China's 9-ministry plan brings embodied AI into retail and logistics
China's commerce ministry and eight other departments issued 18 measures on Oct 19 to put embodied AI into retail, household services and logistics, with sandboxes but no new subsidy. MIIT reportedly put first-half humanoid output above 40,000 units and expects more than 100,000 in 2026, figures MEKAWIRE has not confirmed.
China's commerce ministry and eight other central departments have told the country's retailers, wholesalers and logistics firms to put embodied AI to work, in a policy document posted on the Ministry of Commerce (MOFCOM) website on Oct 19, 2026. The Opinions on Accelerating Technology-Driven Commerce, document Shangjianhan [2026] No. 551 dated Oct 13, set out 18 measures, and several of them name robots directly: robot companionship services, smart warehouses and logistics equipment, unmanned delivery, and embodied AI in household services and commercial logistics.
On the same day a Ministry of Industry and Information Technology (MIIT) official gave the volume behind the push. At a State Council Information Office (SCIO) briefing, Wei Wei, director general of MIIT's science and technology department, said China had more than 400 humanoid robot models on the market in the first half, output above 40,000 units, and that the industry expects more than 100,000 for the full year, according to reports by Gasgoo, Yicai and other outlets. MEKAWIRE could not open the SCIO transcript, so those figures are unconfirmed here. The question is what the policy actually puts on the table, and whether the numbers mean what they seem to.
What the nine departments signed
The document is signed by MOFCOM, the Cyberspace Administration, the National Development and Reform Commission, the education, science and industry ministries, the culture and tourism ministry, the People's Bank of China and the market regulator. Measure 1 asks for more smart goods, robots among them, under the "AI plus manufacturing" action. Measure 6 backs "healthy and well-regulated" robot and virtual-human companionship services. Measure 7 promotes smart automated warehouses, smart logistics equipment and unmanned delivery. Measure 10 is the core for our sector: it calls for embodied AI to be deployed in commerce and to expand household-service and commercial-logistics scenarios. Measure 16 asks for faster transfer of embodied AI and brain-computer interface research to the market, and measure 17 for new occupations such as logistics robot application technician.
What the text does not do matters as much. Measure 18 funds the plan through "existing fund channels" and asks banks to lend on market terms. There is no new subsidy pool, no procurement target and no unit number. Measure 15 instead allows stress tests and sandbox pilots for new technology "within a controllable range". The policy offers demand and room to experiment, not cash.
Reading the 100,000 figure
If the reported numbers are right, the second half has to deliver at least 60,000 units, 1.5 times the first half, for the full year to pass 100,000. That is a 2.5 times multiple of first-half output.
Set against the only industry-wide count measured on one rule, the gap is large. The International Federation of Robotics counted about 7,000 full-size humanoids, taller than 140 cm, sold worldwide in all of 2025, as our [shipments explainer](/news/humanoid-robot-shipments-2026-explained-7000-full-size-units-sold-china-ships-thousands-us-counts-hundreds-20261017) records. MIIT's first-half figure is 5.7 times that. UBTech, a listed maker that reports full-size sales, sold 921 such units in the first half, 2.3% of MIIT's output figure. Unitree shipped more than 5,500 humanoids in 2025, per its filing. The arithmetic says MIIT's count is dominated by smaller, cheaper machines for education, research and display, not the full-size workers that dominate headlines.
- Humanoid output, China (MIIT)
- H1 2026
- More than 40,000 units
- SCIO briefing Oct 19, outlet-reported, unconfirmed
- Humanoid output expected, China (MIIT)
- Full year 2026
- More than 100,000 units
- SCIO briefing Oct 19, outlet-reported, unconfirmed
- Full-size humanoids sold worldwide (IFR, taller than 140 cm)
- 2025
- About 7,000
- IFR release Sept 30, 2026
- Unitree humanoid shipments
- 2025
- More than 5,500
- Company filing
- UBTech full-size humanoids sold
- H1 2026
- 921
- Company filing
MIIT counts all humanoid products; IFR counts only full-size units. The two are not comparable one to one.
As of Oct 20, 2026
The state apparatus around the push
The commerce opinion lands in a week of Chinese robot policy. China's first six national humanoid standards, recommended rather than mandatory and effective April 1, 2027, were [published on Sept 28](/news/china-first-humanoid-robot-national-standards-gb-t-48317-unitree-ubtech-xiaomi-drafted-all-six-20261019). The National Bureau of Statistics said on Oct 19 that [industrial robot output reached 794,781 sets](/news/china-industrial-robot-output-794781-sets-jan-sep-2026-up-29-percent-above-all-of-2025-20261019) from January to September while service robot output fell 17.8% year on year. According to the outlet reports, Wei also said MIIT will issue a 2026 national humanoid standards system guide, set up industry monitoring and run product, network and data security testing.
MEKAWIRE analysis
Beijing is now treating shops, warehouses and homes, not factories, as the demand outlet for the humanoids its makers are already building, and it is paying with access rather than money.
The evidence is in the text: commerce scenarios, sandboxes and new job titles, but "existing channels" for funding. The service robot output decline in the NBS data shows why a demand signal is needed. The strongest counter-argument is that access is worth more than subsidies in China: if state-linked retailers and logistics groups open sites, a maker can collect deployment data quickly, as AgiBot is doing with its [university placement of more than 1,000 robots](/news/agibot-1000-robots-20-universities-rmb200m-yuansheng-aima-talent-programme-20261016).
Bull case: retail and logistics pilots convert into repeat orders and give Chinese makers the deployment data their foreign rivals lack. Bear case: the volume is mostly small display units, the pilots stay pilots, and the 100,000 figure says more about capacity than demand. Signals to watch: the SCIO transcript and any MIIT release with a definition of "output"; the 2026 standards system guide; the first named retail or logistics sandbox pilots under measure 15; and China's full-year output figure in early 2027. The open question is how many of those 100,000 units will be bought by a customer rather than shipped to a showroom.
- H2 needs at least 60,000 units, 1.5x H1
- Existing fund channels and market-based bank lending; no new subsidy, no unit target
- MOFCOM and 8 other central departments, MIIT among them
- Measure 15 allows stress tests and sandbox pilots within a controllable range
- Opinions on Accelerating Technology-Driven Commerce, Shangjianhan [2026] No. 551, dated Oct 13, posted Oct 19
- 18; robots named in measures 1, 6, 7, 10, 16 and 17
- Measure 10: deploy in commerce, expand household-service and commercial-logistics scenarios
- H1 2026 humanoid output above 40,000, 400+ models; 100,000+ expected for 2026
Why it matters
The opinion turns commerce into an official deployment channel for embodied AI. For Chinese humanoid and service-robot makers, a nine-department document signed by MOFCOM, MIIT, NDRC and the central bank gives local commerce bureaus and state-linked retailers political cover to open stores, warehouses and delivery routes to pilots. Combined with the recommended national humanoid standards effective April 2027 and the sandbox clause, it lowers the cost of getting a robot into a real customer site, which is the scarce input for every humanoid programme.
Rival analysis
The direct beneficiaries are the makers with service and logistics products already in volume: Unitree (more than 5,500 humanoids shipped in 2025), UBTech (921 full-size units sold in H1 2026, plus logistics and education lines) and AgiBot (more than 1,000 robots being placed in universities). Foreign rivals such as Figure AI (more than 350 robots delivered by April) and Agility have no comparable access to Chinese retail sites. In Korea and Japan, public-sector demand is being prepared through safety rules and procurement rather than through commerce scenarios.
Valuation context
No financial terms are attached to the opinion, so it changes demand expectations rather than cash flows. MIIT's reported 100,000-unit expectation, if it holds, would be about 14 times IFR's 2025 global full-size sales count, but the definitions differ and most of the Chinese volume appears to be smaller units. Investors pricing Chinese humanoid makers on unit growth should separate full-size from small-format volume; UBTech's 921 full-size units equal 2.3% of MIIT's H1 output figure.
Supply-chain implications
Retail and logistics deployments favour cheaper, smaller and wheeled form factors, which shifts component demand toward low-cost actuators, cameras and edge compute rather than high-torque full-size joints. Chinese suppliers already dominate those lower tiers. For Korean, Japanese and German component makers the policy is a volume signal for China but not an opening: the document stresses safe and reliable open-source operating systems and domestic supply.
Signals to watch
The SCIO transcript or an MIIT release that defines 'output' and confirms the 40,000 and 100,000 figures. Publication of the 2026 national humanoid standards system guide. The first named commerce sandbox pilots under measure 15, by city and retailer. The full-year 2026 humanoid output figure in early 2027.
Analyst view
Thesis: Beijing is buying demand for humanoids with access, not money, and the access is in commerce rather than manufacturing. Confidence: medium. The policy text is primary and clear; the volume figures are outlet-reported and their definition is unknown. A confirmed MIIT definition showing mostly full-size units would raise the importance of the 100,000 figure; evidence that pilots remain showroom deployments would lower it.
Questions you should be asking
How does MIIT define humanoid output, and how many of the 40,000 units are full-size? Which retailers and logistics groups will open sites first, and on what commercial terms? Will local governments add subsidies on top of the 'existing channels'? How will the sandbox clause interact with product safety testing under the new standards?