The financial adviser kept a full diary this spring. On her busiest day she received four government delegations, from Tianjin, Shenzhen, Chengdu and Xiamen. Most were district-level investment promotion officers. A few ran whole prefecture-level cities.
Some of them said it plainly. They had no thesis of their own, so they followed whatever the policy documents said. Others arrived with an entire industrial map of their county: factories standing empty, suppliers already clustered, a workforce in place. All that was missing was a flagship company to pull the chain together. Two thousand years after a king paid handsomely for the bones of a thoroughbred to prove he was serious about buying the horse, the same logic still prices Chinese venture rounds.
For most of the last two decades, state capital stood away from the venture market. Consumer, entertainment and internet money rose and fell without it. Hard technology broke that pattern, because market funds alone could not carry the policy objectives stacked on top of them. Embodied AI, the loudest hard-tech sector of the past two years, is where the state became most visible.
Three hundred deals, RMB 93.5 billion
IT Juzi counts more than 300 funding events in Chinese embodied AI in the first half of 2026, up 137 per cent year on year, with RMB 93.5 billion disclosed. State institutions took part in 42 per cent of the large rounds. The standard structure became a local state platform, an industrial investor and a leading VC signing the same sheet.
The national team moved in late 2025. Shanghai’s embodied AI fund and Shenzhen’s robot fund cluster deployed first and became the largest single source of new money. Beyond the first-tier cities, robot industry funds in Hangzhou, Suzhou, Hefei, Wuxi, Ningbo and Nanjing joined the same race.
For a while, investing in embodied AI felt like an open-book exam. One state fund investor described it as set-text investing: back Musk or back Marx, but stand with the direction. Several others repeated the same line in different words.
The structure of the demand explains the appeal. A city that wants a position in the 15th five-year plan for new productive forces has few options. Commercial space, quantum computing and fusion all require national facilities, elite research teams and a specialised industrial base, which only a handful of cities can host. Embodied AI splits into airframes, components, software and data, and each piece can land somewhere different.
That means a city does not need a full humanoid maker. A factory making joint motors or reducers will do. It meets the current round of recruitment targets, plants the seed for tax revenue and a cluster, and provides enough material for a plausible story about the future.
Three kinds of state money, three clocks
The first kind answers to a group’s own supply chain. These are the industrial investment platforms inside central state enterprises, and their scorecard is not the financial multiple on a single deal but whether the investment fills a gap in the parent’s operations. The CRRC upgrade fund that backed Zhiyuan Pingfang said as much, pointing to the fit between a high-end manufacturing base and embodied AI technology.
The second kind answers to local recruitment targets. District and county platforms are the most active investors in this cycle. They operate under state asset preservation rules, so they prefer later stages, valuations in the billions of renminbi, and projects close to a pre-IPO profile. One investment promotion official in a second-tier city put the yardstick directly: can the company put an R&D centre, a production line and a support team in the district, and can it pull component and data suppliers in behind it. Whether the valuation is high or the listing is near matters less.
The third kind is the market-oriented money inside the state system, where financial return is the stated objective. Shenzhen and Shanghai are piloting a classification reform that strips some state funds of their recruitment mandate and scores them on investment performance instead. Local factories stop being a precondition and return requirements loosen. Shenzhen’s investment holding group, the AI devices fund run by Shenzhen Capital Group and several Shanghai state sub-funds have backed Stardust Intelligence, Agibot and Ziyuan Robotics on those terms.
The September turn
The story changed direction in September. Unitree lost half its market value after listing. The founder of Mech-Mind publicly attacked the state of the industry. Hong Kong’s 18C route for specialist technology companies tightened. Investors began saying out loud what had been an open secret: the IPO case for the second and third tier of companies no longer holds. Several said the real funding pressure arrives after November 2026, because most companies still hold cash from earlier rounds, and the next raise will bring valuation write-downs and a closed exit window at the same time.
Unitree still ships more humanoids than almost anyone. It delivered more than 5,000 units in the first half of 2026 and its market value is still double the pre-listing mark. The collapse nonetheless became the point where the sector braked.
Circular 54 from the State Council General Office changed the paperwork as much as the mood. Signing off a base, a production line and a batch of airframes used to close a recruitment file. Now whoever signs has to explain why the money went in and whether the industry actually runs.
The dumbbell
State money is now concentrating at two ends. At the early end, funds want to write the first cheque, set the price and own the upside. At the late end, several institutions club together into pre-IPO names where the listing expectation, not the fundamentals, carries the valuation. The middle of the curve is where the risk has pooled: companies whose growth depends on custom orders, whose cash depends on the next round, and whose product has not yet produced repeat purchases.
There is a mismatch in the middle of that. Most founders still believe state money belongs at the back end, for listing endorsement and shareholder structure. Several said early state cheques come tied to capacity, tax and location commitments they do not want while the product is still unproven.
What the state will not do is underwrite the sector without limit. As one investor put it, everyone is looking upward at the policy language, at which level of government is talking about what, at which park needs a project, at which sector writes well into a report. They are not discovering the future. They are guessing it from the documents.
Panel makers, solar manufacturers and semiconductors all travelled this road and all grew up. The step that mattered in each case was weaning. Robots are now at that step.
Editor’s note: This is an adapted translation of the original 36Kr report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.sohu.com/a/1092756863_114778.