There is a state-backed robotics testbed in Hangzhou that booked less than RMB 450,000 of revenue in the first nine months of 2026 and lost more than RMB 15 million. Seventeen companies have just paid to own a piece of it.
The National AI Application Pilot Base for Embodied Intelligence raised RMB 200 million, lifting registered capital from RMB 100 million to RMB 300 million. The investors are not venture funds. They are the companies that expect to need the base’s production lines, data and standards work over the next three years.
Who paid, and how much
The base was set up in Hangzhou in May 2025 as the country’s only national pilot platform dedicated to embodied intelligence. Its founding shareholders were Hangzhou High-Tech Sci-Tech Innovation Group, Hangzhou Data Group, Unitree Robotics and Transfar Group.
In this round, the high-tech group put in RMB 73 million, the data group RMB 39 million, Unitree RMB 20 million and Transfar RMB 8 million. The largest single cheque, RMB 60 million, came from a new holding vehicle called Jushen Wuxian that was registered only in December 2025 and holds 16 industrial shareholders. Unitree deliberately held its stake at 10 per cent.
The shareholder list is the story. Galbot brings robot bodies. MetaX and Xiwang Xinke bring GPUs. Infinigence brings compute services. Jinglianwen brings data production. Hangcha Group brings forklift deployments. Wasu and Huace bring content. Anheng Information brings security. Bodies, chips, compute, data, scenarios, content and security are all represented in one cap table.
What they are actually buying
Not equity returns. Access.
The Chinese word for what the base does is zhongshi, pilot-scale validation, and the manufacturing industry treats it as the expensive gap between a working prototype and a shippable product. Can the motion that works in simulation work on real hardware. Can one unit become a hundred identical units. Before a robot enters a factory it has to be broken repeatedly somewhere safe.
That somewhere costs money and time. Collecting full-body motion data for one humanoid needs one or two operators. One simple task needs thousands of real demonstrations. The base offers a one-to-one replica production line, mine shaft, shop and kitchen, plus a library of bodies, datasets, models and tool chains. It says it has run more than 40 application scenarios and has built a joint innovation centre with Huawei and an accelerator with Alibaba DAMO Academy.
Buying in means locking a validation slot, a licence to the scenario data and a chair in the room where industry standards get argued. RMB 60 million does not buy shares. It buys a position in the queue.
The three kinds of money
Set this against two other facts from the same month. Financial investors are withdrawing, because valuations will not close and diligence now asks only whether delivery happened and whether anyone paid. And the funding wave has not stopped: the sector raised RMB 20 billion in the first two months of the year, Galbot took a single RMB 2.5 billion round, and the third phase of the national big fund has entered directly.
Three kinds of capital, three kinds of arithmetic. Financial money runs a valuation model and leaves when it does not clear. Industrial money runs a strategic model: I do not need this company to grow tenfold, I need somewhere to validate my robot, data to train it on and a standard to certify it against before my own production line starts next year.
That is a shift from a funding race to a capacity race.
Hangzhou’s own position explains the confidence. The country’s first local regulation on embodied intelligence took effect in May. Four of every ten humanoid and quadruped robots sold worldwide are built in the city.
The base’s debt ratio is close to 90 per cent and it has almost no revenue, so the RMB 200 million is part working capital and part entry ticket. Seventeen companies sharing one shareholder register will also have to settle how much data competitors get to see and who validates first.
Editor’s note: This is an adapted translation of the original OFweek report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-11/ART-898890-12001-30710329.html.