On September 9 we covered the first reports that China’s securities regulator had informally raised the bar for humanoid IPOs, telling banks that listing candidates needed recurring revenue, narrowing losses, or real technical innovation. At the time, the mechanism was unnamed and the affected companies unconfirmed. Eleven days later, Reuters named both.
The mechanism has a name: window guidance
Reuters reports the China Securities Regulatory Commission is using window guidance, the informal, unpublished channel regulators use when they want an outcome without writing a rule that would require public disclosure. One person close to humanoid-robot investors told Reuters the freeze is effective now, even without a formal ban; another source described it as a slowdown rather than a stop. Either way, the CSRC has not commented, and the companies named have not responded to Reuters.
The trigger is the same one we flagged on September 9: Unitree’s volatile Shanghai debut, which soared more than fivefold at listing a month ago and has since fallen 55% from its peak, up two points from the 53% drawdown we reported at the time.
Who’s waiting: at least six companies, three of them named
Reuters reports at least six Chinese humanoid robotics companies are preparing IPOs. Three are named directly: Deep Robotics, X Square Robot, and AGIBOT. AGIBOT is not a marginal name in this corpus. We’ve tracked it since April, and in August it overtook Unitree on shipment share, delivering 8,400 units for a 44% global slice. A regulator slowdown that touches AGIBOT’s listing timeline is a different order of story than one touching only smaller names.
Beijing’s position, per Reuters, is a needle it is trying to thread: temper investor enthusiasm around the hottest domestic investment theme of the year, while continuing to back humanoid robotics as a stated national priority. The slowdown is aimed at the public markets specifically, not the underlying industrial policy.
The number that matters: 60% to 70%
Our September 9 coverage established that under 10% of Unitree’s 2025 revenue came from industrial applications, with training centres, often co-funded by local governments, filling much of the rest. Reuters now puts a number on how load-bearing that funding is. According to a person close to humanoid-robot investors, regulators believe some robot companies could see revenue fall 60% to 70% if revenue tied to data-collection centres were stripped out. Those centres, and the joint ventures around them, can see local governments cover 80% to 90% of the initial investment.
For scale: Morgan Stanley’s China humanoid shipment forecast, doubled in June, was built on demand numbers that don’t distinguish a factory buying units to run a production line from a local government buying units to fund a demonstration center. Reuters’ 60-to-70% figure is the first attempt by anyone official to separate those two buyers inside a specific company’s books.
That is the number worth sitting with. A revenue line that evaporates by two-thirds if you remove one funding source is not a diversified customer base. It is one customer, structured to look like several. Regulators asking prospective issuers to show “recurring revenue” are, in plain terms, asking them to prove they have a second customer.
”Campaign-style innovation”
Reuters quotes Leo Wang, a venture capitalist at Qianchuang Capital, describing the investment rush as “campaign-style innovation,” his term for the pattern of companies and capital piling into a sector because government policy favors it, not because unit economics justify the valuation. Wang’s read is that hype around embodied AI has now exceeded what China saw during its internet and new-energy investment waves. Industrial-robot makers pivoting into humanoids, and startups commanding valuations disconnected from shipped volume, are in his framing a symptom of the same pattern, not a new one.
That framing lines up with what we found in our $230M procurement count from late August: the state money in this sector has been flowing into demonstration and training infrastructure, not production lines. Reuters’ reporting adds the number that shows how concentrated that flow has become inside individual companies’ revenue statements.
What changed in eleven days, and what didn’t
The delta from September 9 is real: a named mechanism, three named companies with listings now in question, and a specific revenue-concentration number regulators are testing against. What hasn’t changed is the underlying finding from our first article: humanoid robotics in China remains, for now, a job-creating industry rather than a job-destroying one. The training-centre economy this Reuters reporting scrutinizes is itself made of jobs: teleoperation, data labeling, scene construction, deployment engineering. A regulator asking “is this revenue real” is not yet asking “is this labor real.” Those are different questions, and the second one is still years off.
For readers tracking this as a labor story rather than a market story, the practical takeaway is unchanged from September 9: watch vendor revenue mix, not shipment counts or valuations. The CSRC’s window guidance is effectively forcing that same disclosure onto the companies closest to a public listing. When a prospectus has to show what share of revenue survives without a local government as the buyer, the industry gets its first audited answer to the question this column has been asking since April.
There is a narrower reading worth flagging too. A slowdown in public listings is not a slowdown in the underlying spending. Beijing’s own framing, “national priority,” signals the procurement side of this story keeps running even while the IPO side cools. The training centres, the joint ventures, the 80%-to-90%-funded local government deals all continue regardless of whether Deep Robotics, X Square Robot, or AGIBOT list this year or next. What the CSRC has paused is the moment investors outside that state-backed loop get audited numbers to check the story against. That is a market-integrity fix, not a spending cut, and the distinction matters for anyone reading valuation headlines as a proxy for how many robots are actually doing paid work.
Sources
- China scrutinises humanoid robot IPOs as valuations soar, commercial demand questioned (Reuters via ANI News, September 21, 2026)
- China slows humanoid robot IPO rush as hype outruns reality (Reuters via Investing.com, September 20, 2026)
- Our September 9 coverage: China’s regulator raised the humanoid listing bar
