China expects embodied AI robot output to exceed 100,000 units in 2026, MIIT says
China's industry ministry said on July 20, 2026 that more than 230 companies have released over 400 embodied AI robot models and that full-year output of complete machines is expected to exceed 100,000 units. Official data show industrial robot output rose 28% in the first half and 29.0% to 729,352 sets in the first eight months, as Beijing pushes humanoids into real factories through a joint programme with the state asset regulator.
China expects output of complete embodied AI robots to exceed 100,000 units in 2026, the Ministry of Industry and Information Technology said on July 20, giving an official full-year volume marker for a sector that Beijing treats as a strategic industry. Wang Weiming, the ministry's chief engineer, told a State Council Information Office press conference on first-half industrial performance that more than 230 companies have released over 400 complete embodied AI products and that products are iterating quickly.
Wang also said China now accounts for more than 400 humanoid robot models, which he described as over half of the global total, according to the ministry's transcript. The ministry did not break the 100,000 figure down by humanoid, quadruped or wheeled form, and it did not say how the number would be counted.
The statement came alongside broader data showing robotics outgrowing the rest of manufacturing. Industrial robot output rose 28% and service robot output 11.9% year on year in the first half, the ministry said, while value added of industrial enterprises above designated size grew 5.4% and high-tech manufacturing 13.3%. The ministry added that more than 30% of large manufacturers now use AI technology.
The numbers since July
Later statistics confirm the pace. The National Bureau of Statistics reported on September 15 that China produced 96,174 sets of industrial robots in August, up 34.6% year on year, and 729,352 sets in the first eight months, up 29.0%. Service robot output was 1,656,132 sets in August, up 5.0%, and 13,931,863 sets in January to August, up 12.1%. Industrial robot output therefore averaged about 91,169 sets a month this year (729,352 / 8).
The policy behind it
The volume target sits on top of a deployment programme launched in June. On June 8 the ministry and the State-owned Assets Supervision and Administration Commission issued a joint notice, numbered 工信厅联科函〔2026〕256号, for a 2026 special action on real-world training of humanoid robots and embodied AI. It aims to develop more than 100 high-value application scenarios and build the capacity to deploy robots at the scale of tens of thousands of units. Each province must select at least 20 key scenario sites covering at least two of the industrial, service and special-purpose sectors, and central state-owned enterprises at least 10, with verification and deployment to be completed by the end of 2026.
The notice lists the places where robots are meant to learn real work: manufacturing, inspection, maintenance, logistics, retail, healthcare and emergency response. It asks users, manufacturers, component suppliers and research institutes to form application consortia and to develop packages of operational skills and embodied AI models, with attention to standards and talent.
How the companies compare
Company disclosures show how far the industry is from 100,000 units on its own. UBTech Robotics, the Hong Kong-listed maker of Walker humanoids, sold 921 full-size humanoids in the first half of 2026, according to its interim results. That equals about 0.92% of the ministry's full-year expectation (921 / 100,000). Unitree Robotics, which listed in Shanghai in August, reported first-half revenue of RMB1,152.25 million but did not give unit numbers in its listing announcement, and XPeng said its IRON humanoid will enter mass production only at the end of 2026.
The gap suggests that the 100,000 figure includes many smaller and lower-priced machines, such as research platforms, quadrupeds and wheeled service robots, as well as units built for data collection and training sites. That is a reasonable reading of the ministry's wording, which refers to complete embodied AI products rather than humanoids alone, but it remains an inference until the ministry publishes a definition.
The technology behind the push is simple to state and hard to deliver. Embodied AI means robots that learn tasks from data gathered in the physical world rather than following fixed programs. The training action is designed to produce that data at scale by placing robots in real factories and state-owned sites, so that models improve and costs fall as volumes rise.
The risks are classic for Chinese industrial policy. A volume target can encourage output that outruns paying demand, and dozens of companies chasing the same scenarios can trigger price wars that erode margins. Local governments selecting scenario sites may favour visibility over productivity.
ROBOTNESS analysis
Beijing's 100,000-unit expectation is less a forecast of commercial demand than a statement that the state will create the first wave of demand itself.
The evidence is in the design of the programme: a quota of scenario sites for every province and central state-owned enterprises, a target of more than 100 scenarios, and a deadline at the end of 2026. Combined with industrial robot output growing 29.0% in eight months, the state is building both supply and the sites to absorb it.
The strongest counter-argument is that disclosed company volumes are far below the target. UBTech, the most transparent listed seller, sold 921 full-size humanoids in six months, under 1% of the national figure.
Bull case: state sites generate real task data, embodied AI models improve quickly, and costs fall enough that private factories start buying in 2027. China would then own the largest real-world training base for robots anywhere.
Bear case: many units end up in showrooms and training halls rather than productive work, price competition destroys margins, and investors who priced listings such as Unitree on growth are disappointed when demand outside state programmes proves thin.
- October 2026: NBS September and third-quarter industrial output data, the next official read on robot volumes.
- December 31, 2026: deadline for completing verification and deployment under the MIIT and SASAC training action.
- January 2027: full-year 2026 statistics and the ministry's annual review, which should show whether embodied AI output passed 100,000 units.
- Industrial robots
- August 2026
- 96,174
- 34.6
- Industrial robots
- January to August 2026
- 729,352
- 29
- Industrial robots
- January to June 2026
- No data
- 28
- Service robots
- August 2026
- 1,656,132
- 5
- Service robots
- January to August 2026
- 13,931,863
- 12.1
- Service robots
- January to June 2026
- No data
- 11.9
- Complete embodied AI products (MIIT expectation)
- Full year 2026
- 100,000
- No data
- UBTech full-size humanoids sold (company)
- January to June 2026
- 921
- 1946.7
Monthly and January to August output from the National Bureau of Statistics release of September 15, 2026. First-half growth rates and the over-100,000 full-year expectation from MIIT at the July 20, 2026 SCIO press conference (no unit totals given for H1). UBTech from its interim results announcement of August 28, 2026. Figures disclosed, not estimated.
As of Oct 1, 2026
- Wang Weiming, MIIT chief engineer
- July 20, 2026, State Council Information Office press conference
- More than 100 scenarios, completion by end of 2026
- More than 230 companies, over 400 complete products
- Complete embodied AI products to exceed 100,000 units in 2026
- Up 11.9% year on year
- 96,174 sets, up 34.6% (NBS)
- Up 28% year on year
- 729,352 sets, up 29.0% (NBS)
Why it matters
A ministry-level volume expectation changes behaviour across the supply chain. Component makers plan capacity, local governments design subsidies and investors price listings against it. The 100,000 figure will be used in prospectuses and pitch decks for the rest of the year, whether or not it is met.
It also signals the shape of Chinese demand. Through the MIIT and SASAC action, state-owned enterprises and provinces are being told to create sites for robots, which means the earliest large orders in China are likely to come from public and quasi-public buyers rather than private factories.
Rival analysis
Among companies that disclose numbers, UBTech leads in full-size humanoids with 921 units sold in the first half, and Unitree leads in profit with RMB274.00 million attributable net profit in the same period. XPeng is building a production line but will only start mass production at the end of 2026.
Outside China, US developers such as Figure AI, Agility Robotics and Apptronik announce pilots and funding but not units, which makes China's official statistics the only national series on robot output growth at this scale.
Valuation context
The policy underpins the valuations now being set in China's capital markets. Unitree listed at 219.23 times 2025 earnings against an industry average of 38.56 times, according to its listing announcement, a premium that assumes sustained growth in demand of the kind the ministry is projecting.
If the 100,000 figure is met mainly with low-priced research and wheeled units, revenue per unit will be far below the roughly RMB641,000 UBTech achieved per full-size humanoid in the first half (590.3 million / 921). Investors should separate unit targets from revenue potential.
Supply-chain implications
Industrial robot output of 729,352 sets in eight months already creates large demand for servo motors, reducers and controllers, and embodied AI machines add demand for joint modules, dexterous hands, cameras and batteries. The training action explicitly calls for consortia that include component suppliers, pulling them into deployment rather than leaving them as anonymous vendors.
The risk for suppliers is overcapacity. If dozens of integrators expand for the same state-sponsored scenarios, component prices could fall faster than volumes rise, repeating patterns seen in solar and batteries.
Signals to watch
Monthly NBS industrial output releases around the middle of each month give the most frequent official read on robot production. A slowdown in industrial robot growth below 20% would suggest the capex cycle is turning.
The second signal is publication of the selected scenario sites and consortia under the training action, and the third is any ministry definition of what counts toward the 100,000 complete-machine figure.
Analyst view
Thesis: China will likely report embodied AI output above 100,000 units in 2026 because the state is organising the demand, but the figure will say little about commercial adoption. Confidence: medium. The ministry statement and statistics are official and clear; the definition of the target and the mix of machine types are not published.
The more useful metric for investors is revenue from paid deployments outside state programmes. That number is not reported nationally, so company filings such as UBTech's interim results and Unitree's quarterly reports remain the best evidence.
Questions you should be asking
Which machine types count toward the 100,000 complete embodied AI products, and will the ministry publish a breakdown?
How many of the units deployed under the training action are bought by the sites rather than lent by manufacturers?
Will the selected scenarios be published with performance data that allow comparison of robots against human labour?