China's robot work order

A joint plan from China's industry ministry (MIIT) and its state-asset regulator (SASAC), issued in early June, hands ordinary institutions a homework assignment with a deadline. Every province must nominate at least 20 real work scenarios for humanoid robots — across at least two of three domains: factory and warehouse, service, and specialized jobs like disaster relief. Every central state-owned enterprise must nominate at least 10. Implementation plans were due to the ministry by the end of June; progress reports are due by the end of November. The national target: 10,000 robots in routine operation and more than 100 high-value applications by December 2026.

Verify one, deploy a batch, drive a region.

The earlier policy line, from 2023, set production and research goals — build the robots, achieve the breakthroughs. This inverts the verb. It is the first time Beijing has pinned deployment quotas and reporting dates on named local governments and companies, turning a market it hoped would grow into one it is compelling into existence.

The robots are arguably not the point. The plan's real lever is data: it orders 'innovation application consortiums' of users, integrators and algorithm teams, and a 'verify one, deploy a batch' rollout written explicitly to harvest real-machine operating data and build training datasets. China already ships an estimated 90 percent of the world's humanoid units; routing them into mandated, observed work would generate operational data at a scale no commercial market has produced — fuel for the models that control the next generation. The catch is that, by independent estimates, only about a tenth of the humanoids built in 2025 did real-world work, and most sit at autonomy level 0 to 1: scripted or remote-piloted. Beijing is ordering the routine deployment of machines that mostly cannot yet work unsupervised — betting that the data from trying is worth more than the robots are ready.