Beijing's 2028 Robot Plan Sets 200,000 Units of Annual Capacity

On November 11, Beijing's economic and information bureau issued a 2026-2028 plan for embodied-AI robots: at least 5 world-class and 10 domestic-leading humanoid makers, 200,000 units a year of whole-machine capacity, and a 2,000-hour failure-free floor. It names no employment target.

Beijing's 2028 Robot Plan Sets 200,000 Units of Annual Capacity

On November 11, Beijing’s Municipal Bureau of Economy and Information Technology issued its Embodied Intelligence Robot Industry High-Quality Development Action Plan for 2026 to 2028. The plan names one capacity number, 200,000 whole-machine units a year, and one company target, at least 5 internationally first-class and 10 domestically leading humanoid makers by 2028. It also sets a reliability floor: more than 2,000 hours between failures and at least 4 hours of continuous running.

We could not open the bureau’s own text. The figures here come from press summaries, 21st Century Business Herald (reprinting China Business Journal) and Labor Noon News via Sina (quoting Beijing Daily and bureau officials), which agree with each other on every number below. Where one carries a figure and the other does not, we say so.

The numbers in the plan

On industry structure, the plan sets at least 5 internationally first-class and 10 domestically leading humanoid whole-machine makers, at least 200 small and specialized firms with a one-to-one IPO service channel, and 100 robotics research teams at Beijing universities and institutes. On output, Beijing’s whole-machine capacity reaches 200,000 units a year through contract manufacturing plus own lines, and the Beijing-Tianjin-Hebei region takes 25% of national whole-machine capacity with key-component localization above 60%. The summaries do not say whether the 200,000 counts humanoids only.

On the machines, the target platforms must carry a rated payload above 40% of their own weight, run more than 2,000 hours between failures, and work at least 4 hours on a charge. Labor Noon News adds a design life above 5 years. On data and compute, the plan wants more than 1M hours of high-quality data collected across at least 100 real scenarios, and a new embodied-AI compute cluster above 50,000 PFlops.

On deployment, it sets at least 20 benchmark industrial scenarios (handling, inspection and sorting, assembly, auxiliary machining), 20 service scenarios (elder care, healthcare, retail, security patrol), 10 special scenarios and 15 scalable solutions. Funding is promised but no amount is given.

Against the cohort

Start with 200,000. MIIT counted about 40,000 humanoids produced nationally in the first half and forecast more than 100,000 for the full year. One city’s 2028 capacity is double the whole country’s 2026 forecast. Capacity is not output, and the two outside counts of the first half already disagree, with Frost & Sullivan at 36,000 and IDC at 25,000. A capacity target says what a city will build, not what anyone will buy.

UBTech’s Liuzhou plant is designed for more than 10,000 a year. Beijing’s number is 20 such plants, our arithmetic. Two days ago UBTech and China Telecom’s terminal arm put 100,000 humanoid-class units on a three-year target with no customer, and Beijing’s capacity is twice that.

For a demand-side comparison, Shanghai’s October plan targets 8,000 humanoids working in factories by 2028. Beijing’s capacity is 25 times Shanghai’s in-factory target. The two plans measure different things, and that gap is the story: one city is planning how many robots it can make, the other how many it can put to work.

The reliability numbers are the more useful ones. A failure-free floor of 2,000 hours is 250 eight-hour shifts, about a working year with no breakdown. A 4-hour battery is half a shift. So the plan’s own spec says a robot on a full shift needs a swap or a recharge, and those minutes are where a human colleague stays in the loop.

The career read

The plan counts companies, scenarios, standards and compute. It does not count jobs. In the summaries we read, the only people it addresses are robotics engineers: urgently needed talent gets support for Beijing residency, schooling, housing and medical care.

The scenarios say where the exposure sits. Handling, sorting, inspection and assembly are warehouse and factory tasks. Retail service, security patrol and care assistance are front-line service tasks. The plan wants 20 benchmark scenarios in each family, and a benchmark is where a task becomes something a robot is graded on.

The plan runs through 2028. What to watch: the bureau’s full text, the first named benchmark sites, whether the 2,000-hour reliability floor shows up in a vendor’s spec sheet, and any procurement tied to the plan.

Sources

Keep reading