Its starting ingredients include sodium sulfate and ferrous sulfate, two chemicals produced as by-products of other industrial processes. Ferrous sulfate is an iron salt already used to treat drinking water and wastewater. It is also left over from making titanium dioxide, the bright white pigment used in paint. Pigment manufacturers remove iron from their ore, leaving an iron-rich by-product that can become a battery ingredient. Zoolnasm uses these ingredients to make sodium iron sulfate, known as NFS, for a battery’s positive electrode, the cathode.

Sodium batteries generally store less energy for their weight than lithium-ion batteries but perform better in the cold. Their raw materials are abundant and could lower costs, although making a finished cell cheaply is still hard. Those traits suit short-range vehicles, backup power and stationary storage, where a battery’s weight counts for less than in a long-range car.

The company sells this cathode material to other battery makers and also offers its own batteries. A cell is the individual unit that stores energy; multiple cells are assembled into a battery pack. Zoolnasm makes some cells itself and works with established cell manufacturers for larger-scale production. Its finished products include e-bike packs, backup batteries and storage systems. It lists 0.5 GWh of cell capacity against a 5 GWh target, with cell-line expansion the first stated use of the new financing.

The latest investors and valuation are undisclosed. Zoolnasm says it has raised more than ¥1B (about $149M) cumulatively. Earlier backers include energy-storage system supplier Pylontech (派能科技, 688063.SH), which agreed to invest ¥30M (about $4.5M) in 2023 at a ¥1.5B (about $224M) pre-money valuation.

Whether batteries made from industrial leftovers pay depends on what they cost to make and whether buyers will pay more for longer life and cold-weather performance. Zoolnasm is pursuing longer-lived e-bike packs and cold-weather backup power, while competing with cheap lead-acid batteries and lithium batteries whose raw-material price has fallen.

Why this chemistry?

Sodium batteries can be made with different cathode materials. Zoolnasm’s NFS uses sulfate, while the more common NFPP uses phosphate. Metals data provider SMM estimates that NFPP accounted for 81% of China’s sodium cathode output in December 2025. Cathode supplier JNION (珈钠能源), which specializes in NFPP, said in May that its annual capacity was already in the tens of thousands of tonnes. Zoolnasm lists 13,000 tonnes of NFS capacity, with a target of 30,000.

Read more

Founded in January 2024, Sand.ai makes models that generate video with audio from a text prompt, or from a prompt plus a still image. As of writing, its MAGI-2 Preview ranks 10th overall and second among open-weight models on Artificial Analysis’s image-to-video leaderboard. MiniMax H3 leads the open-weight models. Chinese developers including Alibaba, ByteDance and MiniMax feature prominently near the top of the video-model rankings.

Sand.ai founder Cao Yue attributes China’s strength to timing and its short-video market. In video, he argues, Chinese teams started at roughly the same time as their overseas rivals, so they had less catching up to do than in language models. He also credits China’s more active short-video ecosystem with getting the models into use faster.

Alongside its research, Sand.ai runs VidMuse, a paid video-creation service with tools for music videos, product ads and anime dramas. The service offers access to competing companies’ models. Wangsu’s stock-exchange announcement of the proposed investment includes Sand.ai’s financial results.

Ten Seconds, Eight GPUs

Sand.ai’s profile of Cao says he is an author of Swin Transformer, the Microsoft Research Asia paper that won the best-paper award at the 2021 International Conference on Computer Vision.

The MAGI-2 research preview produces ten-second clips with sound. Sand.ai has released its weights and generation code under Apache 2.0, so developers can download the model and run it on their own infrastructure.

The published setup requires eight NVIDIA Hopper GPUs, and the model’s weights total roughly 307 GB.

As of writing, MAGI-2 ranks 21st on Artificial Analysis’s text-to-video leaderboard. Both rankings cover video with audio and rank models by pairwise human preference votes on generated clips.

Sand.ai has also changed the approach we covered in June. MAGI-1 generated video in successive chunks, using earlier footage to guide what came next and extend the video. MAGI-2 did not directly carry that method over. Sand.ai says retaining it would complicate the combination of joint audio-video generation and distributed training at the 100-billion-parameter scale.

Joint audio-video generation is already shared by models including ByteDance’s Seedance 2.0 and MiniMax H3. H3 also runs video and audio through a single-stream architecture. H3 and MAGI-2 take different approaches to scaling: H3 uses a dense 33-billion-parameter model, while MAGI-2 uses a mixture of experts with 114 billion parameters, activating about 6 billion per token. This lets Sand.ai increase model capacity while limiting how much of it is used for each calculation. The company says the design explores an efficient path to scaling video generation.

Why Offer Other Companies’ Models?

In June, Yicai identified VidMuse as Sand.ai’s core revenue product.

VidMuse’s Pro plan costs $33 a month when billed annually and includes 4,000 monthly credits for generating and analyzing media. How far those credits go depends on the model, video length, resolution and settings; subscribers can buy more.

Read more

Z-TRAK’s proposed rocket, Fission-A (裂变号-A型), is designed to carry one or more satellites with a combined weight of up to 450 kg into low Earth orbit. The first-stage booster powers the initial ascent, then separates; the upper stage carries the satellites the rest of the way into orbit. Z-TRAK plans to land the booster and fly it again, spreading the expense of building it across multiple launches.

A satellite owner can buy space alongside other customers on a shared launch, or book a whole launch around its own mission. Sharing spreads the cost among customers, but the satellite has to fit the flight’s orbit and timetable. A dedicated launch gives the customer more control over both, while leaving it to cover the launch bill. Z-TRAK says rideshare customers wait an average of 6.8 months; it wants to make booking a whole launch a more affordable alternative.

The amount a rocket can carry depends on the orbit it needs to reach. Fission-A’s planned maximum is 450 kg to low Earth orbit, falling to 300 kg for a sun-synchronous orbit. That is a type of low Earth orbit used by imaging satellites to revisit a location under consistent sunlight, making changes on the ground easier to compare.

For a whole launch carrying up to 450 kg to low Earth orbit, Z-TRAK targets a cost below ¥15M ($2.24M). SpaceX’s published rideshare tariff charges ¥14.1M ($2.1M) for a customer to send 300 kg to sun-synchronous orbit alongside other customers’ satellites.

SpaceX already reuses Falcon 9 boosters on both dedicated and shared launches; the upper stage is used once. Z-TRAK is trying to bring booster reuse to a much smaller rocket, so a small-satellite customer could afford the whole launch rather than a share of a larger one.

Zhang says he grew up building rocket models and wrote the business plan with classmates in a Shenzhen co-working space in October 2024. He says he is still studying aerospace engineering at Hong Kong Polytechnic University. The venture has since moved to Zhuzhou, Hunan, through a local government program that uses investment to attract businesses.

On September 29, 2026, Z-TRAK announced a “¥100M-level” (~$14.9M) angel-plus round, its latest round of early-stage funding. Investors bought about 10% of its Hunan operating company at a reported valuation of ¥1B ($149M) after the investment, up from ¥300M ($44.7M) in May. Returning investor Zhuzhou Beidou Group (北斗集团), a local state-owned investment company, is helping establish Z-TRAK’s development and manufacturing base. The team has tested an engine on the ground; the rocket has yet to fly.

Batteries behind the burn

Z-TRAK uses batteries to power the pumps that feed methane and liquid oxygen into its engine’s combustion chamber. The methane burns to produce thrust; battery-powered electric motors drive the pumps, replacing the gas turbine used in many liquid-fueled rocket engines. This simplifies the machinery and makes thrust easier to adjust, which matters when slowing the booster for landing. Zhang said in a July 2026 interview that the team built its first engine prototype for under ¥500,000 ($74,600).

Read more

This was its sixth financing since incorporation in December 2025. Its fundraising adviser says investors have put more than ¥1B (~$149M) into the company in total. Its valuation rose from ¥3B (~$446M) on June 11 to ¥5B on July 31 and ¥6B in the latest round, doubling in about three months.

Mind over motor

The round was announced on September 17, the day LiberAI’s Liber-0 Preview model took first place on RoboDojo, which tests AI models controlling robot arms on 42 simulated tasks. A day earlier, RoboDojo’s team had added OpenAI’s GPT-6 Astra to the board. Tested without additional task-specific training, Astra achieved a 22.5% average success rate, putting it above every public entry at the time. Liber-0 Preview reached 25.5%.

Read more

Chen Tianrun (陈天润), born in 2000, is a Zhejiang University PhD student under Pan Yunhe (潘云鹤), a senior figure in Chinese AI research. He registered KOKONI in 2021 and the next year began selling a consumer 3D printer with an app that turned photos into printable models; kokoni.ltd still opens on the printer, its copyright notice stopped at 2024, and nobody has touched the old business since.

Then investors began pouring money into world models, AI that predicts how a scene will change, and Chen had something most model labs did not: a library of 3D objects and the tools to make more of them quickly. By his account KOKONI used that library as its training data: assemble the objects into scenes that move and you have the raw material for a world model, produced in house rather than scraped. The company pivoted to spatial intelligence in 2024. Customers so far, by the same account, are a few embodied-AI companies, digital factories, and surveyors who now turn three photographs of a site into a 3D map in minutes, work that used to take days with dedicated sensors.

Read more

The low altitude economy

China’s 低空经济 (”low-altitude economy”) is the government’s umbrella for all civilian use of low-altitude airspace, from eVTOL passenger aircraft and delivery drones to power line inspection, agricultural spraying, and emergency response.

Chinese megacities are running out of…

Where capital flows is where talent concentrates, where production capacity gets built, and where the next commercial technologies will come from.

The China Round is published weekly, 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.

This week’s 183 deals ranged from a fold-away home robot to AI shrunk into a pair of headphones, but a handful kept circling the same question: who draws the line between foreign and domestic technology, and which side of it a company lands on. China draws it one way, pulling the work home. A design-software maker hit an ¥8B (~$1.15B) valuation, close to the listed domestic leader’s market value in early September, on a CAD package (the 3D software factories draw their parts in) whose core engine it licenses from Siemens, just as the state’s buy-domestic drive makes owning that engine the whole point. A seed-stage startup that says it serves three million users across Pakistan, Nigeria and Bangladesh moved its AI off Google’s cloud onto Chinese chips in Shanghai, months before the city made that shift official policy. Japan draws it the other way, and a Shanghai developer is wiring Chinese cells into Japan’s grid through a Japanese operating company, hoping to clear a security certificate no Chinese battery maker has yet passed.

The largest cheques, though, went to teams priced on a résumé more than a product. An OpenAI returnee pointing the large-model recipe at robot bodies raised roughly ten times the typical round at his stage, and the single biggest new round skipped the model labs for the compute layer beneath them.

Where the Money Went

Semis 41 | AI 33 | Robotics 30 | Adv Mfg 30 | Energy 28 | Frontier 12 | Aero 9

183 deals | 23 disclosed a combined $963M (~¥6.7B) | median disclosed $14M (~¥100M)

Semiconductors were the week’s densest lane by count, 41 of 183 deals, but robots pulled the most disclosed capital, about $346M (~¥2.4B) across their rounds, just ahead of AI’s $313M (~¥2.2B). The single biggest new round went to AI, over ¥2B (~$288M) to AGIC micro (谦合益邦), a chip designer building compute-in-memory processors that do the arithmetic where the data already sits instead of shuttling it to a separate chip, the bottleneck that makes AI inference slow and power-hungry. The state’s China Structural Reform Fund anchored it, one of several teams this week chasing the same problem. Robotics’ total was spread wider and mostly priced on pedigree, with INFIFORCE’s ¥1B (~$144M) A+ round and Knowin’s ¥500M (~$72M) angel both backing pre-shipping robot-AI teams out of Huawei and the big model labs. Energy and materials ran quieter but pointed offshore, Inxtech’s $100M (~¥695M) round for solid-state batteries aimed at drones and eVTOLs (battery-powered air taxis), Taoke building and trading grid batteries inside Japan. Only 23 of the 183 rounds published an amount, so the $963M is a floor, most of the week’s capital is undisclosed, and the state funds behind the largest cheques are doing most of the disclosing.

Spotlights

Omnimind (万象智维)

Angel+ · Tens of millions RMB (single-digit millions USD) · Valuation undisclosed

A 30-person Tsinghua spinout building the software layer that runs AI agents on the device itself rather than in the cloud, with a bone-conduction headphone maker leading the round.

Omnimind builds the plumbing for running AI models and agents on phones, PCs and gadgets instead of on cloud servers. Doing the work on the device is cheaper, with no per-token bill to a cloud provider and a reach to hosted models only when the local one cannot cope, and more private, since anything handled on the phone never leaves it, which matters most when the agent is operating your other apps for you. Incubated in 2025 out of Tsinghua’s ubiquitous-intelligence lab, it is…

Why build a company around one joint? Because the hand is where a humanoid does its job. Legs get the robot to the work; the fingers do the work, and packing real dexterity into a human-sized hand is one of the hardest parts of the whole machine. Every joint needs a reducer (the small gearbox that turns a motor’s fast spin into a joint’s slow, controlled movement), and every reducer costs space, money and power.

One that is smaller, cheaper and self-locking is much more than a spec-sheet nicety. Without it, a joint either burns motor power to hold a grip, which drains the battery over a shift, or needs a separate brake bolted on, another part, more weight, one more thing to fail. A self-locking joint holds that grip for free, whether it is clamped on a tool for a minute of fine assembly work or carrying a box across the floor, and it lets a hand pack more independent fingers into the same space. The early buyers point to where that lands first: the dexterous-hand makers themselves, and electronics assemblers that want robots doing fine-motor work on the line.

Global humanoid robot shipments reached about 13,000 units in 2025, per market researcher Omdia, led by Chinese manufacturers, and Morgan Stanley expects China alone to ship 50,000 this year. Each one needs dozens of reducers. The dominant technology is the harmonic reducer, a coaxial gear (input and output share the same axis) that sits inline between motor and joint. One listed Chinese player, Leaderdrive (绿的谐波, 688017), shipped 425,000 harmonic reducers in 2025 and, per a 2026 disclosure, its share of the harmonic reducers going into Chinese humanoid robots had risen to 80-90%, up from around 70% in 2025.

The coaxial design’s structural constraint means power moves in a straight line, not around a corner. For joints that need a 90° turn, a wrist rotating perpendicular to the forearm or a finger bending off the palm, a harmonic reducer cannot do the job. Among reducer-based designs, the current workaround, the second…

Where capital flows is where talent concentrates, where production capacity gets built, and where the next commercial technologies will come from.

The China Round is free, published weekly, 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.

222 filings this week. The single biggest check went to a company that’s already public, and it went into a subsidiary. A robotics professor raised $100 million twice in 33 days. A June government directive may be tying this week’s AI, chip, and robotics capital into one policy-driven stack. And two city governments wrote checks into the same startup’s round, nine weeks after Beijing told them to stop competing. Here’s where the money went.

Where the Money Went

AI 53 | Semis 39 | Adv. Mfg 42 | Robotics 33 | Energy 29 | Frontier 13 | Aero 13

222 deals | 24 deals disclosed a combined $2.04B (~¥14.7B)

AI led the week on deal count (53 of 222 filings), but most of that volume was small and undisclosed. Only 24 of 222 deals disclosed a figure at all. Semiconductors, with only 39 deals, pulled in nearly half the week’s disclosed capital, $973M of the $2.04B total. Almost all of that came from one deal: ESWIN Materials’ ¥6.5 billion (~$903M) capital increase into its Wuhan wafer subsidiary. Robotics saw two standouts: DISCOVER Robotics raised roughly $100 million twice in 33 days, and Konvery brought three robot manufacturers onto its cap table as strategic investors. In AI, Gradence reported the top coding benchmark score, though we couldn’t find the company on the benchmark sites it references. Five quantum computing deals landed this week, all on the Radar. Aerospace stayed state-anchored, with i-Space closing another ¥1 billion (~$139M) tranche of its E round on the way to a STAR Market listing (the Shanghai exchange’s board for high-growth tech companies).

Spotlight

Gradence (梯度回音)

Angel Round, Undisclosed Amount, Valuation Undisclosed

A multi-agent coding harness running on top of OpenAI’s model claims the top benchmark scores in software engineering.

The China Round is free, published weekly, 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 USD AUM. Now running ZenGen Labs, a cross-border strategy studio covering AI, robotics, and deep tech.

The structural stories this week sit further down the cap table. Three competing humanoid robot makers put money into the same AI-model supplier, a cooking-robot company reported turning profitable, and a Chinese cockpit chip appeared inside a Volkswagen Group vehicle for the first time. Moonshot’s $3.5 billion F round dominates the numbers.

Where the Money Went

AI 35 | Semis 28 | Mfg 35 | Energy 21 | Robotics 14 | Frontier 11 | Aero 8

152 deals | $4.18B disclosed | $683M ex-Moonshot

Three fusion companies raised in one week, with CATL leading its first fusion bet into a 7-month-old startup. Robotics capital went beneath the robot. Three rival humanoid makers backed the same AI-model supplier (Delta Intelligence, nearly ¥500M at a ¥3B valuation) while cooking robots and dexterous hands split the remaining $97M in disclosed robotics capital. Semiconductors stayed state-heavy at later stages, anchored by SiEngine’s $100M for the cockpit chip inside a Volkswagen and Angstrom Excellence’s ¥1B for wafer metrology equipment. AI was almost entirely Moonshot’s $3.5B F round; strip it out and the week split between agent infrastructure and biological AI. Corporate strategics led or co-led an unusually large share of headline deals, with CATL in fusion, Alibaba Cloud in AI ops (while running a competing internal product), China Merchants Capital in cooking robots, and Sungrow and Tianqi Lithium putting a combined ¥805M (~$112M) into battery maker Sunwoda EVB.

Spotlight

Delta Intelligence (德塔智能)

Angel Round (6th round), Nearly ¥500M (~$69M), ¥3B (~$417M) valuation

Humanoid robot foundation-model platform, six rounds in six months.

It is free, published weekly, 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.

A seven-month-old company raised $80M at angel stage, two rival startups in Shanghai raised in the same week for competing visions of AI memory, and AI-generated 3D content produced the category’s largest round to date. Elsewhere, a medical device maker and a landscaping contractor turned up among the investors in an AI chip company that has yet to sell anything. Several deals were timed to coincide with WAIC, where Xi Jinping announced the creation of the World Artificial Intelligence Cooperation Organization (WAICO). The other kind of memory had its own week. As this issue goes out, CXMT (长鑫科技), China’s largest DRAM maker, saw its shares begin trading in Shanghai at ¥49.5 against an ¥8.66 issue price, a 471% jump that briefly made it the most valuable company on the A-share market at over ¥3.3 trillion (~$458B).

Where the Money Went

191 deals this week across AI, robotics, semiconductors, energy, and frontier tech. 22 deals with published exact amounts disclosed a combined $1,469M (~¥10.6B), a strong but not exceptional week by 2026 standards. AI dominated both the deal count and the disclosed capital, and advanced manufacturing spread across the widest range of sub-sectors.

🧠 AI (50 deals)

A cluster of large rounds arrived within days of each other, several timed to coincide with WAIC in Shanghai. Meshy raised ~$400M (~¥2.9B) for AI-generated 3D content, the largest known round in that category globally. FlovaAI, a 7-month-old company founded by the developer of CapCut, raised $80M (~¥576M) at angel stage. Two “AI memory” startups, Red Bear AI and MemTensor, both raised in Shanghai in the same week with competing architectures for persistent AI memory, the ability for language models to remember prior conversations and accumulated knowledge. HiDream.ai (智象未来) pulled in ¥1.5B (~$208M) with China’s Social Security Fund co-leading, a marker of how Beijing categorizes generative AI as strategic infrastructure. Calculet (原粒半导体) raised over ¥700M (~$97M) for AI chips built to run agents on local hardware. Design Order (设序科技) raised its B round for industrial AI software that embeds inside Siemens NX, with plans to enter the European market this year.

🤖 Robotics (33 deals)

The highest-valued robotics round went to Psibot (灵初智能), which raised ~$100M (~¥720M) at a $1.48B valuation with Chery Auto leading and Lens Technology co-investing. A car manufacturer and an Apple/Tesla components supplier are writing the checks for an embodied AI company (AI systems built to control physical robots rather than run as software alone), not traditional VCs. Elsewhere, Xynova (曦诺未来) raised ¥500M (~$69M) for dexterous robot hands, extending a pattern where each of its rounds has been led by a different platform company (CATL, JD, Li Auto, and now Meituan). Yimu Tech (一目科技) hit ¥10B (~$1.39B) valuation on its E round for tactile sensors. Futuring Robot (未来不远机器人) raised ~¥1B (~$139M) at Pre-A for home service robots.

⚡ New Energy & Materials (17 deals)

Mostly undisclosed amounts, with state and corporate capital dominating the cap tables. Xuandian (玄电新材) raised near ¥100M (~$14M) for conductive carbon black, a specialty chemical added to lithium-ion battery electrodes to improve conductivity. TCL’s venture arm led the A round for Tianjiao Xinneng (天交新能), a new energy company. China General Nuclear Power Group (CGN), a state-owned nuclear operator, opened its Yunnan renewable energy subsidiary to outside capital.

🔬 Semiconductors (27 deals)

Mostly undisclosed amounts and mostly state-adjacent capital. Tongyong Semiconductor (通用半导体) disclosed its B round at several hundred million RMB (~$40M+). The rest were smaller or undisclosed rounds across analog chips, micro-sensors, and lithography components. Zhongxin Juyuan (中芯聚源), a fund linked to SMIC, China’s largest chipmaker, appeared in two deals.

💡 Frontier Tech (9 deals)

LiangYi (两仪万象) raised its A+ at ¥2B (~$278M) valuation after setting a world record of 11,000 trapped atoms for neutral-atom quantum computing. Trapped atoms serve as quantum bits (the quantum equivalent of a transistor), so more of them, controlled precisely, means more computing power; LiangYi’s 11,000 nearly doubles the previous record of 6,100, held by Caltech. Lei Jun’s Shunwei Capital has invested in every one of LiangYi’s rounds. Jingjie Juneng (竞界聚能) raised a hundreds-of-millions-of-RMB (~$40M+) angel round at ¥1.5B (~$208M) valuation for a fusion reactor that uses an unconventional approach to confining the superheated plasma that powers the reaction. Jingjie Juneng’s round drew an unusually broad 9-investor syndicate led by Loongson VC, the venture arm of China’s flagship domestic CPU company.

🚀 Aerospace & Defense (8 deals)

Xingchen Daohe (星辰道合) raised a near-¥100M (~$14M) angel round for satellite communications equipment. The remaining seven deals were smaller or undisclosed.

⚙️ Other Advanced Manufacturing (47 deals)

The largest disclosed round went to Lanhe Tech (蓝河科技), which raised a B+ of several hundred million RMB (~$40M+) for UV-LED equipment used to cure coatings and etch circuits in electronics manufacturing. The round was led by the China Internet Investment Fund, a fund under China’s State Council (the country’s top government body). IBDTEC (博顿光电) raised a B+ for ion beam equipment used in manufacturing superconducting tape, a material that carries electricity with zero resistance and is critical for advanced magnets in everything from MRI machines to fusion reactors. The company claims near-100% domestic market share in that niche.

Spotlight

Meshy

B Round, ~$400M (~¥2.9B), ~$1.5B valuation (~¥10B+)

AI 3D content generation platform.

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.

The China Round is free, published weekly, 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.

This is quite a few days late. I appreciate your patience. And fret not, there will be two editions this week. I was busy with WAIC and, more importantly, getting married.

The World AI Conference opened in Shanghai this week with Xi Jinping delivering the keynote, calling for open-source AI cooperation and pledging 5,000 AI training places for developing countries over the next five years. Ten days earlier, at a pre-WAIC press conference, the Ministry of Industry and Information Technology gave its most concrete production target yet for humanoid robots: 100,000 units in 2026, more than four times the roughly 24,000 deployed worldwide at the end of 2025.

Where the Money Went

This week, 159 deals landed across seven sectors, and 25 of them disclosed a combined $1.02 billion (~¥7.3 billion). Most deals came without published figures, so the actual capital deployed was considerably higher. AI led in deal count with 40 transactions, followed by semiconductors (28), robotics (25), and other advanced manufacturing (24). Industrial customers, not venture funds, led several of the largest rounds this week, buying equity directly in the suppliers they need.

🧠 AI (40 deals)

The two largest disclosed AI rounds went to on-device models, not cloud-scale foundation models: ModelBest (面壁智能) accumulated over ¥5 billion (~$694 million) across H1 2026 for models that run on phones and cars, and AIsphere (爱诗科技) closed its C+ at ¥2.98 billion (~$414 million) with Alibaba leading. TASHAN (他山科技) closed a B round with industrial customers as investors. At the early stage, seed and angel deals skewed toward embodied intelligence, with multiple rounds backing “world models” (AI systems that simulate physical cause-and-effect so a robot can predict what happens before it acts) and vision-language-action systems (models that turn a spoken instruction directly into physical movement).

⚡ New Energy & Materials (20 deals)

Power, batteries, and the physical infrastructure beneath the AI boom. Marvel-Tech (慕帆动力) closed a Sequoia-led C round for on-site AI datacenter power generation. ENN Fusion (新奥聚变) opened its first external round at ¥10.6 billion after more than a decade of internal R&D. Ganfeng Battery (赣锋锂电) raised twice in three weeks for solid-state batteries. Linkcharging (领充新能源) raised a strategic C round with multiple automakers. Generating, storing, and delivering power is becoming its own investment category, driven by AI compute demand.

🔬 Semiconductors (28 deals)

State-capital coalitions and strategic investors dominated. Liuyuewei (六岳微) raised ¥500 million in a Series A for domestically developed digital signal processors (the chips that convert real-world signals like sound and radar into digital data), backed by a Guangdong government coalition. Cygnusemi (星思半导体) completed its third round of 2026 for satellite baseband chips (the component in a phone or terminal that decodes wireless signals), reaching ¥12 billion. Shuoguang Precision (烁光精密) attracted two listed-company strategic stakes within weeks for co-packaged optics components, the light-based interconnects replacing copper inside AI data centers. Multiple deals showed listed companies taking direct equity in their own suppliers rather than routing through venture funds.

🤖 Robotics (25 deals)

Capital came from unusual directions. LimX Dynamics (逐际动力) raised $200 million in a Pre-IPO, over 70% from overseas investors. CATL took RoboParty’s (萝博派对) entire Pre-A alone. Shanghai Electric led DexRobot’s (灵巧智能) Series A and co-founded two manufacturing joint ventures with the company. Seven of 25 deals were at the angel stage, suggesting a new wave of company formation. Industrial companies are buying equity in their future robot suppliers, locking in components before the production ramp.

💡 Frontier Tech (11 deals)

The largest round came from one of the oldest companies: Wentian Quantum (问天量子) raised ¥600 million in a Series B after 17 years in quantum communication. Bama (八匹马) raised over ¥100 million for superconducting magnets (magnets cooled to near absolute zero to generate extremely strong magnetic fields) and cryogenics serving quantum computing, fusion, and semiconductor fabrication. Shuyanjuchuang (束研聚创) raised ¥100 million at the angel stage for neutral beam injection systems (equipment that shoots high-energy particles into a fusion reactor’s plasma to heat it enough to sustain a reaction). Two of three deals are component suppliers to other frontier sectors, a sign that the infrastructure layer beneath quantum and fusion is now investable on its own.

🚀 Aerospace & Defense (11 deals)

LegendSpace (临界航天), a six-month-old startup, raised ¥200 million (~$28 million) at the angel stage for AI-accelerated liquid rocket engine development. Galactic Energy (星河动力), China’s most-launched pre-IPO commercial rocket company with 21 flights and 89 satellites delivered to orbit, took an investment from GAC Capital, Guangzhou Automobile Group’s corporate venture arm. GAC Group launched its own flying car brand earlier this year and has backed autonomous driving companies including Momenta and Pony.ai, a portfolio that puts rocket launches closer to the core business than it first appears.

⚙️ Other Adv. Manufacturing (24 deals)

Hardware teams leaving big companies to build consumer products for export. Deeplore (蓝洞探索), three former DJI engineers, raised a seed for an AI diving mask. LightMake (轻量智造), a former AnkerMake lead, raised an angel for 3D printers targeting North America and Europe. Whale (帷幄) raised $40 million from Hyundai and Japanese bank corporate venture arms, bringing its total Series C to over $100 million, more than half its revenue now international. Chinese hardware expertise built at scale inside large companies is spinning out and shipping directly to Western customers.

Spotlight

Deeplore (蓝洞探索)

Seed, Tens of millions of RMB (several million USD), Val. ¥150M (~$21M)

AI diving mask with a heads-up display and real-time marine species recognition.

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.

Happy Friday. China’s first unified outbound investment regulation took effect on July 1, restricting how technology can leave the country. The same week, 141 deals closed across seven sectors. More on those rules below.

Where the Money Went

141 deals across seven sectors. 24 disclosed exact amounts, totaling $2.2 billion (~¥15.6B), with most of the capital concentrated in a satellite mega-round and several hundred-million-dollar AI raises.

🧠 AI (26 deals)

Shengshu Technology (生数科技) closed a $500 million (~¥3.6B) B+ round for Vidu, its AI video platform. Tripo AI raised $150 million (~¥1.08B) in an A+ round for AI-generated 3D assets, its third raise in four months. DeepCtrls (深度智控) pulled in hundreds of millions of RMB with Jinko Solar as a strategic investor (covered below). MindRank (德睿智药) raised $52 million (~¥374M) for AI drug discovery, with its lead candidate now in Phase III (covered below). Below the growth rounds, a cluster of Pudong-backed seed and angel deals continued Shanghai’s pattern of systematic early-stage deployment.

⚡ New Energy & Materials (23 deals)

Starchens (星辰新能) led the sector with a ¥500 million (~$69M) B round backed by CITIC Securities Capital, ICBC Capital, and East Money, a lineup of securities and bank-affiliated investors more typical of pre-IPO rounds. The rest split between battery materials, hydrogen, energy storage, and specialty chemicals, with state-linked capital from Hefei, Nanjing, and Jiangsu backing early- and growth-stage rounds.

🔬 Semiconductors (22 deals)

Twenty-two deals, nearly all with undisclosed amounts. Sicreat (新美光) closed a ¥1 billion (~$139M) C+ round for CMP (chemical-mechanical planarization) polishing pads, a key semiconductor consumable (more below). EDA (chip design software), chiplet packaging, and optoelectronic chips each drew a handful of deals. Several semiconductor materials companies are raising in Suzhou, continuing that city’s emergence as a vertically integrated chip-materials cluster.

🤖 Robotics (18 deals)

RobotEra (星动纪元) closed a ¥1 billion (~$139M) B round led by Chengtong, a fund under SASAC (China’s state-asset regulator), part of ¥2.5 billion (~$347M) raised across three rounds in roughly 60 days. Morphi (墨奇智能) pulled in hundreds of millions of RMB in what is reportedly one of the largest disclosed angel rounds in China’s embodied-AI sector (AI designed to operate physical hardware like robots), with Alibaba and Tencent co-anchoring for a six-month-old company. Motorevo (泉智博) completed its seventh round in 18 months for humanoid robot joint modules.

💡 Frontier Tech (8 deals)

Two brain-computer interface companies raised in the same week: Gestala (格式塔科技), a non-invasive ultrasound BCI startup, closed a ¥442 million (~$61M) angel+ round (covered below), while NeuroXess (脑虎科技) raised an A+ round as it began Class III clinical trials at Huashan Hospital. QBoson (玻色量子) closed a Pre-IPO round and filed for IPO coaching (the formal process of engaging an underwriter to prepare a company for listing) three days later. SynMetabio (贻如生物), a synthetic biology company making cultivated leather for luxury brands, closed its A round with Ordos Group co-leading (covered below).

🚀 Aerospace & Defense (16 deals)

Changguang Satellite (长光卫星) dominated with a ¥5 billion (~$694M) strategic round, the largest single raise in the sector this week. The company operates 161 satellites, the world’s largest sub-meter-resolution commercial remote-sensing constellation. Space Pioneer (天兵科技) added ¥200 million (~$28M) in strategic capital. Kaipu Dynamics (开普动能), an eVTOL (electric vertical takeoff and landing) electric propulsion startup founded by a former Rolls-Royce division director, raised an angel round (covered below). Most of the remaining deals were seed and angel rounds for early-stage space-tech and low-altitude economy startups, including Ruiqi Deep Space (瑞启深空), a four-month-old constellation startup that raised ¥220 million (~$31M) at a ¥1.1 billion (~$153M) valuation from Suzhou state funds.

⚙️ Other Adv. Manufacturing (28 deals)

Snapmaker (快造科技) raised ¥1 billion (~$139M) in a C round for consumer 3D printing (covered below). CLAWLAB (浪爪智能) raised hundreds of millions of RMB with miHoYo leading (covered below). Thinkhead (上海星合), a 17-year-old bootstrapped machine-tool manufacturer, took its first outside capital. Coffee robot maker InSpace/XBOT (影智科技) closed a B round. The rest spanned optics, precision measurement, industrial equipment, and microfluidics (tiny channels etched into chips to manipulate fluids for medical diagnostics and lab work).

Spotlight

Gestala (格式塔科技)

Angel+, ¥442M (~$61M), Valued at ¥4.2B (~$583M)

Non-invasive ultrasound brain-computer interface.

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.

Happy Friday everyone. Let’s get straight into it.

Where the Money Went

163 deals across seven sectors. 37 disclosed exact amounts, totaling $5.9 billion (~¥42.4B). Strip out Kling AI’s $3 billion debut raise and the remaining 36 deals still disclosed $2.9 billion (~¥20.8B). Robotics led on volume (37 deals).

🧠 AI (35 deals)

Kling AI’s $3 billion first external raise dominated the sector and the week (covered below). Several ¥1 billion+ growth rounds closed, including DexForce Technology and Tripo AI (both in Radar below), and Xellar Biosystems (耀速科技), whose organ-on-chip platform screens drug candidates on chip-based organ models (covered below). At the other end, a cluster of seed and angel rounds backed brand-new entrants. Klaren Intelligence (厘清智能), a startup building world models, raised several hundred million RMB with HongShan (红杉), Hillhouse, and Shunwei all on the cap table for a seed round. Founded in April 2026. Eighteen AI companies raised at valuations of ¥100 million (~$14M) or less.

⚡ New Energy & Materials (17 deals)

Mostly sub-¥100 million (~$14M) rounds across battery materials, energy storage, and specialty chemicals. Yingong Technology (隐功科技) raised ¥100 million (~$14M) in a Pre-A for nanocomposite materials, and Anputai (安普态) closed a near-¥100 million A round in advanced insulation, both backed by state-linked funds.

🔬 Semiconductors (31 deals)

31 deals, led by Gowin Semiconductor’s Pre-IPO (covered below). EDA, chiplet packaging, and MEMS sensors each drew 1 to 3 deals, alongside camera sensor chips and power management ICs. Qili Semiconductor (齐力半导体) raised ¥100 million (~$14M) for chiplet packaging at Zhangjiang, and Jiyi Semiconductor (集溢半导体) raised ¥130 million (~$18M) for analog chip design. State capital appeared in investor lists across the sector, with Guangdong, Jiangsu, and Zhejiang government funds all active.

🤖 Robotics (37 deals)

The most active sector by deal count, with three mega-deals (AI2 Robotics, X Square Robot, and Boundless Motion, all covered in Radar below). Below those, the deals split between platform-backed humanoid companies and smaller startups building components, sensors, or manipulation software. Fifteen deals closed at valuations of ¥100 million (~$14M) or less.

💡 Frontier Tech (11 deals)

Eleven deals. SpinQ (量旋科技) raised ¥1 billion (~$139M) in a D-round for quantum computing (more in Radar below). Xinghe Fusion’s (星核聚变) ¥833 million (~$116M) debut was the largest first-round raise in Chinese private fusion, backed by 25 investors including SAIC Motor. Two other fusion companies also raised this week, Andong Jubian (安东聚变) and Chaoci Xinneng (超磁新能), making it a three-fusion-deal week. SynSense (时识科技) closed several hundred million RMB for neuromorphic chips, event-driven processors that react only to changes in a scene rather than analyzing every frame, cutting power consumption for always-on cameras and sensors.

🚀 Aerospace & Defense (8 deals)

Eight deals, dominated by Hongqing Technology’s ¥1.3 billion (~$181M) Pre-B for its satellite constellation (covered below). Puxing Aerospace (谱星航天) raised several hundred million RMB across combined rounds for optical remote-sensing payloads, with plans for a 1,024-satellite constellation.

⚙️ Other Adv. Manufacturing (24 deals)

24 deals spanning optoelectronics, industrial equipment, and consumer hardware. Sinoscience Fullcryo’s ¥1 billion (~$139M) Pre-IPO for cryogenic equipment (covered below) was the sector’s largest. CAYE raised ~¥400 million (~$56M) in a B-round for coffee machines (covered below). At the other end, AIVELA raised several million dollars for its smart ring.

Spotlight

Kling AI (可灵AI)

First External Round, $3B (~¥21.6B), Valued at ~$18B (~¥129.6B)

Kuaishou’s video generation platform raised its first outside capital, with $500M in annualized revenue and a Hong Kong IPO planned within 12 months.

A quarter of Xinjiang’s annual generation is exported east across transmission lines, capturing generation revenue but little else for the province. Much of the renewable surplus beyond that is curtailed: the transmission lines are at capacity and local demand is too low to absorb it.

Renewables account for over three-quarters of Xinjiang’s installed power capacity but only a quarter of its actual generation: wind and solar produce intermittently, and when they do, there is often nowhere for the power to go. Of what is exported east, less than a third is green.

A policy framework introduced in China’s 2026 government work report proposes capturing some of that surplus by converting electricity into AI compute. The framework is called 算电协同 (”compute-power coordination”) and Xinjiang is the first pilot region.

If all goes to plan, Xinjiang aims to reach 60 exaflops of compute capacity by the end of this year, with construction underway across four cities.

AI’s power problem

For most of the past few years, AI infrastructure headlines centered on chip supply, but that constraint is easing. Nvidia has scaled production, hyperscalers are deploying custom silicon, and China’s Huawei shipped the Ascend 950PR in Q1 2026 with roughly 2.8x the inference throughput of Nvidia’s H20 and its own domestically produced HBM. Government mandates require at least 80% domestic AI chips across the planned data center buildout.

Memory remains tight globally: SK Hynix and Samsung still control over 95% of HBM supply outside China, and Chinese memory maker CXMT’s placement on the Pentagon’s 1260H list has complicated supply chains enough that Apple is lobbying the White House for guarantees it can continue sourcing from the company. But the harder constraint is power.

In the US, new grid connections can take years. At the World Economic Forum in Davos in January, Elon Musk warned that the industry would soon be “producing more chips than we can turn on.” He put it bluntly: “The limiting factor for AI deployment is fundamentally electrical power.” He added: “except for China.”

In the US, over 2,000 GW of generation and storage capacity is waiting to connect to the grid. The median project that reached commercial operation in 2025 spent over five years in the queue. The International Energy Agency estimates that around 20% of planned data center projects could be at risk of delays. FERC responded in June with orders to fast-track data center grid connections, but new facilities may still need to bring their own power or pay for grid upgrades.

Pent-up energy

China generates more than twice as much electricity as the United States. BloombergNEF projects it will add almost six times US electricity generation capacity over the next five years. A lot of that will be renewables. In 2025 alone, China increased wind and solar capacity by over 430 GW, more than half of all global renewables additions that year.

Much of this capacity is concentrated in western provinces, where land is vast, populations are sparse, and wind and solar resources are among the strongest in the country. Xinjiang’s geography makes the case: 1.66 million square kilometres of mostly arid basin and steppe, flanked by the Tianshan mountains, with some of China’s highest solar irradiance and wind corridors like Dabancheng that rank among the country’s best. There is plenty of room for turbines and panels, and plenty of wind and sun to power them. But local demand is thin.

Xinjiang crossed 100 GW of wind and solar installed capacity at the end of 2024, a first for Northwest China. Including hydro and other renewables, total clean energy capacity reached 173 GW within a total grid capacity of 219 GW by mid-2025. But the province’s ability to use all that power locally has lagged behind its ability to generate it. Conditions have improved from historical lows, but not enough: as of early 2025, only 67.6% of available renewable generation was actually used, meaning a third still went to waste.

The traditional solution has been to export the surplus east via ultra-high-voltage DC transmission lines: Hami-Zhengzhou (commissioned 2014), Changji-Guquan (2019, at 3,293 km and 12 GW capacity the world’s longest), and Hami-Chongqing (2025). Changji-Guquan alone has delivered 84 billion kWh of green electricity since commissioning, all consumed at the far end in Anhui.

At projected 2026 curtailment rates, Xinjiang’s renewable fleet will leave untapped the equivalent of more than eight times the annual electricity consumption of Elon Musk’s xAI Colossus supercomputer.

Building compute where the power is

China’s first attempt to build compute in the west was 东数西算 (”East Data, West Compute”), launched in 2022. The program designated 8 national compute hub nodes, including Inner Mongolia, Guizhou, Gansu, and Ningxia, and built real infrastructure: 633 hyperscale and large data centers delivering 268 exaflops of national computing power by 2024.

The program built real capacity, but utilization lagged. The Oxford Institute for Energy Studies found energy savings of 4 to 12% from moving data centers west, though western centers often lacked the fiber infrastructure to serve eastern clients in real time, and the companies, talent, and customers remained concentrated on the coast. Western data center utilization rates settled at 20-30%. Over 100 data center projects have been scrapped nationally in the last 18 months, and the government imposed a moratorium on new large-scale facilities in cities where existing centers operate below 50% utilization.

Compute-power coordination, the successor framework, takes a more focused approach. Rather than trying to solve those demand-side problems directly, it asks whether energy economics alone can pull enough latency-tolerant AI workloads west. It bypasses the moratorium by granting energy-rich provinces the same policy benefits as the original 8 hub nodes based on their power resources, not administrative designation. All new data centers in hub regions must source at least 80% of their electricity from renewables, a target western provinces already meet by default.

Xinjiang, which was not one of the original 8 nodes, was granted the same pilot status. It then launched what its authorities describe as China’s first multi-user 绿电直连 (”green electricity direct connection”) pricing, bypassing the traditional grid to connect power generation directly to the data center at 0.31 yuan per kWh (~$0.045/kWh). That is 0.04 to 0.07 yuan below the standard grid-delivered tariff and cheaper than Inner Mongolia’s 0.35 yuan per kWh (~$0.051/kWh), which Xinhua describes as one of the lowest among the national data center clusters.

Despite frequent claims that China has an energy cost advantage over the West, the Oxford study found that China’s industrial electricity prices are broadly comparable to those in the US. A data center in Guangdong or Jiangsu faces delivered costs of RMB 0.60-0.75 per kWh (~$0.08-0.14/kWh), near or above industrial rates in Virginia, the largest US data center market, because coastal Chinese provinces layer grid charges, capacity fees, and cross-subsidy surcharges on top of generation costs. Western green direct connection pricing at 0.31 yuan is roughly half the eastern rate, which is the real incentive for moving workloads west.

In April 2026, four central agencies led by the National Energy Administration jointly published a 29-item action plan targeting a “safe, green, economical energy guarantee system” by 2027 and “world-leading” clean energy supply for AI compute by 2030. The plan calls for million-kW-scale AI compute facilities with coordinated energy systems, multi-year green power contracts, and infrastructure REITs as a financing vehicle for compute facilities.

The NDRC is drafting a plan to spend 2 trillion yuan (~$295 billion) over 5 years on a nationwide AI data center grid, funded by sovereign debt and ultra-long special government bonds, with China Mobile and China Telecom as primary operators. The total could exceed 5 trillion yuan (~$740 billion) if power grid upgrades are included. Compute-power coordination is one of the frameworks that determines how energy-rich provinces participate in this nationwide buildout.

The demand wave

The infrastructure is being built to meet a compute demand curve that is still steepening. Deloitte estimates that global demand for AI compute is rising four to five times per year through 2030, driven primarily by inference rather than training. Inference workloads now account for roughly two-thirds of all AI compute, up from a third in 2023 and half in 2025. The market for inference-optimized chips alone is projected to exceed $50 billion in 2026. In China, computing-related electricity consumption rose from 82.4 billion kWh in 2019 to 196 billion kWh in 2025. The China Academy of Information and Communications Technology projects it could reach 500 to 700 billion kWh by 2030, up to 5.3% of total national demand.

China’s domestic AI ecosystem is generating much of that demand. Chinese platforms processed 140 trillion tokens per day in Q1 2026, a more than 1,000-fold increase from 100 billion per day in early 2024. Chinese open-source models surged from 1.2% of global usage in late 2024 to nearly 30% by mid-2025, and by June 2026 accounted for roughly 48% of token usage on OpenRouter, overtaking US models, with DeepSeek, Qwen, and Zhipu’s GLM leading the expansion.

Goldman Sachs estimates that China’s top internet firms will invest over $70 billion in data center capital expenditure in 2026, and reports that Chinese data center power demand rose 25% in 2025.

The growing dominance of inference might seem to work against Xinjiang’s case, since inference typically runs close to the end user. But much of the fastest-growing inference demand is not real-time. User-facing inference (search, voice assistants) needs to run close to the user and will stay in eastern cities. Training is less latency-sensitive but faces its own constraint: as one executive told Caixin, the datasets are so large that “companies sometimes have to physically transport hard drives” rather than transmit them east to west. Planned network infrastructure investment aims to close that bandwidth gap. But once the data arrives, training runs are exactly the kind of sustained, power-hungry workload that suits cheap western electricity. The same is true of batch processing, agent workflows, and offline industrial analytics, and these workloads are growing fastest. In January 2026, eight ministries set a target of 1,000 industrial AI agents deployed across manufacturing, logistics, and public services by 2027. Some of these, like real-time robotics control, need to run locally. But many, like logistics optimization, demand forecasting, and document processing, are batch workloads that need cheap, reliable power and a network connection, not proximity to the end user. That is the market compute-power coordination is designed to serve.

The funding landscape

In April 2025, China launched a ¥60 billion (~$8.3 billion) state-backed AI fund to invest across the full AI industrial chain, with its first-year focus on semiconductors and computing infrastructure. ByteDance’s 2026 capital expenditure alone is ¥160 billion, over half of it on AI chips. Alibaba has committed ¥380 billion over three years to cloud and AI hardware infrastructure.

Below the hyperscalers, venture capital is filling the infrastructure stack. Some notable raises this year across the energy and cooling layers of AI compute:

Qingqi Power (清启动力): micro gas turbines for distributed data center power. Pre-A round, nearly ¥100 million (~$14 million), ¥500 million valuation. Facilities that cannot wait on the grid use on-site turbines to generate power where the connection has not yet arrived.

Xinhan Intelligence (芯寒智能): cooling systems for AI compute clusters. Pre-A round, hundreds of millions of yuan, ¥1.5 billion valuation. AI accelerators run hot and pack tightly, and at cluster scale the heat load outstrips what air cooling can handle. Cooling also sets PUE, the main lever on a facility’s operating cost.

Afalight (埃尔法光电): co-packaged optics. D round (sixth total), ~¥100 million, ¥1.2 billion valuation. Supplies a 2.4 Tbps optical engine to Alibaba’s AI data centers. As clusters scale, the copper links between racks and the network become the bottleneck on data movement, and co-packaged optics moves the transceiver onto the switch package to push interconnect bandwidth past that limit.

MetaPWR (沛塬电子): power modules for AI compute. Strategic round, undisclosed amount, ¥500 million valuation. AI racks pull very high current at low voltage, and the rack’s power conversion stage has to deliver that cleanly and reliably. Generic modules are not rated for the load profile.

Four cities, one value chain

Four cities are dividing the work: Karamay runs the most efficient facilities, Changji is building new capacity with major tech partners, Hami dispatches compute to Chongqing, and Urumqi connects westward to Central Asia.

1. Karamay (克拉玛依) has some of the most energy-efficient data center infrastructure in China. Carbon and Network Tech runs facilities with a PUE under 1.1 (PUE measures how much energy goes to compute vs. cooling and overhead; a perfect score is 1.0, and typical facilities run 1.3-1.5), with cooling and overhead energy one-fifth that of a conventional data center thanks to liquid cooling and Xinjiang’s cold, dry climate. The company has used 100% green power via green electricity certificates since 2024 and was listed on the 2025 National Green Compute Facilities roster. Karamay dispatches compute to Shanghai under a cross-regional framework.

2. Changji (昌吉) broke ground in May 2026 on a green direct-connect photovoltaic project paired with a China Unicom 16,000P intelligent compute infrastructure build at the Xinjiang Information Industry Park. Alibaba, ByteDance, and Zhipu have all reached cooperation agreements, with scheduled completion in November 2026.

3. Hami (哈密) has been designated by the National Data Bureau as an implementation city for both the Xinjiang-to-Chongqing compute corridor and compute-power coordination. A dedicated all-optical network link delivers under 20 ms one-way latency to Chongqing across 4,000 km, with 20,000P of compute already operational and a target of 40,000P by year-end.

4. Urumqi (乌鲁木齐) is building a national-level internet backbone direct connection point alongside a BRICS digital industry cooperation network city node, positioning itself as a compute hub serving Central Asia and Europe. Xinjiang borders Kazakhstan, Kyrgyzstan, Tajikistan, Pakistan, and Mongolia, and has more than 20 cross-border international fiber cables connecting to those neighbors, recently upgraded to 1.2 terabits (China-Kazakhstan) and 800 gigabits (China-Pakistan). The provincial government describes its cross-border compute services as “算力出海” (”compute going overseas”). So far, the only confirmed services crossing these borders are satellite remote sensing and smart agriculture from Xinjiang’s Space-Air-Ground Integration Lab. The fiber infrastructure exists; commercial AI compute demand from Central Asian markets has yet to materialize. Eight intelligent compute center projects are under construction, and companies including Beijing Wanjie Data and Hefei-based Leinao have established operations. The Xinjiang Daily describes Leinao’s facility here as the first compute center using domestically produced chips. Leinao’s Bitahub platform aggregates self-built and external compute resources and reports a utilization rate above 90%, but that is a platform-level figure, the result of a marketplace that resells compute across multiple facilities. Facility-level utilization in western hubs remains at 20-30%. Whether a platform model can close that gap at scale beyond a single marketplace is untested.

Electricity to compute to tokens

Jin Qian, president of Carbon and Network Tech in Karamay, told the Xinjiang Daily: “Xinjiang is thinking about how to convert electricity into compute, compute into tokens, and serve the world via the internet.”

China does not need to export chips, models, or hardware to serve AI demand beyond its borders. It can deliver tokens over fiber, keeping the infrastructure, energy, and jobs domestic. Each step retains more value locally than the last: raw electricity captures generation revenue, compute adds infrastructure and operations, tokens add the service layer. Geopolitical and data-sovereignty constraints will shape which markets those tokens can realistically reach, but the economic logic of the conversion chain holds regardless.

The electricity-to-tokens framing has taken hold in other compute hubs too. Inner Mongolia’s Ulanqab, just over 200 km from Beijing with 4.2 ms fiber latency, has branded itself China’s “Token之都” (Token Capital) under a three-year action plan (2026-2028), with over 5 million standard server racks signed and 165,000P of running compute. But Ulanqab’s success leans on proximity to Beijing. Xinjiang’s pitch is different: the cheapest green power in the country, direct-connect pricing that undercuts even Inner Mongolia, and cross-border fiber to Central Asia that no eastern hub can match.

What has to go right

SMIC’s co-CEO Zhao Haijun warned on the company’s Q4 earnings call that rushed AI data center capacity could sit idle, describing the pace as “10 years of capacity crammed into two.”

Caixin, citing expert sources, describes compute-power coordination as being in an “early exploration period,” and draws a technical distinction between 电量 (”electricity volume”), the total energy available, and 电力 (”electrical power”), managing instantaneous demand. Xinjiang has plenty of 电量, but the real challenge is 电力. AI training clusters draw power in sharp, irregular spikes that solar and wind cannot match on demand. The green direct connection model that bypasses the grid also bypasses this buffer, so facilities need either substantial battery storage or a robust grid fallback. Some are already building this: a State Power Investment Corporation facility in Xinjiang operates entirely off-grid with on-site solar, wind, and battery storage, and CATL committed over $1.5 billion to data center energy infrastructure in two months, including $577 million into data center power supplier Zhongheng Electric in April and up to $942 million into data center operator VNET Group in May.

Data centers still account for only 1.68% of China’s total electricity consumption, projected to reach 3% by 2030, a narrower scope than CAICT’s broader “computing-related” estimate of up to 5.3%, which includes telecom and edge infrastructure. Compute-power coordination will not solve the curtailment problem in volume terms. The case rests on value density: a kilowatt-hour converted into AI tokens captures far more economic value than one exported as raw electricity, even if the total kilowatt-hours consumed remain a sliver of the surplus.

The timeline mismatch compounds the challenge: large computing centers can be built in 8 to 24 months, but traditional power substations require 3 to 5 years for planning, approval, and construction. Green direct connection lets data centers start operating before the substation arrives, but the grid fallback they will eventually need still takes 3 to 5 years to build.

The Xinjiang Daily itself acknowledges the challenges: “算力基础设施投资大、回报周期长” (”compute infrastructure investment is large, payback period is long”), along with the need for better cross-regional networks and talent.

Conclusion

The question this framework has to answer is the same one East Data, West Compute could not: whether workloads will actually move. Compute-power coordination addresses the cost side of that equation with 0.31 yuan/kWh green direct-connect pricing and equivalent hub-node status. What it does not address is the commercial gravity of the eastern seaboard, the dense concentration of developers, customers, and corporate headquarters that resisted the first wave of westward migration.

If the framework succeeds, it establishes a model with implications well beyond Xinjiang: that energy-rich, compute-poor regions can compete for AI workloads on electricity price rather than proximity, provided they target latency-tolerant workloads, whether training or batch inference, rather than real-time services. That would reframe western China’s renewable surplus as upstream infrastructure for a national or even global AI services layer. The energy still reaches demand, but as tokens over fiber rather than electrons over high-voltage lines, and the bottleneck shifts from the rigid capacity of a transmission corridor to the elasticity of a network. At the token layer, remoteness stops being the obstacle and becomes the advantage: stranded power converted into tokens, carried over fiber to markets the grid was never built to reach.

Where capital flows is where talent concentrates, where production capacity gets built, and where the next commercial technologies will come from.

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.

This week, a robotics company founded in 2015 with over ¥1 billion (~$139M) in annual revenue filed for its Hong Kong IPO at a ¥19.5 billion (~$2.7B) valuation. The same week, a robotics company founded 90 days ago with zero product raised at a ¥9 billion (~$1.25B) valuation. Both are rational bets. 127 deals spanned every sector, early-stage rounds continued to swell to growth-stage sizes, and the most interesting capital flows crossed sector boundaries.

Where the Money Went

127 deals this week, 49 of which disclosed amounts totaling approximately $1.75 billion (~¥12.6B). Strip out COOWA’s $600 million (~¥4.3B) IPO round and Tongwei Solar’s ¥5.06 billion (~$703M) restructuring, and 47 deals disclosed a combined ~$442 million. Most Chinese deals don’t disclose amounts, so the actual total is higher.

Stage distribution: 42 angel, seed, and pre-A rounds, 49 at Series A/A+, 15 at Series B or later, and 21 strategic investments. The early-stage dominance reflects a market where the embodied AI and physical AI boom is still generating new companies faster than it is graduating existing ones.

🧠 AI (25 deals)

The largest AI cluster spans physical AI infrastructure, video generation, and 3D content. Lightwheel AI (光轮智能) raised ¥1 billion (~$139M) for robot simulation data from a syndicate that includes three listed gaming companies. Sand AI raised over $100 million for video generation that competes with OpenAI’s Sora. Deemos (影眸科技) raised hundreds of millions of RMB for 3D generative AI with clients including Unity, Figma, and Canva. Jiangxing AI (江行智能) closed its third fundraise of 2026, backed by CATL-affiliated capital, Jinkosolar, and Yanzhou Energy as customer-investors.

⚡ New Energy & Materials (16 deals)

The headline is Tongwei Solar’s ¥5.06 billion (~$703M) restructuring, which dwarfs everything else. Beyond that, the energy transition stories are more interesting than the numbers suggest. HYDOtech (海德氢能) attracted a repeat investment from Saudi Aramco’s venture arm alongside Conch Group’s private equity arm, one of China’s largest cement producers. HYLICreate (合源锂创) landed a Xiaomi fund-led round for solid-state batteries with a live factory. Solide (索理德) raised for silicon-carbon anode materials at volume production scale.

🔬 Semiconductors (26 deals)

The second-largest sector by deal count spans optical interconnect chips, AI inference silicon, and specialized sensors. Aluksen Technology (傲科光电子) closed a B++ for high-speed optical chips that connect GPUs inside AI data centers. AIPhotonix (慧芯激光) raised to build its own manufacturing line for gallium arsenide and indium phosphide optical chips, targeting the same AI data center interconnect market. Transtreams (凌川科技) closed an A+ for a video compression chip that outperforms NVIDIA on compression efficiency benchmarks. Juxin Semiconductor (聚芯半导体) filed for a Hong Kong IPO as the world’s fourth-largest optical sensor company by revenue.

🤖 Robotics (20 deals)

COOWA (酷哇科技) is raising $600 million (~¥4.3B) ahead of a Hong Kong IPO at a ¥19.5 billion (~$2.7B) valuation, with SoftBank China Venture Capital backing. At the other extreme, Kunlun Xing (昆仑行机器人) raised an undisclosed amount at a ¥9 billion (~$1.25B) valuation 90 days after founding, with zero product. AGILINK (临界点) hit unicorn status at ¥10 billion in its B round, five months after founding, though unlike most instant unicorns it actually has revenue and 8,000+ dexterous robot hands shipped.

💡 Frontier Tech (4 deals)

The smallest sector by count but one deal carries outsized signal: Juhe Jubian (聚合聚变), a fusion energy startup in Hefei, raised an angel round co-led by Hillhouse Ventures and Sequoia China, the third fusion deal in three consecutive weeks across three distinct fusion approaches.

🚀 Aerospace & Defense (8 deals)

Argo Space Technology (吉天星舟) raised a B round as a profitable space-tech component supplier with 55 payloads delivered and 30 in orbit at a 100% success rate. Jitai Aviation (吉太航空) raised for a modular electric vertical takeoff and landing aircraft, an air taxi that takes off like a helicopter but runs on electric motors, targeting China’s government-designated “low-altitude economy” covering commercial airspace below 1,000 meters.

⚙️ Other Advanced Manufacturing (28 deals)

The largest sector by deal count spans industrial automation, optoelectronics, and specialized manufacturing. Highlights include Fubusi (辅布司), an AI textile cutting company, and Botler Robotics (博特勒), building high-temperature superconducting magnets for fusion reactors.

Spotlight

COOWA (酷哇科技)

Latest round, $600M (~¥4.3B), ¥19.5B (~$2.7B) valuation

Autonomous urban robots that actually make money.

Where capital flows is where talent concentrates, where production capacity gets built, and where the next commercial technologies will come from.

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.

This past week, DeepSeek raised over $7 billion (~¥51B) in its first-ever external funding round, the largest single financing event in Chinese AI history. The rest of the data is just as revealing: 93 deals in total, a wave of new company formations in physical AI, and at least six companies raising in the same week to build “world models,” AI systems that predict physical cause and effect so robots can plan actions before executing them.

Where the Money Went

Ninety-three deals landed between June 12 and 18. Nineteen disclosed amounts, combining for $8.7 billion (~¥62.6B) in total disclosed capital. DeepSeek alone accounts for 82% of that figure. Strip it out and 18 deals still disclosed a combined $1.6 billion (~¥11.5B), a strong week by any measure.

Roughly 43% of deals were angel or seed rounds, well above the typical weekly mix. This skew reflects a physical AI company-formation wave, with new teams spinning out of established labs and raising within weeks of incorporation.

🧠 AI (24 deals)

AI dominated both deal count and capital. DeepSeek’s ¥51 billion (~$7B+) A round accounted for the vast majority of that capital, but the sector had depth beyond it: LiblibAI raised $300 million (~¥2.2B) for its AI creative platform, SiliconFlow closed a ¥2 billion (~$278M) B round for inference infrastructure, and 卜拉格科技 raised ~$220 million (~¥1.6B) in its first institutional round with no shipped product. A cluster of companies raised to build “world models” for robotics. Physis, Manifold AI, and Synapx all closed world model rounds in the same window. More on this pattern in The Bigger Picture.

🤖 Robotics (13 deals)

Thirteen robotics deals disclosed $278 million (~¥2B). SEAHI Robotics pulled in ¥1 billion (~$139M) for its marine robots, while GigaAI closed a ¥1 billion (~$139M) B2 round for its humanoid platform. ACE Robotics, which builds AI software brains for robots, raised nearly $100 million (~¥720M) at angel stage with a ¥2.6 billion (~$361M) valuation. NOITOM Robotics, a leading player in the global professional motion capture market, raised several hundred million RMB (~$30-130M) for its pivot into robot training data. At the smaller end, a half-dozen angel-stage robotics companies each raised single-digit millions.

🚀 Aerospace & Defense (8 deals)

Eight deals, $264 million (~¥1.9B) disclosed. Emposat (航天驭星) led with a ¥1.4 billion (~$195M) D+ round for satellite ground operations ahead of an IPO filing. ZEROG (零重力飞机工业), which makes both conventional electric fixed-wing aircraft (already shipping, over 1,000 letters of intent) and electric aircraft designed for vertical takeoff and landing (in development), closed a ¥500 million (~$70M) B round led by three Hefei government funds. Juneng Tianqing (聚能天擎) raised several hundred million RMB (~$30-130M) for methane rocket engines at angel stage.

💡 Frontier Tech (6 deals)

Six deals, $125 million (~¥900M) disclosed. Taiyi Quantum (太一量生) raised ¥300 million (~$42M) at Pre-A for its neutral-atom quantum computer, which uses ytterbium atoms suspended in a laser trap rather than the superconducting qubits favored by IBM and Google. The approach runs at room temperature and makes fewer calculation errors, the main bottleneck in current quantum systems. Xeonova (星能玄光) closed a ¥500 million (~$70M) A round for commercial nuclear fusion. Nuclear fusion, the reaction that powers stars, would produce nearly unlimited clean energy if it can be sustained cheaply. Most approaches require enormous reactor chambers costing billions; Xeonova uses a compact alternative reactor design that reaches initial plasma ignition at far lower cost.

🔬 Semiconductors (6 deals)

Six semiconductor deals, mostly with undisclosed amounts. The most notable chip deal of the week was NanoCore Chip (微纳核芯), which raised ¥1 billion (~$139M) in a B+ round for its RISC-V processor platform. RISC-V is an open-source chip architecture, the open alternative to ARM’s proprietary designs, important because it avoids licensing dependencies on Western IP and gives Chinese manufacturers a path around US export controls on chip technology. The sector was quieter than in recent weeks.

⚡ New Energy & Materials (3 deals)

Three deals, all with undisclosed amounts. TCL’s venture arm backed an early-stage new materials company, and two strategic rounds landed in specialty chemicals and clean energy.

⚙️ Other Adv. Manufacturing (33 deals)

Thirty-three deals across optoelectronics, industrial equipment, quantum instruments, display materials, and precision manufacturing. Most were early stage with undisclosed amounts. The Hefei government continued its aggressive cluster-building, leading or co-leading rounds in advanced materials and quantum instrument companies alongside its aerospace investments.

Spotlight

DeepSeek (深度求索)

A round, ¥51B (~$7B+), valuation ~¥400B (~$55.6B)

China’s top AI lab takes its first outside money, on its own terms.

Where capital flows is where talent concentrates, where production capacity gets built, and where the next commercial technologies will come from.

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.

As this report went to press, China’s drug regulator accepted an IPO application from Neuracle (博睿康), the company behind the world’s first commercially approved invasive brain-computer interface. It filed for a ¥2.5 billion (~$347M) listing on Shanghai’s STAR Market, the tech-focused board that serves a similar role to Nasdaq for Chinese companies. That timing is coincidental, but the signal is not. China’s frontier medical devices are moving from lab to market to public listing faster than most investors expected.

Where the Money Went

This week’s scan captured 109 deals across all sectors, with 27 additional stragglers from last week’s gap. Origin Quantum’s ¥3 billion (~$417M) quantum computing Pre-IPO dominated the headline numbers. Strip that out and 16 deals disclosed a combined ~$443M (~¥3.2B), a solid mid-tier week. Robotics and AI continued to lead deal volume, but the week’s most interesting stories came from frontier tech and advanced manufacturing.

🧠 AI (22 deals)

AI deal flow stayed strong at 22 rounds, though this week skewed toward application-layer companies rather than foundation models. The standout was CreativeFitting (井英科技), which closed both an A and A+ round backed by Ant Group for its overseas AI short drama app, Reel.AI. HiDream.ai (智象未来) received a strategic investment from Huace Film (华策影视), one of China’s largest private TV drama producers. A cluster of early-stage deals in spatial intelligence, AI education, and virtual humans filled out the sector.

⚡ New Energy & Materials (19 deals)

Materials and new energy logged 19 deals, almost entirely early-stage. MATTERENE (烯材科技), co-founded by Nobel laureate Andre Geim, raised a Pre-A+ from Aviation Industry Investment for its graphene film production facility in Shenyang. Advanced ceramics, specialty metals, and battery materials accounted for the bulk of deals, with state-backed funds leading most rounds.

🔬 Semiconductors (18 deals)

Semiconductor deal count held steady at 18. The week’s standout was Anatrix (安纳智芯), a six-month-old Peking University spinout whose RRAM analog computing chip earned a Nature Electronics paper and a Series A led by Matrix Partners China (经纬创投). Raysees (瑞识科技), the Samsung Galaxy Ring sensor supplier, took strategic capital from GAC at a ¥3.5 billion (~$486M) valuation. Wisewave (芯潮流), which designs the high-speed data link circuits that let chips inside servers and AI clusters talk to each other at hundreds of gigabits per second, raised a Series A from Sequoia China and CITIC Securities.

🤖 Robotics (16 deals)

Robotics logged 16 deals and produced some of the week’s biggest stories. Dexmal (原力灵机) closed a B round at ¥6 billion (~$833M) valuation backed by three of China’s top large language model companies: Zhipu AI, StepFun (阶跃星辰), and SenseTime. LiberAI hit a ¥3 billion (~$417M) valuation on a Pre-A. Inspire Robotics (因时机器人) raised a C+ at ¥2.6 billion (~$361M), having shipped over 10,000 dexterous hands in 2025. Qianjue Robotics (千诀科技) raised an A round at a ¥5 billion (~$694M) valuation for its robot decision-making and planning platform.

💡 Frontier Tech (9 deals)

Frontier tech produced the week’s largest round and its strongest stories. Origin Quantum (本源量子) raised ¥3 billion (~$417M) in a Pre-IPO with China North Industries Group (中国兵器集团), a defense conglomerate, among the lead investors. A nuclear energy cluster formed: Dongsheng Fusion (东昇聚变) raised $100 million (~¥720M) to build a tokamak, a doughnut-shaped reactor that confines superheated plasma with magnetic fields to fuse hydrogen isotopes into helium, releasing energy. Its D-He3 fuel choice produces near-zero radioactive waste. Weilan Zhidian (蔚蓝支点) raised hundreds of millions of RMB for a small modular fission reactor, and straggler Dongxi Fusion (东曦聚变) raised hundreds of millions for a fusion approach that uses high-powered lasers to compress fuel pellets until they ignite. In BCI, Shenfu Jianxing (神复健行) raised ¥300 million (~$42M) at ¥1.8 billion (~$250M), the largest angel round in Chinese BCI history.

🚀 Aerospace & Defense (8 deals)

Eight aerospace deals, led by Expace (科工火箭), the commercial launch arm of state defense conglomerate CASIC, which raised ¥300 million (~$42M) in a C round. Tianmu Turbine (天目涡轮) raised a near-¥100 million (~$14M) angel for aviation engines, an unusually large angel for propulsion hardware.

⚙️ Other Adv. Manufacturing (17 deals)

Seventeen deals in optics, 3D printing, precision equipment, and factory automation. The headline was Goertek Optical (歌尔科技), which took ¥1 billion (~$139M) in strategic capital from parent Goertek and Sunny Optical at a ¥8.79 billion (~$1.22B) valuation, positioning for the AR/VR optics cycle.

Spotlight

Shenfu Jianxing (神复健行)

Angel, ¥300M+ (~$42M), ¥1.8B valuation (~$250M)

Implantable brain-spinal interfaces that help paralyzed patients walk again.

Where capital flows is where talent concentrates, where production capacity gets built, and where the next commercial technologies will come from.

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.

This was one of the busiest weeks of the year, with 132 deals in seven days. Two humanoid robot unicorns, a $200 million 3D AI round, and a wave of quantum deals backed by something closer to industrial policy than venture capital.

Where the Money Went

132 deals crossed the wire this week. About half published some indication of the amount raised, with 25 deals disclosing exact figures for a combined total of ~$1 billion (~¥7.5 billion). The real figure is substantially higher: most early-stage Chinese rounds never publish an amount, so treat that as a floor. The biggest individual checks went to humanoid robotics, but the broader story is breadth: every sector had at least eight active deals, and several had more than twenty.

🧠 AI (29 deals)

AI led the week by deal count, anchored by two outsized rounds. VAST (also known as Tripo AI) raised $200 million (~¥1.44B) for its 3D AI foundation model, while LightLink Tech (光联芯科) raised ¥500 million (~$69M) for optical chip-to-chip interconnect, a technology that replaces the copper wires between processors with laser light to keep up with AI compute demands. The rest of the sector split between AI agents, life science automation, and data infrastructure. Several deals went to companies building the training data that physical AI systems need: embodied intelligence is generating its own supply chain.

⚡ New Energy & Materials (17 deals)

Battery materials and energy storage continued their quiet grind. No single blockbuster, but steady capital into cathode and anode materials, lithium-ion cell makers, and advanced-materials plays. The largest disclosed amounts were modest: ¥140 million (~$19M) for a flexible photovoltaic company and ¥88.2 million (~$12M) for a lithium iron phosphate subsidiary. The sector’s capital arrives incrementally, round after round, without the headline figures that robotics and AI command.

🔬 Semiconductors (25 deals)

The second-busiest sector and the most strategically dense. A subsidiary of chip-design house Montage Technology (澜起科技) raised ¥500 million (~$69M) for switch chips that control how processors and memory share data inside an AI data center, the routing layer that determines whether thousands of chips work together efficiently or spend most of their time waiting on each other. Supercore (超摩科技) raised ¥305 million (~$42M) for processors built from smaller, stackable tiles rather than one large chip, a design that lets manufacturers match high-end performance using fabrication processes they have access to domestically. LEDA Tech (立芯软件) took ¥300 million (~$42M) for EDA tools, the software engineers use to design chips before fabrication. And at the other end of the spectrum, Sibai Chuang (偲百创) was acquired at a distressed valuation by listed RF chipmaker Wisecore for vertical integration. The stage distribution skewed early, with several angel and Pre-A rounds targeting advanced chip packaging, processors that compute directly inside memory chips, and display semiconductors.

🤖 Robotics (25 deals)

The biggest individual checks of the week. Spirit AI (千寻智能) raised ¥1.5 billion (~$208M) at a ¥18 billion (~$2.5B) valuation, its sixth round in twelve months. Astribot (星尘智能) closed a ¥1 billion (~$139M) Series B at ¥12 billion (~$1.67B). Below those flagships, capital concentrated in components and data rather than finished machines: dexterous hands, tactile sensors, force-torque sensors, tendon-drive mechanisms, and the training data that teaches robots to manipulate real objects all drew more rounds than complete humanoid systems.

💡 Frontier Tech (8 deals)

Eight deals, three of them quantum. HyQubit (华翊量子) raised its A+ from a 10-investor national syndicate for quantum computers that trap individual atoms with lasers to perform calculations. FieldQuantum (量坤科技), a five-month-old quantum-AI middleware company with no product, raised hundreds of millions at a ¥1.8 billion (~$250M) valuation. GGQuanta (国光量子) completed its third round in three months from industrial and strategic listed companies. The non-quantum deals included two brain-computer interface startups and a simulator for modeling nuclear fusion reactions.

🚀 Aerospace & Defense (10 deals)

Satellite communications, eVTOL (electric vertical-takeoff aircraft, essentially flying taxis), and space. Interstellor (穿越者) raised ¥100 million (~$14M) for crewed spacecraft that fly to the edge of space and return without entering orbit, a market aimed at space tourism and high-altitude research. SparkSpace (星火空间) took a Pre-A for liquid rockets that use electric motors to feed propellant rather than traditional turbopumps, a simpler design suited to small satellite launches. The rest was early-stage: satellite-based Earth observation, rocket propellant tanks, and UAVs. Disclosed aerospace capital was modest, about $28 million (~¥200M), but the deals are concentrated in hardware categories that take years and billions to reach scale.

⚙️ Other Adv. Manufacturing (18 deals)

The catch-all sector, and this week mostly components. Kunwei Technology (坤维科技) raised ¥100 million (~$14M) for six-axis force-torque sensors, the kind of touch-sensitive joint that lets a robot arm know exactly how hard it is pressing and adjust in real time. Servo systems, fiber-optic temperature sensors, and industrial automation filled out the list. Stage distribution was scattered: from angel rounds for polishing robots to a Series D for an optical sensing company.

Spotlight

Spirit AI (千寻智能)

A+, ¥1.5 billion (~$208M), ¥18 billion (~$2.5B) valuation

The humanoid robot company that beat global benchmarks and raised nearly ¥5 billion in three months.

It was here that CATL had built a lab to burn, freeze, overload and electrocute batteries, then document exactly how they die.

A Surge in Demand

The world adds more than a Japan’s worth of electricity demand every year. Global use is rising by about 1,100 TWh annually, up from ~700 a year over the previous decade, the fastest sustained growth in more than 10 years. The cause is the electrification of almost everything: industry, air conditioning, electric vehicles and computing.

One of the fastest-growing drivers is the AI data center.

A decade ago, a large data center drew about 50 MW. Today a single AI campus is designed for 10-20x that, and the biggest are planned for 1 GW or more, the output of a nuclear reactor, on one site. A single rack of Nvidia’s latest AI chips draws up to 140 kW, against 10-15 kW for a rack of conventional servers. A 500 MW facility running flat out uses about as much electricity in a year as 360,000 American homes, roughly the number of households in San Francisco.

OpenRouter, a marketplace that routes requests across most of the major AI models, now processes more than a quadrillion tokens a year, the units of text and data a model reads and generates, roughly 15x the volume of a year earlier. Goldman Sachs expects token consumption to rise another 24x by 2030.

The obvious counterargument is efficiency: each new generation of chips and models does more per watt. But usage is growing far faster than efficiency, so total power demand climbs rather than falls. The International Energy Agency expects data centers worldwide to roughly double their electricity use by 2030, to about 945 TWh, growing four times faster than demand from everything else.

Even so, data centers at that scale would still account for less than 3% of the world’s electricity. The total energy is not the problem. The problem is concentration: hundreds of megawatts of demand arriving at a single point on the grid, with load profiles that swing by the second, faster than any grid was designed for.

That power still has to reach the chips.

Power Struggle

Between the data center and the power it needs sit three problems, each distinct, each getting worse.

1. Power delivery is years behind demand. In the United States alone, more than 2,600 GW of generation and storage capacity sits in the interconnection queue, waiting for approval to connect to the grid. The median wait to commercial operation is about five years. Google has reported potential delays of up to 12 years. Building a data center takes 12 to 24 months. Connecting it to the grid can take three times longer. In Ireland, EirGrid restricted new data-center connections in Dublin because the transmission network could not absorb them. The queue is not just long: more than 90% of interconnection applications arrive with technical deficiencies needing multiple rounds of revision before a grid operator will even begin a study.

2. The demand itself is a new kind of problem. A conventional data center drew steady, predictable power. An AI training facility does not. Training runs ramp consumption up and down by megawatts in seconds as jobs start, checkpoint, and restart. Gas turbines, the primary source of flexible generation, take minutes to ramp. The grid was engineered for loads that change slowly and predictably. The load profile of an AI data center is unlike anything it was built to serve. It is also more expensive to run: electricity pricing penalizes peaks, and a facility whose consumption swings unpredictably pays for the spikes, not the average.

3. The cost of the mismatch does not stay inside the facility. US electricity prices rose about 13% in 2025. In Georgia, residential bills climbed roughly sixfold in two years to about $175 a month while Georgia Power proposed $15 billion in expansion driven largely by data-center demand. More than 20 data-center projects across the country have been delayed or canceled amid local opposition. In March 2026, AWS, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed the White House Ratepayer Protection Pledge, committing to cover generation and grid-upgrade costs rather than pass them to households.

Batteries Lead the Charge

Batteries do not generate electricity. They store it, release it, and do both faster than any other asset on the grid. That combination solves several of the problems above at once.

1. Time-to-power. A battery system can be installed in months. Paired with on-site generation, it lets a facility begin operating while the grid connection is still years away. It also raises effective capacity beyond what the connection alone can deliver: if the grid provides 50 MW but the facility needs 200 MW at peak, the battery charges when load is light and discharges when it exceeds the connection’s limit. At the extreme, a facility goes fully off-grid with solar and batteries, bypassing the queue entirely.

2. Shock absorption. When a training run spikes or drops demand by tens of megawatts in seconds, batteries match it instantly. Turbines take minutes. The rest of the power system sees a smoother load instead of raw spikes. That smoothing saves money: batteries store cheap off-peak energy and discharge during spikes, reducing peak charges.

3. Grid services. A facility with co-located storage can provide frequency regulation and voltage support back to the grid, turning it from a burden into a resource.

4. Protecting the compute. A power cut mid-training means a lost checkpoint, potentially days of wasted work. Batteries bridge outages in milliseconds, far faster than a turbine can start, and deliver cleaner power through solid-state inverters than rotating generators produce. For racks of GPUs sensitive to voltage fluctuations, that matters.

Batteries have a hard constraint. They store hours of energy, not days. They cannot sustain a facility through a prolonged outage or replace the grid as a primary power source. Gas, grid, and eventually nuclear provide that sustained generation. But batteries are the fast-responding layer that makes the rest of the stack viable.

The Trust Problem

If batteries can do all of this, the obvious question is why the storage market is not larger. The answer is not technology, cost, or capacity.

Storage batteries and EV batteries began developing in parallel around 2009. Today the storage market is only about a fifth the size of the power-battery business. The bottleneck is proving they work. Vendors claim cycle lives and efficiencies that nobody can independently verify. Some equipment underperforms in the field. Buyers cannot tell good from bad, so they price everything as if it might be bad.

That shows up directly in what people will pay. Tesla’s Megapack sells at several times the per-Wh price of comparable Chinese storage products overseas. Tesla has five years of deployed operating data, a track record asset managers can underwrite. Chinese manufacturers, many of them technically competitive, do not. Someone has to close that trust gap.

Back to the Lab

That burned-out shipping container in Xiamen was not a failure. It was a test. The hall I walked through is one of five laboratories in what CATL and the Xiamen municipal government have jointly built as the world’s largest energy storage validation platform: the Empirical Energy Storage Technology Research Institute (实证储能科技研究院). Ten hectares, roughly RMB 3 billion, and built from the start as industry-shared infrastructure (行业共享的基础设施), open to any manufacturer willing to put its equipment through the process.

Chen Xiaobo (陈小波), the institute’s dean, describes it as a physical exam center plus hospital (体检中心加医院): install a prototype, run the full test suite, identify failures, fix them on site, iterate. Another expert called it a wind tunnel (风洞).

The five labs test a storage system the way the real world would break it:

Grid connection. A 35kV/100MVA grid simulator that tests more than 10 large containers simultaneously, simulating a 1,000-node grid topology across 15 to 60 Hz to cover domestic and international standards. A system destined for Fujian’s grid gets tested against Fujian’s actual grid behavior. The lab serves a joint State Grid (国网) / China Southern Power Grid (南网) storage innovation consortium.

High-voltage safety. Testing from 1kV to 500kV, tracing the root mechanisms of ignition and explosion under extreme voltage. The engineering goal: equipment that does not catch fire and does not explode (不起火、不爆炸).

Thermal safety and combustion. The hall from the opening scene. A 20MW calorimeter, the first of its kind, inside 100,000 cubic meters of indoor burn space. Nine large storage containers can be deflagration-tested simultaneously. This is where batteries are documented failing under thermal stress.

Environmental adaptability. Five chambers (climate, salt spray, rain, dust, environment) covering minus 50°C to 100°C, altitudes up to 7,200 meters, and full-grade dust and water protection. A container bound for the Gobi Desert or a Norwegian coast gets tested under the conditions it will actually face.

Electromagnetic compatibility. The only EMC lab in the world that can take a full 40-foot shipping container as a single unit. A 65-ton turntable, 5MW power supply, and an anechoic chamber running real high-power charge/discharge cycles. EMC testing at the scale the equipment actually ships at, not on a downsized sample.

Chen names three beneficiaries:

1. Equipment makers. Faster iteration, fewer field failures, deeper knowledge of their own products.

2. Long-term asset investors. Returns depend on uptime (带电率) and how quickly a station moves from grid connection to revenue. Validation shortens that timeline. “Real money they can actually earn” (实实在在可以赚到的钱).

3. The grid. Confidence that when a major event hits and storage needs to hold the grid up, it can.

The institute is already in talks with a European insurer about recognizing validation data. A second expert predicted a national grading body will eventually rate storage equipment on safety, reliability, and grid functionality. The chain: validated asset, rated asset, insurable asset, financeable asset class. Solar walked this path over the past decade. Storage is at the stage solar was before institutional capital flooded in.

CATL’s strategy: make the lab open, but the investment unrepeatable. The consortium model follows GB 38031, the power-battery safety standard that became a national standard through industry collaboration. Chen insists the intent is “not to block others” (不卡别人). Another expert: “it will be hard for anyone to invest in a facility this complete again.” Both are true. To reinforce the platform’s independence, CATL brought in TÜV Süd, TÜV Rheinland, CSA, and CGC (鉴衡) as co-supervisors. Storage is ~20% of CATL’s business. The lab’s value is raising the bar for the entire industry.

The Money Is Moving

Capital is already flowing. Nextpower acquired Prevalon Energy for up to $365 million in May 2026, buying into BESS and AI data-center infrastructure. Prevalon, spun out of Mitsubishi Power in 2024, has over 6 GWh deployed globally with firm contracts supporting hyperscaler data centers.

In Oregon, Aligned Data Centers and Calibrant Energy deployed a 31 MW/62 MWh battery system at a Pacific Northwest campus, the first US project where a battery was purpose-built to accelerate interconnection and bring a data center online years earlier. Operational in 2026.

LG Energy Solution signed a $1.6 billion deal with Michigan utility DTE Energy for 1.5 GW/6 GWh of battery storage across eight projects, using LFP cells manufactured in the US and Canada. LG shares surged 15% on the announcement.

In Croatia, the Pantheon AI project backed by KKR-owned Greenvolt is building a €50 billion hyperscale data-center campus with 500 MW of solar and 8 GWh of battery storage behind the meter, designed to operate fully off-grid to bypass interconnection constraints. It is the largest private US investment in Europe.

Google acquired Intersect Power for $4.75 billion, gaining 2.2 GW of operating solar and 2.4 GWh of battery storage co-located with data centers, targeting $20 billion in renewable infrastructure by the end of the decade. In Minnesota, Google partnered with Xcel Energy to deploy a 300 MW/30 GWh iron-air battery from Form Energy, the world’s largest battery by energy capacity, capable of discharging for 100 hours. Iron-air stores energy by rusting iron and releases it by reversing the reaction, a technology that directly addresses the “hours not days” constraint of lithium-ion.

Jefferies estimates that hyperscalers represent a ~20 GW BESS opportunity through 2035, with ~9 GW through 2030. BloombergNEF raised its US storage projection to 204 GW by 2035 even while cutting solar and wind forecasts.

In China the capital is flowing even faster, though across the storage ecosystem broadly rather than into named data-center deals. Chinese venture databases tracked more than 50 energy-storage-related fundraises in the first five months of 2026 alone, spanning cells, materials, systems, inverters, testing, and AI-driven power management. Sunwoda Power (欣旺达动力), one of the companies building AI-data-center-specific storage solutions, closed a RMB 1.68 billion (~$230M) C round at a RMB 26.7 billion valuation. Haier New Energy (海尔新能源) raised RMB 1 billion at a RMB 6 billion valuation. Saudi Aramco and French PE firm Eurazeo co-led a B round valuing 碧澄能源 (Bicheng Energy) at RMB 10 billion.

One startup, 京清数电 (JingQing Digital Power), a storage-inverter maker, raised three rounds in five months. 新能先锋 (New Energy Pioneer) is being funded to build battery testing services, the same layer CATL’s Xiamen lab targets. The data-center-specific market in China is forming through CATL itself: a $600 million acquisition of Zhongheng Electric for data-center power (Zhongheng’s HVDC systems already run in facilities operated by Alibaba, Tencent, ByteDance, and Baidu), a $942 million stake making it the largest shareholder of 21Vianet (世纪互联), one of China’s biggest carrier-neutral data center operators, and a 60 GWh sodium-ion order with HyperStrong (海博思创), the largest sodium deal globally. Together the Zhongheng and 21Vianet deals total roughly $1.5 billion and put CATL on both sides of the data center power equation: supply infrastructure and demand.

What This Changes

The economist George Akerlof called it a market for lemons: when buyers cannot verify quality, they price everything toward the worst case, and good products get punished. The energy storage market has exactly this problem. Akerlof’s fixes are certification, warranties, brand, and licensing. CATL’s lab is reaching for nearly all of them.

The dean called the price gap with Tesla “enormous” (巨大) and framed the lab as infrastructure for 出海 (going overseas). A credible validation record gives Chinese integrators like Sungrow, HyperStrong, and Envision bankability they currently lack. CATL’s numbers show where value is heading: ESS gross margin 26.7% against 23.8% for EV batteries, overseas revenue at 31.4%. The premium is in the validated system, not the cell.

The lab is also a competitive move, and not the only one. In EVs, CATL expanded steadily from cells into materials, components, financing, and partnerships with automakers. It did not need to control the entire industry to gain influence over it. The data center investments suggest a similar pattern: Zhongheng extends its reach into power systems, 21Vianet connects it directly to demand, and the validation lab sets the quality standard. Integrators are diversifying cell supply as the market matures. Cell leadership alone is not enough at the system layer. If validation becomes an industry norm, the early mover that built the infrastructure and co-opted the certifiers has a structural advantage, even under an open-access model. Standards may ultimately be set by neutral bodies, and the geopolitical risk on Chinese exports remains regardless of validation quality. But the lab exists, and nobody else has built one.

The burned-out shipping container is still sitting under the test structure in Xiamen, in 38-degree heat, inside a hall the size of a small aircraft hangar. It was not a failure. It was a product that went through the worst conditions anyone could engineer, and the data from its death was recorded. Before batteries become infrastructure, they have to become trusted assets. CATL is building that trust.