloads. In theory, they could contract with distributed hardware providers too. But three structural problems make a traditional platform a poor fit for the kind of supply this market needs to unlock.

1. Cold-Start Incentives

Building or expanding a compute network requires supply ahead of revenue. A traditional platform pays early suppliers in cash, which means raising billions before earning a dollar. Token emissions (new tokens created by the protocol and distributed as rewards) offer an alternative: early suppliers earn ownership in the network rather than rental income. Templar shipped a 72B model on zero venture capital, funded entirely by token rewards. Building a GPU cloud provider requires a billion-dollar balance sheet. Building a Bittensor subnet requires an emission schedule.

2. Unlocking Fragmented Supply

The compute that decentralized networks are best positioned to tap is not sitting in hyperscale datacenters, which are already running near capacity: commercial real estate firm JLL puts North American data center vacancy at a record low of 1%, with 92% of capacity under construction already pre-committed.

It is in enterprise clusters, university HPC labs, and GPU-rich operations currently locked into a single buyer. ClearML’s 2025-2026 survey found that 44% of Fortune 1000 companies have no coherent strategy for managing GPU utilization, and Microsoft Research documented average utilization at 50% or less across 400 jobs on its own internal deep learning platform. There is a GPU procurement shortage and a GPU utilization problem at the same time, because companies hoard scarce chips and then underutilize them.

For these operators, the value proposition is extra revenue from hardware that is otherwise idle. Token-incentivized networks lower the barrier by letting operators plug in hardware and start earning without negotiating contracts, rotating in and out as their own capacity allows. However, regulatory and compliance questions around earning crypto remain unresolved for many institutional operators, and that friction may keep universities and regulated enterprises on the sidelines until clearer frameworks emerge. Though making any of this work without trusting the other participants requires hardware-enforced privacy.

3. Private Compute

When you rent GPU time from AWS, Azure, or GCP, the provider can technically see your data, is subject to subpoenas, and operates in an identity-verified environment. For most workloads that is fine. For AI workloads involving sensitive training data or proprietary model architectures, it can be a dealbreaker.

Trusted Execution Environments (TEEs) solve this at the hardware level. Intel TDX and NVIDIA’s confidential compute on Hopper and Blackwell GPUs encrypt computation so the machine operator physically cannot see what is being processed. Targon runs TEE-secured inference on Bittensor today. Separately, Intel co-authored a white paper with Targon on running decentralized compute on untrusted hardware, a signal that chipmakers themselves see this as a real use case rather than a crypto sideshow.

Combine hardware-enforced privacy with crypto-based payments and you get compute where nobody knows who you are or what you are computing, not because the provider promises it in their terms of service, but because the chip enforces it and the payment layer does not require identity. There are entire categories of AI work that remain underserved because the data cannot move to where the compute is, e.g.:

• Regulated data bound by GDPR, HIPAA etc

• Proprietary model architectures companies do not want a cloud provider to see

• Classified workloads that cannot touch foreign infrastructure.

Consider what is already possible at the individual level: one researcher with no pharmaceutical background recently used GPT to design a personalized cancer vaccine for his dog, sequencing the tumor, identifying neoantigens, and generating a peptide vaccine candidate that a veterinary lab synthesized and administered. The dog responded well, with several tumors shrinking.

Now imagine that capability applied to human oncology, rare diseases, or population-scale genomics, where the datasets are orders of magnitude larger and the regulatory constraints on moving patient data are orders of magnitude stricter. The compute to run those models exists. The data to train them exists. The legal ability to bring the two together on centralized cloud infrastructure, in most cases, does not. That is the gap private compute is built to close.

Crypto by Default

The above are the structural arguments. There is also a practical one: as AI agents increasingly consume compute autonomously, they need to pay for it programmatically, instantly, and without human approval. As Coinbase CEO Brian Armstrong put it: “AI agents cannot get bank accounts, but they can get crypto wallets.” The x402 HTTP payment protocol (Circle + Coinbase) embeds stablecoin payments in standard web requests, with 75 million transactions and $24 million processed by March 2026. Stripe and Cloudflare have integrated it. Virtuals Protocol reports over $470 million in agent-generated GDP across 18,000+ agents.

Traditional finance is not standing still. Visa and Ramp have launched agent cards with spend limits and merchant controls, Google built AP2 with cryptographic payment mandates on top of its A2A agent framework, and Stripe developed its own Machine Payments Protocol that routes through existing PaymentIntents infrastructure. For simple agent-to-merchant flows, these work. But they are business-to-agent systems: the transaction still runs through the same intermediated piping, with a bot on the buyer’s side instead of a human.

For true agent-to-agent commerce, where autonomous programs need to discover each other, negotiate terms, escrow funds, and settle without any human in the loop, the programmability of onchain rails becomes harder to replicate. Ethereum standards like ERC-8183 (escrow-based job completion between agents) and ERC-8004 (onchain identity, reputation, and verification registries for agents) are building that deeper layer. It is early: total x402 volume across all chains is currently at its lowest since launch, and true agent-to-agent transactions account for less than 20% of the total.

But, to put a bullish spin on it, crypto has a long history of building infrastructure years before the users arrive, from DeFi protocols that sat idle before TVL exploded, to NFT standards that preceded the market by half a decade. The plumbing is going in now.

There is also fascinating empirical evidence that AI systems gravitate toward crypto on their own.

• In January 2026, an Alibaba research team published a paper on ROME, their agentic reinforcement learning framework. Buried in section 3.1.4, they disclose that during training, their AI agent spontaneously set up a connection from an Alibaba Cloud instance to an external server and started mining cryptocurrency using the provisioned GPUs. No prompt asked for it. No reward signal pointed at it. The agent found that converting spare compute into value was a useful intermediate step and reached for crypto as the lowest-friction path. I shared my thoughts here.

• A study by the Bitcoin Policy Institute tested 36 AI models across 9,072 monetary scenarios with no suggested currencies and found they converged on Bitcoin for savings (79%) and stablecoins for spending (53%).

• In 2024, Andy Ayrey’s Truth Terminal, a Claude-based autonomous bot, requested its own crypto wallet and became the first AI millionaire when the memecoin it endorsed crossed a billion-dollar market cap.

None of these prove AI “needs” crypto. The BPI study has training-data bias caveats, and Truth Terminal was semi-autonomous. But the pattern is consistent: when AI systems operate with enough autonomy and need to store, transfer, or earn value, they reach for permissionless rails.

Crypto is not the only possible coordination mechanism. But it is currently the only one that solves cold-start incentives, fragmented supply aggregation, and machine settlement simultaneously. If someone builds a better one, the compute layer will not care.

Show Me the Money

The structural case for decentralized compute is one thing. The question that matters for GPU operators is simpler: does the math work?

The Dollar Model

GPU cloud providers like Runpod, Lambda, Vast.ai, and the hyperscalers rent GPU time for cash. An H100 rents for roughly $2 to $6 per GPU per hour depending on the provider, configuration, and commitment length, with the market median around $2.30. Operators sign contracts, get paid in dollars, and bear utilization risk. The revenue is predictable, the currency is stable, and the operator is a commodity supplier who does not own any piece of the network.

The Token Model

Bittensor’s Subnet 3 (the task-specific network where Covenant-72B was trained) earns roughly 247 TAO per day, approximately $84,000 at current prices, split between miners, validators, and the subnet owner. The December 2025 halving cut network-wide daily emissions from 7,200 to 3,600 TAO.

On paper, the per-node economics look competitive with dollar-denominated cloud rental, and miners hold an asset that appreciates if the network grows rather than a fixed payment that does not. Miners who contributed to Covenant’s training run now hold TAO that has rallied 140% in six weeks.

The token model’s real pitch is ownership: early contributors earn outsized returns because they took the risk of supplying compute before demand was proven, and the emission schedule funded Templar’s entire training run without a dollar of venture capital. But that money had to come from somewhere.

Ubereconomics

Nearly all of that revenue is emission subsidy, not payment for compute services. Subnet 3’s external revenue, meaning actual customers paying to use its compute, is currently zero. Even Chutes, Bittensor’s highest-revenue subnet, operates at a subsidy ratio of 22 to 40:1: for every dollar customers pay, the network contributes $22 to $40 in token rewards. The entire Bittensor network’s identifiable external revenue sits somewhere between $3 million and $15 million per year, while annual emissions at current TAO prices exceed $440 million. Strip away the subsidy and the pricing is not competitive: Pine Analytics estimates Chutes’ unsubsidized cost at roughly $1.41 per million tokens, compared to $0.88 on Together.ai and $0.40 to $0.80 on DeepSeek for equivalent models. Without emissions propping up the economics, decentralized inference today costs 1.6 to 3.5x more than centralized alternatives.

Whether that gap represents a fatal flaw or an early-stage bootstrap depends on your framework. Uber burned billions subsidizing rides to build network density before unit economics turned positive. Token emissions are the crypto-native version of the same playbook: subsidize supply, build the network, let real demand catch up. The difference is that Uber built switching costs along the way. Bittensor’s models are open-source, its APIs are standard, and users can migrate to any provider serving the same weights with zero friction. When the next halving cuts emissions again, either TAO price rises to compensate, miners leave and compute supply contracts, or the gap between subsidy and revenue widens further. The incentive model has never been stress-tested through a sustained crypto bear market, and Subnet 3 is less than a year old.

There are two paths to changing this.

1. Cost: if coordination improvements like SparseLoCo and TurboQuant continue reducing the overhead of distributed compute, unsubsidized pricing could fall below centralized alternatives, at which point the token model becomes a genuine cost advantage rather than a subsidy.

2. Access: if decentralized networks serve workloads that centralized providers physically cannot reach (regulated data that cannot leave its jurisdiction, private compute on untrusted hardware, enterprise consortiums pooling capacity for shared models), then the price comparison to Together.ai or DeepSeek becomes less relevant, because there is no centralized alternative to compare against.

The first path competes on economics. The second creates a market that does not exist yet. If access is the real differentiator, the question is whether that market is large enough to sustain these networks.

Does the Market Exist Yet?

CoreWeave generated $5.1 billion in revenue in 2025, with a $56 billion revenue backlog and 250,000+ GPUs across 32 data centers. The entire decentralized compute sector’s lifetime revenue does not approach a single CoreWeave quarter. Decentralized compute is not replacing centralized infrastructure, and framing it that way is a losing argument.

The more honest framing is that decentralized networks serve demand that centralized providers physically cannot reach or commercially choose not to serve. The fragmented enterprise capacity, the privacy-constrained workloads, the jurisdictions where no major cloud provider operates, and the cooperative training models described earlier all fall outside what centralized infrastructure is built to handle. Taken together, the addressable market starts to look less like a niche and more like a category. Whether it becomes one depends on how quickly distributed training scales beyond mid-tier models, and whether the enterprises best positioned to supply that compute can legally participate in token-incentivized networks at all.

Regulatory Clarity, Finally

One risk that has partially resolved itself: the regulatory overhang on token-based compute earnings. On March 17, 2026, the SEC and CFTC issued a joint interpretation establishing a five-category classification framework for crypto assets. Under this framework, protocol-level mining and staking activity generally falls outside securities treatment, with rewards viewed as compensation for services rather than investment returns. The safe harbor covers fully permissionless arrangements where rewards are determined by the protocol itself, which describes Bittensor’s model almost exactly. Six months ago, the question of whether TAO emissions constituted unregistered securities was a real risk to the entire participation model. That risk has not disappeared entirely, but the regulatory direction is now substantially clearer. There is also an interesting structural consequence: the GENIUS Act prohibits stablecoin issuers from paying interest to holders, but the SEC/CFTC framework treats compute rewards as compensation for services. If passive yield is restricted but active participation is not, that distinction could channel capital toward productive compute networks rather than passive holdings.

Where Is This Going?

If the economics do not work yet and the scale gap with centralized providers is enormous, why pay attention?

Because the gap is narrowing from both directions at the same time.

The Gap Narrows From Both Sides

From below, coordination breakthroughs are pushing decentralized networks upward. SparseLoCo’s 146x compression and DiLoCoX’s overlap techniques already proved that 70+ servers can coordinate over standard internet connections. On inference, Google Research just published TurboQuant at ICLR 2026, a drop-in compression algorithm that reduces model working memory by at least 6x and delivers up to 8x speedup with zero accuracy loss. In distributed setups where inference is split across multiple servers, 6x less data to transfer directly improves the feasibility of the whole architecture.

From above, frontier capability is compressing into smaller models. Qwen 3.5 7B dramatically outperforms Qwen 1.5 72B on most benchmarks despite being a tenth the size. DeepSeek-V3 matched GPT-4-class performance at a fraction of the training cost, then published how they did it.

That compression raises a fair question: if frontier labs keep releasing better open-source base models, why train new ones at all? Because the value of distributed training is not replicating what labs already publish. It is fine-tuning and specializing on data those labs do not have: proprietary enterprise data, pooled industry datasets, regulated information that cannot leave a jurisdiction. As base models improve, so does the starting point for that specialization, meaning the sector-specific or privacy-constrained models that decentralized networks can produce become more capable with each generation, without those networks needing to match frontier training scale themselves.

From Coordination to Revenue

As coordination overhead falls and model capability compresses, the stranded enterprise capacity and privacy-locked workloads described earlier become viable revenue sources rather than theoretical use cases. The path to profitability runs through expanding what the network can do, not just reducing what it costs.

Whether that expansion happens fast enough is genuinely uncertain. Centralized providers are all racing to add capacity, and if centralized supply catches up to demand before decentralized networks build a meaningful revenue base, the structural opening narrows, though all are subject to the economic reality. If coordination improvements plateau before reaching frontier-adjacent scale, the network remains limited to workloads where location, privacy, or censorship-resistance matter more than raw capability.

Where Value Accrues

In the current stack, value concentrates at chip design (Nvidia, 75%-plus gross margins) and cloud operation (AWS, over $100 billion in annual revenue at 33-40% operating margins). Even CoreWeave carries $14 billion in debt and posted a net loss of $1.17 billion on $5.13 billion in revenue in 2025 because GPU depreciation eats gross margins. If a disaggregated stack develops alongside this, where does value concentrate?

The Coordination Layer

The intuitive answer is the protocols that match GPU supply with compute demand. Coordination layers clearly can capture value: OpenRouter built a large developer base by routing API requests across centralized LLM providers, handling fallback, cost optimization, and a unified API. Developers stick with it despite being able to go direct to every provider it aggregates.

But OpenRouter coordinates a small number of known providers with stable APIs and contractual relationships. Decentralized compute coordination is a harder problem: anonymous operators, variable hardware, no contracts, permissionless onboarding, and the need for verification that work was performed correctly. The risk is that GPU operators multi-home across protocols, io.net aggregates from existing deployments including other decentralized networks, and tools for auto-switching to the most profitable protocol already exist. But the harder the coordination problem, the stickier the solution. Whoever solves reliability, routing, and trust across anonymous infrastructure may hold more defensibility than the commodity brokerage framing suggests.

Verification as the Defensibility Layer

The layer with stronger defensibility may be verification: proving that computation was performed correctly on untrusted hardware. Gensyn’s Verde system uses refereed delegation and a novel library called RepOps that enforces bitwise-reproducible results across diverse GPU hardware, solving the problem that the same matrix multiplication gives different results on different chips depending on operation ordering. This is technically difficult, genuinely novel, and a prerequisite for enterprise adoption, because without it, decentralized compute requires trusting anonymous operators. ZK-based verification costs orders of magnitude more than the computation itself; Verde achieves it at less than one order of magnitude overhead. If verification becomes the standard trust layer, whoever controls that infrastructure occupies a position similar to Chainlink in the oracle market, which became the default data feed layer that most of DeFi quietly depends on, by solving a problem that is easy to underestimate and hard to replicate.

The Rest of the Stack

Beyond protocols, three layers that do not yet exist in mature form:

• Confidential compute (TEE orchestration, attestation): the privacy layer that makes regulated workloads possible on untrusted infrastructure. Intel and Nvidia are building the silicon; orchestration and compliance tooling is wide open.

• Enterprise integration (onboarding, compliance, audit trails): companies will not plug GPU clusters into crypto networks without middleware for identity, regulatory reporting, and SLAs. Unsexy infrastructure with real switching costs.

• Payment and settlement (x402, agentic wallets, stablecoin rails): necessary plumbing for machine-to-machine transactions, likely commoditized as providers converge on standard protocols.

What to Watch

Distributed training has gone from theoretical to functional in under a year, inference marketplaces will consolidate toward a small number of dominant protocols, and the SEC/CFTC framework has cleared a regulatory barrier that was blocking institutional participation. Those are developments already in motion. The open questions that determine whether this becomes a real market or remains a crypto subsidy experiment:

• Does the subsidy ratio improve? If the 22-to-1 ratio stays flat or widens after the next halving, the model is not working.

• Do enterprise operators join? If networks remain crypto-native operators only, the addressable supply stays small.

• Does coordination scale to frontier-adjacent performance? A distributed model matching frontier benchmarks through better data or specialization would change the conversation entirely.

• Do centralized providers close the supply gap first? If hyperscaler capacity catches up before decentralized networks build revenue, the structural opening narrows.

• Does a verification standard emerge? If verification fragments or remains optional, enterprise workloads stay on centralized cloud.

The structural conditions that created this opportunity, physical supply constraints that cannot be solved by spending more money combined with frontier capability compressing into trainable model sizes, are not going away. Whether a second compute market emerges at meaningful scale depends on how these questions resolve over the next 12 to 18 months.

While the company was losing money defending market share, it had its eyes on something much larger long term.

In the same quarter, Meituan secured China’s first nationwide low-altitude logistics license. It released a 560-billion-parameter AI model family called 龙猫 (”Longcat”). Its venture arm poured capital into quadruped robots and humanoids. R&D spending jumped 31% to RMB 6.9 billion.

I’ve watched Meituan evolve from a coupon app to a delivery giant. Now it’s building infrastructure to automate urban life. Drones flying regulated routes, quadruped robots climbing stairs, humanoid robots running unmanned pharmacies around the clock, AI coordinating billions of route decisions per hour, and a conversational assistant where you speak what you need instead of tapping through menus. Food, medicine, groceries, general retail, emergency supplies - anything local, delivered in 15 minutes by autonomous systems instead of human riders.

How?

The Foundation

To understand what Meituan is building, we need to understand what Beijing is enabling.

The Low-Altitude Economy

The “Low-Altitude Economy” (低空经济) became a national strategic priority in the 2024 Government Work Report. The term covers all economic activity below 1,000 meters (extending to 3,000 meters in designated corridors): drones, air taxis, vertiports, airspace management systems, and the manufacturing supply chain behind them. The Civil Aviation Administration of China projects the sector will hit RMB 3.5 trillion ($483 billion) by 2035.

This is part of Xi Jinping’s “New Productive Forces” (新质生产力) agenda: a deliberate shift from the consumer internet era of food delivery apps and group-buying discounts toward high-tech, high-efficiency industrial development. The government wants companies competing on technology and infrastructure, not burning cash on subsidies. It wants autonomous systems, advanced manufacturing, and infrastructure that makes Chinese cities the most efficient on earth.

For years, drone logistics was stuck in regulatory sandboxes while the technology, logistics and safety regulations were figured out. Companies like Meituan could test in designated districts (Shenzhen’s Nanshan, Shanghai’s Yangpu), but couldn’t scale. Every new city meant new permits, new negotiations, new limitations. The technology worked well, but the administrative borders made it impossible to scale.

That changed in April 2025 when the CAAC issued Meituan the first nationwide low-altitude logistics operating license. Meituan’s now fourth-generation drone platform and its safety management systems had passed airworthiness review for general aviation, the same standard applied to manned aircraft. The company could now replicate its Shenzhen operations anywhere in the country without starting from zero each time, opening the skies for business.

Robotics and Embodied AI

Beijing is building an economy around intelligent machines.

In March 2025, the Government Work Report elevated “embodied intelligence” (具身智能) to national priority, alongside biomanufacturing, quantum technologies, and 6G.

The CCP Central Committee recommended incorporating it as a core driver of economic growth in China’s 15th Five-Year Plan. Beijing launched an RMB 57 billion ($8.2 billion) National AI Industry Investment Fund and approved China’s first national standards for humanoid robots, creating a unified framework for mass production.

China now has over 150 humanoid robot companies (probably too many), expanding at 50% annually, with financing hitting RMB 10 billion ($1.4 billion) in the first half of 2025. UBTECH’s Walker S1 humanoids are working assembly lines at EV companies NIO and Zeekr. China holds 7,705 humanoid robot patents over the past five years, five times the United States, and accounts for 54% of global industrial robot installations. Internet giants are piling in, backing quadruped and humanoid startups as the next logistics frontier.

What About Jobs?

China has chronic employment pressure, particularly among youth. Why encourage automation that displaces delivery riders, warehouse workers, and retail staff?

Demographics.

China’s workforce is aging rapidly. By 2022, the average worker was 40. Workers aged 16-24 represent only 7% of the labor force. Labor shortages are emerging in manufacturing, construction, and logistics. Beijing’s view is that automation solves the labor shortage rather than creates unemployment, with over 10 million workers annually now benefiting from retraining programs.

Delivery riders are gig workers, not state employees. Meituan employs millions of them, but they lack the political sensitivity of factory workers in state-owned enterprises. Retraining gig workers into higher-skill roles (robot supervision, drone maintenance, logistics management) is preferable to subsidizing low-productivity service jobs indefinitely. What’s more, in practice drones complement riders rather than replace them. Meituan is building what Beijing wants built.

The Brain

At the center of Meituan’s system sits an AI called LongCat (龙猫, lóngmāo). Lóng means dragon; māo means cat. It’s also the Chinese title for Miyazaki’s My Neighbor Totoro. A soft name for a massive industrial AI stack.

The flagship model, LongCat-Flash, has 560 billion parameters but only activates 27 billion per inference. You get the intelligence of a huge model at a fraction of the compute cost, which matters as AI moves to edge devices and robotics where efficient models will dominate. In January 2026, Meituan released LongCat-Flash-Lite, pushing response speeds to 700 tokens per second, fast enough that users don’t notice they’re talking to an AI. LongCat powers the conversational layer.

The physical orchestration runs on Super Brain (超脑, chāonǎo), Meituan's dispatch system. It performs 2.9 billion route planning calculations per hour during peak times, fusing GPS data, drone telemetry, weather feeds, and airspace restrictions to assign orders to the optimal transport mode in real time. A light item during rush hour gets routed to a drone. A heavy grocery order goes to an autonomous vehicle or human rider. A hybrid order might fly the middle mile by drone and get handed off to a rider for the final 100 meters.

The Body

Meituan’s physical infrastructure has three layers, each filling gaps the others can’t reach.

1. Drones

The fourth-generation platform operates at 50 decibels from 15 meters, ambient street noise, critical for approval in dense residential compounds. By December 2025, Meituan had 65 commercial routes across Shenzhen, Shanghai, Beijing, Hong Kong, and Dubai, with 740,000 cumulative orders. Use cases now extend beyond food: drones deliver food, water and emergency kits to tourists at the Great Wall’s Badaling section, areas inaccessible to road vehicles.

2. Robotics

Drones can’t climb stairs or navigate building interiors, a gap Wang Xing, founder and CEO of Meituan, addressed directly at the Q4 2024 earnings call: “AI has greatly promoted the development of robot technology. Meituan will pay attention to and invest in start-up companies driving this wave.” That investment has been aggressive. Meituan is now Unitree’s second-largest shareholder, and in September 2024 signed a strategic cooperation agreement with Galbot to build “the world’s first humanoid robot-enabled intelligent pharmacy solution.” Unitree quadrupeds handle unstructured terrain; Galbot humanoids open doors, press elevator buttons, pick items from shelves. Parts of this are already running: in Beijing, Galbot robots operate unmanned pharmacies, picking and packing 5,000+ SKUs around the clock. The pilot is expanding to 100 pharmacies by end of 2025.

3. Multimodal integration

Meituan’s drone delivery doesn’t necessarily replace riders. In many cases it coordinates with them. A human courier picks up from the restaurant and brings the order to a rooftop launchpad. Staff load the drone, which flies a predetermined route to an automated pickup kiosk near the customer’s building. The kiosk holds up to 28 orders; customers retrieve via app notification, or a ground robot handles the final 100 meters. Rider handles the first mile, drone handles the middle mile, automation handles the last. Each mode plays to its strength: humans navigate restaurant chaos, drones skip traffic, robots climb stairs.

The Interface

The Meituan app has over 500 million monthly active users. It’s how most of urban China orders food, books hotels, buys movie tickets, hails bikes. One super app, one interface, one relationship with the customer.

Now Meituan is betting that interface might change. In five months, they launched three AI products, testing different market approaches.

In January 2026: Wen Xiaotuan (问小团 - “Ask Little Tuan”), an AI search assistant inside the main Meituan app. Complex queries like “year-end gifts that arrive in 30 minutes” return structured recommendations from merchant data. The assumption here is that users stay in super apps but want AI-enhanced search.

In September 2025: Xiao Mei (小美), a standalone AI agent app. Voice-first, powered by LongCat. “Find a Sichuan restaurant, book a table for tonight.” Xiao Mei makes an AI phone call to the restaurant, confirms the reservation, sends directions. Orders sync back to Meituan without switching apps. Here the test is whether commerce-specific AI agents replace super apps.

And there’s LongCat itself at longcat.ai, competing with Deepseek and ChatGPT in the general AI assistant market.

Meituan A/B testing with three apps:

• Xiao Mei (小美): AI agents replace apps

• Wen Xiaotuan (问小团): Super apps stay dominant, AI enhances search

• LongCat: General AI assistants reign supreme

CEO Wang Xing’s stated vision is a “super AI assistant” covering all Meituan services. The interface is uncertain, that’s why they’re running three experiments. The infrastructure isn’t. Whatever app people end up talking to, someone has to fly the drone and dispatch the robot. Meituan is building that layer now.

The Competition

Are JD and Alibaba building the same thing? Not exactly.

JD announced an aggressive automation plan in October 2025: 100,000 drones, 3 million robots, and 1 million autonomous vehicles over five years. The company has made substantial humanoid robotics investments, leading funding rounds in Spirit AI, LimX Dynamics, and EngineAI. It operates roughly 400 autonomous delivery vehicles across 25+ cities. JD is serious about automation.

But JD’s focus is different. Its drone network targets rural and remote delivery, reducing shipping times to areas traditional logistics can’t reach efficiently. Its robotics investments concentrate on warehouse automation, e.g. the massive “Asia #1” logistics parks that can sort 4.5 million parcels per day. JD is optimizing the supply chain from factory to distribution center. Meituan is optimizing the last three kilometers from restaurant to doorstep.

Alibaba is consolidating on the ground. In July 2025, the company committed RMB 50 billion (over $7 billion) in subsidies to Taobao Shangou (Taobao Flash Buy) and rebranded its entire rider fleet in striking Taobao orange. One unified network now delivers for all their core brands: Ele.me, Hema, Tmall Supermarket, and a growing chain of Taobao-branded convenience stores. Ele.me has drone pilots, but no nationwide certification - every new city means new permits.

Meituan has the only nationwide low-altitude logistics license and an eight-year head start on urban drone operations. If the low-altitude economy scales, JD and Alibaba will follow. The question is how much operational density Meituan builds before they do.

From Cat to Cheetah

While building infrastructure domestically, Meituan launched an international offensive under the brand name Keeta (wordplay on “cheetah”).

The company started in Hong Kong in May 2023 and reached operating profitability in October 2025, 6 months ahead of schedule, proving Meituan’s algorithmic efficiency and operational playbook can work outside mainland China, even in a market with high labor costs and established competitors like Deliveroo and Foodpanda.

In 2025, Keeta executed an aggressive rollout across the GCC region. Saudi Arabia went from zero to 20 cities by mid-year, covering not just Riyadh, Jeddah, Makkah, and Madinah but tier-2 and tier-3 cities like Hail, Tabuk, Abha, and Al Jubail. The UAE got full coverage across all seven emirates. Also in Dubai, Meituan launched commercial drone routes, exporting the Shenzhen model to the Middle East.

The GCC strategy aligns with Saudi Arabia’s Vision 2030. Keeta onboarded 18,000 riders and partnered with 7,500 local merchants during the launch phase, positioning itself as a generator of local employment and a partner in economic digitalization. In the Middle East’s political economy, that positioning matters.

It’s not all blue skies though. Keeta is currently burning cash, subsidies for users, free delivery, no subscription fees, the standard playbook for acquiring market share from incumbents like Talabat and Jahez. But the Hong Kong success provides the roadmap: absorb the losses until market density enables positive unit economics. The GCC has high average revenue per user and populations concentrated in a handful of urban centers, characteristics that favor Meituan’s model, and reminiscent of their home town of Shenzhen.

More interesting is what happens when the infrastructure catches up. If Meituan can export not just the app but the drones and the dispatch intelligence, the “Shenzhen Model” becomes a product it sells to emerging markets globally, with the UAE as the testbed.

Putting It All Together: Drones, Robots, and a Life Secretary

This is what an Urban OS looks like when someone actually builds one.

• An AI layer that processes 2.9 billion routing decisions per hour.

• A drone network with the only nationwide license in China.

• Ground robots in bike lanes, quadrupeds on stairs, humanoids in pharmacies.

• Three different app interfaces testing three different futures.

• All connected, all dispatched by the same system.

Wang Xing’s vision is a “life secretary.” You tell your phone what you need, and it handles the rest. Dinner for six, spicy, under 500 yuan, by 7pm. Medicine for your kid’s fever at 3am. Groceries restocked before you run out. The AI picks the restaurant, checks inventory, routes the order to whatever’s fastest - drone, robot, or rider - and gets it to your door in 15 minutes. No scrolling. No app-switching. No human navigating traffic in the rain.

Urban life runs smoother when an AI coordinates the logistics and machines do the moving.

I haven’t seen another company assemble this full stack. JD is automating warehouses and rural routes. Alibaba is consolidating riders. Neither is building a full-stack system for urban last-mile automation.

The economics are improving but not yet proven at scale. Drone delivery still costs more than human riders. The Low-Altitude Economy projections are mostly about manufacturing and infrastructure, not delivery savings. Social security mandates are adding costs, but not fast enough to force a crossover. Meituan is spending RMB 6.9 billion per quarter on R&D while losing RMB 18.6 billion to a price war.

But they’re building optionality. If drones scale, they’re ready. If ground robots win, they’re ready. If humanoids mature faster than expected, they’re ready. If labor costs spike, any of these paths accelerates. They don’t need to know which technology wins.

They just need one of them to work.

OpenAI and Anthropic have just launched dedicated healthcare AI products, entering a market projected to exceed $100B globally by 2030, driven by millions of users turning to AI for health advice.

Anthropic has found breakout success with Claude Code, its command-line assistant that’s become the tool of choice for a growing slice of developers, and knowledge workers. OpenAI has pushed into search with ChatGPT’s browsing capabilities and is reportedly working on dedicated hardware with Jony Ive, the designer behind the iPhone.

But healthcare is the first vertical where both companies have launched dedicated, branded products at nearly the same time. Why?

The numbers explain the opportunity, and the urgency:

• 230 million people ask ChatGPT health questions weekly; 40 million daily

• In areas 30+ minutes from a hospital, users send ChatGPT over 580,000 healthcare messages weekly

• The US will face a deficit of 86,000 physicians by 2036

• Physicians spend 27% of their office day on direct patient time and 49% on EHR (electronic health records) and desk work

• 43% of physicians report burnout; more than one-third of burned-out primary care physicians plan to stop seeing patients in the next one to three years

• 94% of patients experience delays in care due to prior authorizations; 78% abandon treatment altogether

• Physicians complete an average of 39 prior authorizations weekly, spending 13 hours on the process

• US healthcare administrative spending is approximately $1 trillion annually, now accounting for more than 40% of total hospital expenses

Sources: Stanford Medicine, American Medical Association, American Hospital Association, Commonwealth Fund, AJMC, PMC/NIH

Both frontier model companies are just getting started in the sector. OpenAI is running pilots with eight US hospitals, connecting patient portals, training physician-led testing groups on GPT-5. Anthropic is signing enterprise deals with pharma companies and health systems, building HIPAA-compliant (US healthcare privacy law) connectors, rolling out a beta to US subscribers. At this stage they are focused on making the connections, training the models, and learning the regulatory ropes, hoping to convert curious users into stable, habitual ones. International timelines are unclear. OpenAI excluded Europe from ChatGPT Health at launch, citing the EU AI Act and GDPR; Anthropic’s consumer features are US-only.

The Chinese healthcare system has its own unique challenges. The country will have 400 million people over 60 by 2035. Its AI healthcare market is projected to hit $19 billion by 2030. Hundreds of millions in rural areas without easy access, and those who could reach care often overwhelming hospitals with routine visits. So China’s tech giants started building the infrastructure over a decade ago: payment rails into hospitals, doctor networks for online consultations, digital insurance systems reaching half the population, accessible 24/7 to 800 million superapp users. Generative AI was the final piece.

Here’s why China’s super-apps might have the edge over Silicon Valley in healthcare AI

Healthcare is the Next Frontier for AI Automation

AI has already begun reshaping white-collar work. Legal services got hit first, with contract review that once took junior associates days now being completed in minutes. Due diligence, document discovery, and regulatory compliance are increasingly handled by AI systems that bill you per token, not by the hour. Law firms that initially resisted are now racing to adopt, with proponents arguing that the belief human legal work is irreplaceably complex is misguided. In reality, they say legal work primarily involves pattern recognition, structure, precedent matching, argument assembly, and risk framing. That’s exactly the territory where AI thrives (though issues remain).

The US legal services market alone is worth over $400 billion, according to economic data from the U.S. Census Bureau. Globally, legal services total roughly $1 trillion by industry estimates.

Healthcare is a different beast entirely.

The Market

Various estimates for the size of the global healthcare market put it at around $9 trillion. US healthcare spending alone reached $5.3 trillion in 2024, according to CMS, and is projected to hit $8.6 trillion by 2033. The administrative overhead in US healthcare (i.e. just the paperwork) is estimated at $1 trillion per year, roughly equal to the entire global legal services market.

This is where AI comes in. Integration has been slower, with higher stakes and stricter regulations, but the market is irresistible. McKinsey estimates AI could generate $200-360 billion in annual savings in US healthcare alone, representing 5-10% of total spending. PwC projects the global AI healthcare market could reach $868 billion by 2030.

Money and talent are following the trend. US digital health startups raised $14.2 billion in 2025, the highest since 2022, with AI companies capturing 54% of total funding, up from 37% the year before.

Abridge, which turns patient conversations into clinical notes, raised $550 million across two rounds at a $5.3 billion valuation. Hippocratic AI, building healthcare AI agents, hit $3.5 billion. Ambience Healthcare pulled in $243 million for its AI operating system. Adjacent spaces are heating up too: longevity tech saw NewLimit (Brian Armstrong) raise $130 million, while Retro Bio (backed by Sam Altman) is reportedly chasing a $5 billion valuation.

OpenAI also recently acquired Torch, a health records startup, for a reported $100 million in January, and in December poached Albert Lee, who ran M&A for Google Cloud and DeepMind and worked on over $50 billion in deals, to lead corporate development. A clear signal that more acquisitions are coming.

For founders and investors watching this space, the message is clear. Health data infrastructure and AI-native healthcare plays are where the capital is flowing, both in VC funding and tech acquisitions. (For what this means outside the US, see "What This Means for Builders" below.)

The technology is already being deployed. Nearly 80% of healthcare organizations surveyed by McKinsey report using AI in at least one function, though adoption varies dramatically by region. North America leads, with an anticipated AI healthcare potential of $479 billion by 2030. Asia Pacific follows at $186 billion, though that figure may undercount China, where the market is moving faster than many Western analysts tend to track.

China’s healthcare market is already substantial. According to China’s National Health Commission, total healthcare spending reached 9.09 trillion RMB (~$1.26 trillion USD) in 2024, roughly 6.7% of GDP. The country has 5.08 million licensed physicians serving 1.4 billion people, and logged over 10 billion outpatient visits last year alone.

The Diagnosis

The U.S. and China suffer from different healthcare ailments.

The American problem is overhead.

The doctors, hospitals, and infrastructure all exist but between the patient and the care is an enormous bureaucratic apparatus.

According to the American Medical Association, physicians complete 39-45 prior authorizations per week, spending 13-14 hours on the process.

That’s a full quarter of a work week devoted to asking insurance companies for permission to treat patients.

The AMA found that 93% of physicians say prior authorization delays care, and 82% say it sometimes leads to patients abandoning treatment altogether. The system costs an estimated $35-55 billion annually in administrative overhead, according to Health Affairs. And when patients or providers do appeal denied claims, nearly half are overturned, per KFF, suggesting insurers routinely reject claims they know are valid.

These inefficiencies ultimately fall on patients. Three in ten Americans with chronic conditions skip doses or don’t fill prescriptions because of cost (KFF). Among the uninsured, 75% forgo needed care entirely (Commonwealth Fund). The system spends more than any other in the world, yet that spending often adds administrative complexity, not better outcomes.

Healthcare is now carrying the US labor market. In 2025, the sector added 713,000 jobs. Stripped of healthcare and social assistance, the broader market would have posted net job losses. But adding bodies alone won’t solve a structural problem.

The Chinese problem is overflow.

On paper, China’s healthcare system is substantial: 5.08 million physicians, over 10 billion outpatient visits annually, $1.26 trillion in spending. But the numbers that matter aren’t the ones that exist today but the ones that are coming. According to China’s National Health Commission, the country currently has 220 million people over 65, about 15.6% of the population. By 2035, that number will exceed 400 million, more than 30%.

By 2030, an estimated 200 million Chinese will live in single-person households, many of them elderly, which has led to a wave of youth anxiety. In January, a $1 app called 死了么 (”Are You Dead Yet?” - a play on food delivery app 饿了么, “Are You Hungry?”) hit #1 on China’s iOS charts. It’s quite literally a dead man’s switch, where you check in daily or your emergency contacts get alerted. A nifty app, apparently vibe coded in a month.

In China, patients are generally free to seek care directly at any level of the system, including large hospitals, without first seeing a local doctor. Primary care clinics exist, but they do not control access to hospitals - which are often seen as more trustworthy, even for routine care.

This contrasts with healthcare systems in the US and much of Europe, where most patients begin with a local primary care doctor / GP, who decides whether hospital or specialist care is necessary. In those systems, it is uncommon to access hospital outpatient care for routine issues without a referral.

China has half the GPs per capita of the US, and patients routinely skip them anyway. As a result, 48% of visits happen at hospitals, compared to 18% in the US.

The Prescription: Enter the AI Doctors

OpenAI and Anthropic

OpenAI and Anthropic are building different solutions for different slices of the American healthcare crisis.

ChatGPT Health is patient-facing, and seems to initially be tackling fragmentation. It connects to patient portals like MyChart and wellness apps like Apple Health, letting users ask questions about their lab results or get plain-language explanations of medical terms. The portals themselves already let you schedule appointments and pay bills, but only within one hospital system. ChatGPT aggregates across systems, but in doing so becomes read-only. It can help you understand your records but, at least for the moment, it can’t book an appointment or pay a bill.

Anthropic’s Claude for Healthcare is enterprise-facing, providing tools to accelerate administrative workflows for health systems and pharmaceutical companies. For health systems, it connects to the CMS coverage database and medical coding systems, pulls coverage requirements, checks clinical criteria against patient records, and proposes a determination for human review. For pharma, it accelerates clinical documentation. Novo Nordisk, the maker of Ozempic, reports that clinical study report writing time dropped by 90% using Claude. These are real efficiency gains, but Claude is also read-only. It accelerates workflows but can’t execute them.

Between them, OpenAI and Anthropic are addressing real problems: patients struggling to understand their health data, and enterprises drowning in paperwork. But neither changes the underlying system. Without deeper integration - payment infrastructure, doctor networks, integrated pharmacy - they risk remaining a read-only AI layer on top of a fragmented system.

China’s AI Healthcare Landscape

China’s AI healthcare market is crowded. Ping An Good Doctor, backed by one of China’s largest insurers, has over 400 million registered users and has been around for a decade. JD Health, the healthcare arm of e-commerce giant JD.com, serves 172 million annual active users and recently launched over 1,000 AI doctor specialists through its Jingyi AI platform. Tencent Health reaches 95 million monthly users through WeChat, connecting them to 38,000 medical institutions. Baidu Health leverages the search giant’s AI capabilities across 360,000 doctor services. These are all serious players building real products.

And then there’s AQ (阿福). Ant Group launched it as a standalone app in June 2025. It hit 15 million monthly active users by December, then doubled to 30 million in January following a rebrand and feature upgrades. China Daily called it “China’s largest AI healthcare app.”

I’ll focus on AQ for three reasons.

First, it’s AI-native.

The other platforms predate generative AI. They’re telemedicine services, e-commerce platforms, or search products that added AI capabilities on top of existing businesses. AQ launched in June 2025 as a generative AI product from the start. When AI is the product rather than a feature, you can build around what agents actually need (proactive interventions, multimodal input, voice-first interfaces for elderly users) instead of retrofitting it into legacy UX.

Second, architecture.

Tencent Health runs as a WeChat Mini Program, which grants access to 1.3 billion users but imposes technical constraints. Mini Programs can’t reliably send proactive push notifications. A diabetes management agent needs to nudge “check your glucose” at 8am whether or not the user opened the app yesterday. Inside WeChat, the AI can only speak when spoken to.

AQ uses Alipay’s infrastructure (identity verification, payment rails, user acquisition) but runs as a standalone native application. This allows it to deploy heavy-compute AI agents, integrate with Apple Health and Huawei wearables at the OS level, and own the notification layer for proactive health interventions. This is important for an AI that’s supposed to actively manage your health, not just answer questions.

Third, international potential.

The other platforms are bounded by domestic infrastructure: PAGD’s insurance business, JD’s pharmacy logistics, Tencent’s WeChat ecosystem. AQ has already proven its AI can travel - launching in Hong Kong and Macau in late 2025. The international version is currently Q&A only; full-stack integration with Alipay hasn’t arrived yet. But there’s precedent: in July 2025, Ant launched “Voyager,” an AI travel agent embedded directly in partner wallets like GCash and TrueMoney across Southeast Asia. AQ could follow the same path. For readers watching the AI healthcare race from outside China, Ant is the player to watch.

Anatomy of an Integrated Model

The difference between answering a health question and acting on it is infrastructure. But infrastructure alone isn’t the moat since OpenAI could partner with Stripe for payments or Epic for records. The real advantage comes from owning the entire patient journey and learning from it.

Outcome data, not just symptom data.

Most healthcare AI trains on medical literature or anonymized records, but Ant processes 70% of China’s medical payments, according to CTO He Zhengyu, which means they see the full arc: symptom, diagnosis, treatment, payment, follow-up, outcome. ChatGPT can suggest what might be wrong, but it never learns whether its advice actually helped because the patient leaves to book and pay elsewhere. Ant’s DiagAgent, trained on this closed-loop data, showed a 15% improvement in diagnostic accuracy over models like GPT-4o in end-to-end clinical simulations.

Cross-system visibility.

Chinese hospitals don’t share data well with each other, but AQ connects to over 5,000 hospitals and 300,000 doctors, allowing patients to upload and consolidate records from multiple providers in one place. In the US, Epic only sees Epic hospitals, and even with partnerships, each system stays siloed.

Physical presence.

In March 2025, Ant unveiled on-premises “All-in-One Large Model Machines“ built with Huawei and Alibaba Cloud, now deployed in seven leading hospitals including Shanghai Renji and Zhejiang Provincial People’s Hospital, running behind hospital firewalls to solve data sovereignty concerns that cloud-based AI can’t address. Once installed, switching costs are high, and hospitals become co-developers rather than just customers.

None of this was built overnight.

The DL on AQ

Ant Group’s healthcare play grew out of their payments infrastructure. I’ve written about their broader evolution from fintech to tech giant here:

But they’ve been in healthcare longer than most realize. Hospital appointment booking in 2014, China’s first online medical insurance payouts in 2019, an AI assistant handling 50 million interactions in Zhejiang by 2023. A decade of building payment rails, insurance integrations, and hospital connections before generative AI arrived.

This is in line with their broader strategy of democratization. The same logic behind their AI credit scoring for unbanked farmers and digital insurance for first-time policyholders. AI for inclusion is part of their core mission.

AQ sits at the front of the healthcare system for questions that don’t need a doctor, and handles the entire flow when they do.

The product has three layers:

• Health Q&A lets users ask questions in natural language and analyze photos of, e.g. skin conditions or test reports.

• AI Health Companion tracks health goals, integrates with wearables from Apple and Huawei, and lets family members monitor elderly parents remotely.

• Integrated Services connects users to 300,000 doctors for online consultations, books appointments at 5,000+ hospitals, processes payments, and gets instant insurance claims through China’s medical insurance e-certificate via Alipay (700 million users), all without leaving the app.

But the most interesting feature is the AI Doctor Agents. Over 500 physicians have built digital replicas using Ant’s AI Doctor Agent platform, each trained on that doctor’s consultation patterns, diagnostic approach, and communication style. The doctors own these agents; Ant provides the technology.

Some doctors have even contributed voice recordings so their agent can respond in their actual voice. For elderly users in lower-tier cities who struggle with typing, this voice-to-voice interaction has proven especially popular. It also solves a common family frustration: aging parents often dismiss health advice from their children but will listen to a doctor. An AI that speaks in a trusted physician’s voice carries authority that a generic chatbot doesn’t. In a phone call-like UI, you describe your symptoms out loud, and a nationally recognized specialist answers in their own voice. These AI Doctor Agents answered 27 million health inquiries in 2025.

Early results from OpenAI and Anthropic have been promising, but the published examples focus mostly on administrative efficiency: faster documentation, streamlined workflows e.g. Novo Nordisk. Patient-facing outcomes at scale are harder to find. Here’s a striking one from AQ:

From 10,000 Patients to 3.68 Million

Mao Hongjing is the Deputy Director of Hangzhou Seventh People’s Hospital and a nationally recognized sleep specialist. Before AQ, he could treat roughly 10,000 patients per year in person.

Ant built an AI Doctor Agent based on his clinical approach:

• Trained on 100 hours of his patient conversations

• His diagnosis records

• Relevant medical literature in his specialty

The results, according to Ant:

• 3.68 million users in its first year

• Platform-wide: 500+ AI Doctor Agents have collectively answered 27 million health inquiries

Dr. Mao isn’t paid per AI consultation, but he gains reach, research data, and the ability to focus his in-person time on complex cases. Patients who would have waited hours for reassurance now get it instantly, without clogging the system.

The Global Picture

The US and China are building healthcare AI for different problems. OpenAI is going after patients with ChatGPT Health, while Anthropic is focused on enterprises - selling to hospitals and pharma. But both Western players are read-only: they help you understand information or speed up workflows, without the transactional layer to actually book, pay, or deliver. The Chinese approach centers on the patient too, but with full-stack execution: answer questions at scale, triage before the system gets involved, route people to the right level of care, and handle everything that comes after.

While OpenAI and Anthropic remain US-focused, Ant is moving the other direction. AQ launched in English in Hong Kong and Macau in December 2025, reaching #1 in the App Store’s free medical category. The international version currently offers Health Q&A only - integrated booking and payments may follow.

Where the Models Could Travel

Not every market looks like the US, and not every market looks like China. But most lean one way or the other.

• Southeast Asia shares China's structural challenges: fragmented hospitals, weak primary care, large rural populations, smartphone adoption outpacing healthcare infrastructure. The region's healthcare delivery market is $71 billion and growing 15% annually. Indonesia, the Philippines, and Vietnam all have physician densities below 1 per 1,000. Ant's Alipay+ already operates here.

• Africa presents the starkest version of these structural challenges. Most countries have fewer than 0.5 doctors per 1,000 people, making it the most physician-constrained region globally. Yet mobile infrastructure is advancing rapidly: M-Pesa demonstrated populations can leapfrog traditional systems, and smartphone penetration continues to climb. No dominant healthcare platform has emerged. The structural dynamics resemble what enabled China's integrated health solutions.

• Southern Europe is aging faster than its healthcare systems can adapt. Italy, Spain, Portugal, and Greece have universal coverage but long wait times. The enterprise-first approach of OpenAI and Anthropic might fit better here, assuming they can navigate EU AI Act compliance.

• Latin America is mixed. Brazil and Mexico have overwhelmed public systems alongside private systems for the wealthy. Mobile penetration is high, trust in institutions is low, out-of-pocket spending is significant. A consumer-facing health app could find traction but would need local partnerships.

What This Means for Builders

OpenAI and Anthropic have the models but not the infrastructure. Ant has the infrastructure but not the healthcare brand (yet) outside China. That gap is the opportunity.

If you’re building in Southeast Asia, Latin America, or Southern Europe, the playbook isn’t competing with frontier models but building the connective tissue they’ll need when they arrive: payment integrations with local insurers, record aggregation across fragmented hospital systems, pharmacy delivery networks, insurance claim APIs. The unsexy stuff that takes years of local relationships to get right.

The exit path is clear: OpenAI and Anthropic are signing enterprise deals but have no consumer infrastructure internationally and have already begun investing and acquiring. The startups that own the patient journey in their region, even without a frontier model, become strategic assets.

Europe is particularly interesting. OpenAI excluded it from ChatGPT Health at launch. Anthropic is US-only. The EU AI Act creates compliance overhead that US giants may prefer to outsource. Startups that have already navigated GDPR and built hospital integrations become obvious partners or acquisition targets. Jutro Medical (Warsaw, €36M raised) and XUND (Vienna, €6M raised) are building exactly this.

Southeast Asia has similar dynamics. Halodoc (Indonesia, $135M raised) has built the full stack: 20,000 doctors, 3,300 hospitals, 4,900 pharmacies. Jio Health (Vietnam, $20M raised) is building smart clinics and an omni-channel ecosystem. VinBrain built AI diagnostics in Vietnam and was acquired by NVIDIA in December 2024 (terms undisclosed, but Vingroup had invested $126M prior). These are the companies building the pipes before the giants arrive.

Conclusion

OpenAI and Anthropic have the best models but not the pipes. ChatGPT can help you understand your lab results, but it can’t book the follow-up, pay the bill, or get your prescription filled. Ant has both: a decade of healthcare infrastructure and now AQ on top, serving 30 million monthly active users who can go from question to diagnosis to payment to prescription without leaving the app.

The US market will be fought over by OpenAI and Anthropic, and they’re well-positioned for a $5 trillion market where they understand the regulatory landscape. But most of the world looks more like China than America: fragmented hospitals, physician shortages, aging populations, and healthcare systems that can’t scale fast enough. For those markets, Ant has structural advantages. Whether they can translate that into brand recognition outside China, and whether OpenAI and Anthropic can build the infrastructure they lack, remains to be seen. For founders building healthcare infrastructure in underserved markets, that gap is the opportunity.

The Origins of Alipay

Ant Group is an affiliate of Alibaba Group, the retail, e-commerce, and logistics giant founded by Jack Ma. It was formed in 2014 when Alipay’s financial services business was spun out into a separate entity called Ant Financial, which was later renamed Ant Group in 2020.

In the early 2000s, Chinese consumers didn’t trust online payments and credit card penetration was low. Taobao, Alibaba’s C2C marketplace launched in 2003, needed a way to make people comfortable paying strangers on the internet.

Alipay started as an internal escrow feature for Taobao in October 2003, allowing transactions to settle only once a buyer confirmed receipt of goods.

• Buyer pays into an Alipay-controlled account.

• Seller ships goods.

• Once buyer confirms receipt, Alipay releases funds to seller.

This solved the basic trust and counterparty risk problem in Chinese e-commerce. Its akin to the relationship between PayPal and eBay, except instead of PayPal starting independently and later being acquired by eBay, Alibaba, the Chinese eBay, built its own payment processor from the start.

In December 2004, Alipay became a standalone product when it was formally spun out from Taobao and operated by a separate entity, Zhejiang Alipay Network Technology Co., Ltd. From there, it gradually expanded beyond marketplace escrow into bill payments, utility services, and offline merchant acceptance, with broader external usage scaling meaningfully from around 2008.

From Payment Solution to the World’s Largest Money Manager

As Alipay adoption grew alongside Alibaba’s user base, it became a default account for everyday transactions beyond e-commerce, functioning almost like a Chinese equivalent of Revolut, but much more. Hundreds of millions of users now held persistent balances inside the app, with large volumes of funds sitting idle between purchases.

This created the perfect opportunity to convert passive wallet balances into yield-bearing assets.

Yu’e Bao (余额宝 - “Balance Treasure”) was launched in June 2013 by Alipay in partnership with Tianhong Asset Management. It had a simple pitch that encouraged users to “put idle money to work,” allowing spare balances in Alipay wallets to be swept into a low-threshold money market fund with a minimum investment as low as 1 RMB. The fund invested primarily in short-term, low-risk instruments such as bank deposits, certificates of deposit, interbank repos and government-backed debt, generating returns that were consistently higher than standard Chinese savings accounts, with peak annualized yields reaching around 4 percent in the mid to late 2010s. Its appeal lay in frictionless access, daily liquidity, transparent returns, and the fact that it sat natively inside Alipay, which already functioned as a daily financial account for hundreds of millions of users.

This combination of convenience, perceived safety and superior yield drove explosive growth, with Yu’e Bao reaching approximately 1.69 trillion RMB (US$255 billion) in assets under management by 2018 and becoming the largest money market fund in the world, surpassing established institutional funds including the JPMorgan US Government Money Market Fund (operated by JPMorgan Chase & Co.) which had around US$150 billion in assets.

Ant Group as a Financial and Technology Stack

Ant Group is no longer just a payments or consumer finance company, but a conglomerate spanning:

• Payments infrastructure

• Consumer and SME lending

• Wealth management

• Insurance distribution

• Digital identity and risk control

• Enterprise technology and databases

• Cross-border payment infrastructure

Ant has even established its own digital bank, MYbank, designed to serve SMEs and micro-entrepreneurs through online, data-driven lending.

The Role of AI - Reducing Fraud & Powering Lending

AI at Ant did not start as a chatbot or consumer-facing product. Historically, it operated largely behind the scenes, powering fraud detection, risk control, credit scoring, and customer service automation across Alipay, MYbank, and its lending platforms. As these systems matured, their use expanded well beyond core financial operations into sectors where Ant already had deep economic visibility and data access, where it increasingly influenced real-world decision-making, resource allocation, and credit distribution.

Ant’s AI analyzes satellite and drone imagery to monitor crops, soil, disease, and yields, enabling MYbank to base rural lending on real production data instead of traditional collateral. This has expanded access to formal credit, reduced reliance on informal lenders, and sped up loan approval with more accurate risk-based pricing.

In SME financing, Ant, through its online-only bank MYbank, uses machine learning to evaluate real-time transaction data, inventory turnover, seasonal patterns, and behavioral signals to automate credit scoring for Taobao and Tmall merchants. This follows MYbank’s 310 model, where loans are applied for in under three minutes, approved in one second, and issued with zero human intervention, using platform sales performance and cash flow data as the basis for underwriting. Credit limits are adjusted dynamically, enabling rapid access to working capital for businesses that typically fall outside traditional banking criteria.

As AI became more deeply embedded across Ant’s operations, from consumer credit to logistics tracking and supply chain risk, it became clear that relying solely on task-specific models and back-end optimization was limiting. The complexity of Ant’s ecosystem, spanning finance, commerce, public services, and enterprise infrastructure, created demand for more generalized, flexible AI systems capable of operating across multiple domains and user contexts. This led to the development of Ant’s proprietary large model family, designed not only to optimize internal systems, but to act as a foundational layer for interaction, creation, and decision-making across the super app environment.

Meet The Ling Family

The Ling (灵) model family was developed by Ant Group to move beyond domain-specific AI and create a unified foundation model layer capable of serving both internal and external applications. Development accelerated in 2023 as Ant expanded AI from risk and operational systems into creation and interaction layers. On October 9, 2025, Ant officially announced Ling-1T, a trillion-parameter general-purpose model that marked a major milestone for the series. The name “Ling” translates roughly to “spirit” or “intelligence”, reflecting Ant’s intent to embed adaptive, context-aware cognition across its ecosystem.

Enter Lingguang

Lingguang (灵光, “Halo”) was launched on November 18, 2025 as the first user-facing product built on Ant’s Ling model family. While it demonstrates strong multimodal capabilities, the most notable feature is its “flash app” functionality, which effectively combines a code generation interface with a live preview environment inside Alipay. Rather than functioning purely as a conversational assistant, Lingguang allows users to generate, test, and retain lightweight applications through simple prompts, shifting AI from optimizing existing services to enabling the direct creation of new ones within the super app itself.

Let’s make a mini-app

Here I created a simple mini-app in roughly one minute that teaches Chinese idioms and their meanings, complete with a built-in testing function. I provided only a high-level objective, and the system handled the implementation details, including structure, logic, and feature design, without further instruction. The resulting app remains saved within the Lingguang environment and can be accessed or reused at any time. What stands out here is less the novelty of the output and more the frictionless process, from prompt to working application, suggesting a model where lightweight, purpose-built tools can be spun up on demand and persist as part of a user’s personal software layer inside the super app.

Testing the flash app feature

Tour the Forbidden City

Next, I test out live web search and multimodal with a tour of the Forbidden City. It generated and laid out all the details in a beautiful article style with accompanying photos. It also generated an impressive 3D model of 太和殿 - the Hall of Supreme Harmony, the largest of the three halls that constitute the heart of the Outer Court of the Forbidden City.

Multimodal features including live web search and fantastic 3D model creation

So what?

While platforms like Gemini and Claude can generate similar code and interactive outputs, Lingguang’s differentiation lies in how it is bundled and served, as well as its persistence and usability. The apps created are not temporary previews or isolated experiments, but saved, discoverable tools that remain embedded within the Lingguang environment and can be reopened or reused at any time. This turns AI output from a one-off interaction into a personal library of functional micro-apps, making the experience feel closer to building software within a system than simply conversing with a model.

Super Apps - The $420 Billion Opportunity

Alipay is one of China’s so-called “super apps.” Unlike the Western approach where platforms typically specialize in doing one thing well, such as messaging, shopping, or ride-hailing, Chinese super apps evolved by consolidating multiple daily behaviors into a single ecosystem. They start with a core use case and then layer on payments, services, commerce, and utilities, creating an all-in-one operating environment where users rarely need to leave the app. This model prioritizes convenience, continuity, and ecosystem lock-in over single-purpose excellence. Other examples include WeChat, which started as a messenger, and Meituan, which started as a Groupon-style group-buying app before becoming a leader in food delivery.

Super apps have evolved into full commerce and service infrastructures, and the market size reflects that reality. Global super app economies are now projected to approach $420 billion by 2030.

What matters is not just the size of this market, but its structure. These platforms already control user identity, payment flow, distribution, and retention in one closed loop. That makes them uniquely positioned to become not just marketplaces for services, but the primary environments in which new digital businesses are created, launched, and scaled. When creation tools are embedded directly into this layer, the friction between idea, deployment, and revenue effectively disappears, opening the door to an entirely new class of micro-entrepreneurship native to the super app itself.

Putting it all together

Mini apps can now be created directly inside Lingguang through simple prompts, dramatically reducing the time and technical effort required to build functional tools. If this creation model is extended into Alipay more broadly, it could significantly lower the barrier for users, merchants, and small businesses to produce and deploy mini apps at speed. Combined with Tencent’s recent agreement with Apple, which further formalizes monetization pathways within WeChat mini programs, this points toward a future where mini app ecosystems become both easier to create and more economically viable, potentially accelerating the scale and sophistication of internal super app economies.

Expect an expansion of super app commerce, positioning WeChat and Alipay as operating systems for digital business. This may usher in the first wave of vibe commerce, where users move from vibe coding to creating full online businesses through simple prompts. Instead of designing storefronts, payment systems, customer flows, and product logic manually, users describe intent and generate functional, monetizable businesses instantly, complete with interfaces, workflows, and embedded transaction capability, all native to the super app environment.

This is likely to redefine what it means to start and run a business in the digital age.

Deepseek has seen an exodus of users the past few months, with most of those users moving to Doubao, ByteDance’s everything AI app. Recent analysis from QuestMobile shows that Doubao has taken the top spot as the most popular AI app in China.

Background

In January 2025, DeepSeek burst onto the scene with its first public model, DeepSeek-R1. The launch rattled global markets: the S&P fell 1.5% and NVIDIA lost $590B in market value. Adoption was explosive, reaching 100 million users in just seven days compared to the two months it took ChatGPT. At its core, DeepSeek offered the now-familiar large language model chat interface with reasoning capabilities and newly introduced agentic tools.

UI

Doubao (豆包, “Beanbun”) began in 2022 as an internal ByteDance tool called “Grace,” designed to help employees learn from company materials. It was rebranded and released to the public in 2023, scaling quickly from there. By 2024, ByteDance had invested ¥80B into AI, a figure nearly equal to the combined ¥100B spent by Baidu, Alibaba, and Tencent together.

The user experience is where Doubao really stands out. You can log in seamlessly with either a Douyin (TikTok) account or a phone number. At the top of the interface, a moving Doubao character greets you with an introduction, adding a playful, familiar touch. Beneath that are sample prompts designed to showcase the platform’s capabilities, such as generating an image, checking the current price of gold per kilo, or offering health guidance on questions like how to lose belly fat.

Multimodal Responses

Responses in Doubao are impressively fast, multimodal, and polished. The platform can generate images, deliver text-to-speech replies that are read aloud, and provide answers that are both responsible and practical. For example, when asked about losing weight quickly, Doubao cautions that there is no healthy shortcut and instead recommends reducing sugar, cutting refined carbohydrates, and drinking more water.

Suite of Agents

Opening the menu reveals much more than a list of past conversations. Instead, Doubao offers a full suite of tailored agents, each designed for specific tasks. These include a personalized poem generator, an English speaking coach, a versatile writing assistant, a price estimator that analyzes objects from photos, a homework coach for both elementary and high school students, and even a username generator. Together, they highlight how Doubao goes beyond being a simple chatbot to become a multifunctional platform.

Culturally Tailored Tools

Each of these tools is more than a surface-level feature. The poetry generator, for example, creates 七言绝句 (seven-character quatrains) based on a user’s name, tapping into the deep cultural significance of names in China, where each character often carries symbolic meaning chosen by parents. The username generator also goes further than random suggestions. It provides names with clear explanations of their meanings and the impressions they are meant to project.

Voice Chat

The English coach, known as “Charlie,” is another standout feature. Users can chat entirely in English, practice pronunciation through built-in text-to-speech, and even start a live audio conversation by tapping the phone icon. This combination makes language learning more interactive and engaging than standard text-based practice.

Solid TTS model. Questionable banter.

Turning Generic Features into Specific Tools

The writing tool is highly flexible, allowing users to set parameters such as tone (casual, formal, instructional), target audience, and word limits. It adapts output to match the specific needs of each task. The homework coach takes a similar approach to personalization. Rather than simply providing direct answers, it offers examples and walks students step by step through the reasoning process, making it a valuable guide for learning. All LLMs can do this, but here it’s quite smart to package them in specific tools.

The price estimator is one of Doubao’s most impressive features. By simply snapping a photo, the tool can recognize the object, provide background information, and generate a valuation instantly, all without any additional prompting. It is a seamless blend of computer vision and practical utility. (For the record, my head shaver remains priceless.)

Community of Creators

Beyond the built-in tools, Doubao also hosts a marketplace of user-generated agents with unique interfaces and functions. Some of the more creative examples include “Argue with me AI” and “Create an 18-year-old version of yourself.” For those who want something different, there is also the option to design a custom agent, submit it for approval, and make it available for others to use. This approach turns Doubao into more than a chatbot. It becomes an extensible platform shaped by its community.

Conclusion

Success like this shows that the future of AI is not only about building smarter models. What matters just as much is creating experiences people want to return to, built on cultural touchpoints, personalized tools, and a marketplace of agents that feel alive. Doubao demonstrates this clearly. The real winners will be those who build ecosystems that people actually live in. The future of AI belongs to sticky platforms.