China’s humanoid robot makers generate most revenue by selling machines to government‑backed training centres that collect and sell training data back to the makers, raising doubts about real demand in a sector Beijing promotes. The model, mirroring Nvidia’s circular financing, has spurred valuations of startups like Agi Bot and Hong Kong‑listed UBTech, but investors are questioning whether policy‑driven purchases translate into commercial traction.

The strategy encourages local governments to build large‑scale training facilities where humans “teach” robots via teleoperation. Nearly 370 startups have emerged in two years, and Unitree surged over 600% after listing on Shanghai’s Star Market, reaching a $50bn valuation. Centres buy robots, generate data, then sell it back, cutting facility and equipment costs but blurring genuine market demand.

Only a small share of data reaches non‑robotics firms, and the model’s sustainability is uncertain. If robots cannot scale on factory floors, valuations may slide. Analysts stress that real‑world data are scarce and often unusable, limiting the loop’s effectiveness.

Government purchases have spurred industry growth, but the long‑term viability of the training‑centre model remains in question.