For two years the Chinese embodied intelligence pitch has been simple: the factory line is expensive, people are expensive, and a machine that replaces both pays for itself. On 23 October, the State Council General Office narrowed that pitch.
The document is titled the Opinion on Adapting to the Development of Artificial Intelligence Technology and Helping Build an Employment-Friendly Development Model, issued as Guobanfa 2026 No. 29. It was signed on 16 October and published on 23 October.
What the document actually says
One sentence carries most of the commercial weight: deployment should be prioritised in posts where labour is in short supply and conditions are hazardous.
The document also introduces a concept that has not previously appeared in Chinese policy at this level, the unemployment buffer period. Before AI is rolled out at scale, workers are to be informed in advance and given support for retraining and redeployment. For labour-intensive industries, the scope, intensity and pace of application are to be grasped steadily and prudently.
The framing is not anti-technology. It is a sequencing instruction. Dirty, heavy, dangerous and understaffed work is where robots are wanted. Work that consists of stacking people into low-value tasks is where robots are told to slow down.
The valuation logic that just broke
Much of the first-tier funding in Chinese general-purpose robotics was raised on a replacement story: dense labour, low automation, enormous addressable headcount. That story now carries a social and compliance cost that it did not carry a week ago.
Companies selling pure headcount substitution will find procurement harder, not because the technology failed but because the buyer now has to justify the deployment internally and publicly. In a market where local governments are often the anchor customer, that justification matters as much as the payback period.
Where the money is being pointed instead
The document effectively publishes a map of approved scenarios. High-altitude power line inspection. Nuclear plant maintenance. Hull derusting on ocean-going vessels. Petrochemical anti-corrosion. Night-shift care work in nursing homes. All of them share three properties: hard to recruit for, hard to retain staff in, and physically damaging over years.
Some Chinese companies are already positioned there. Shineco’s high-altitude robots hold what the company describes as 70 to 80 per cent of the Chinese market across ship derusting, petrochemical anti-corrosion and energy inspection. Unitree has pushed firefighting and grid inspection, and China Southern Power Grid is a shareholder. Youibot builds explosion-proof units for hazardous sites.
The component layer is insulated from this entirely. Whether a humanoid ends up in a nuclear plant or a nursing home, it still needs high-precision reducers, dexterous hands and force sensors. Supply chain exposure is the safer way to own the theme.
The part the document does not solve
Hazardous environments are, by definition, hard to instrument. A robot that works in a lab on a clean floor has to survive heat, dust, radiation, salt and vibration. The data required to make it reliable is expensive to collect and cannot be scraped from the internet.
That is why the policy favours incumbents. A company with ten years of field data from ship hulls or substations has something a well-funded generalist cannot buy, and the directive makes that field data the qualifying asset.
The Chinese robotics industry has spent two years arguing about how many humanoids the world will need. Beijing has now answered a narrower and more useful question: which ones it will pay for.
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/2026-10/ART-898890-8480-30708853.html.