China orders every province and central state firm to put humanoids to work in real sites, targeting 100 use cases and 10,000-unit deployment capacity by end-2026

China's Ministry of Industry and Information Technology (MIIT) and the State-owned Assets Supervision and Administration Commission of the State Council (SASAC) have ordered provincial governments and central state-owned enterprises to move humanoid robots out of showrooms and into working factories, warehouses, hospitals and emergency sites, according to a joint notice the MIIT published on June 8, 2026.

The document, numbered Gongxin Ting Lian Ke Han [2026] No. 256 and dated June 3, launches what the two agencies call a 2026 special action on real-scene training for humanoid robots and embodied intelligence. Its stated aim is to make the routine deployment of such robots in real production and living environments normal practice and to speed up scaling. By the end of 2026 the agencies want validation and regular deployment completed in representative scenarios, more than 100 high-value application scenarios distilled, and what the notice describes as deployment capability at the ten-thousand-unit level.

The quotas are specific. Each provincial-level region must select no fewer than 20 key scenario units, and each central SOE must select no fewer than 10 scenarios tied to its own industry. The fields listed in the notice cover manufacturing, inspection and analysis, repair and maintenance, warehousing and logistics, catering and retail, medical and elderly care, workplace safety, emergency rescue and disaster prevention. Provinces may later recommend up to 10 outstanding training programmes and central SOEs up to five.

The agencies set two deadlines. Work plans were due to the MIIT and SASAC by June 30, 2026, and summaries with performance statistics are due by November 30, 2026. That compresses the whole cycle of site selection, robot procurement, task training and validation into roughly five months.

The notice organises the work around six tasks. Beyond building the training spaces, it asks user organisations and robot makers or application service providers to lead an innovation consortium for each scenario, bringing in model developers, component suppliers and research institutes. It calls for high-quality, high-fidelity datasets, for computing setups that mix cloud, edge and on-device processing including off-grid autonomy, and for safety functions it names as collision detection, force limiting, emergency braking and a black-box recorder.

The remaining tasks deal with proof and money. Agencies want test procedures and pass criteria for application validation, and they explicitly endorse exploring a humanoid-robot-as-a-service model. The notice tells local authorities to coordinate equity, debt and insurance tools to provide full-chain financial services, to push participants into standardisation committees, and to train staff who understand both the core technology and the target industry.

The pairing of the MIIT with SASAC is the notable design choice. SASAC supervises the central SOEs that run China's grids, ports, steel mills, oil fields, railways and telecom networks. By giving each of them a scenario quota, Beijing turns its largest industrial buyers into first customers for domestic humanoid makers at a time when commercial demand outside research and exhibition use is still thin.

The target sits against modest disclosed volumes. UBTech Robotics reported in its 2025 annual results, filed with the Hong Kong exchange on March 31, 2026, that it sold 1,079 full-size humanoid robots in 2025 and had capacity for more than 6,000 units at year-end. AgiBot said at its partner conference on April 17 that its ten-thousandth robot had rolled off the line this year. Unitree's listing documents show 2025 revenue of RMB1.70 billion. A national capability to deploy tens of thousands of units in working sites would therefore require the output of several leading makers combined.

The open questions are about measurement and quality. The notice speaks of capability at the ten-thousand-unit level rather than a count of robots actually installed, and it does not define what qualifies as regular deployment. Nor does it set a budget; financial support is described as coordination of equity, debt and insurance instruments rather than as a central fund. The November 30 reporting date is the first point at which the agencies will gather comparable statistics.