On October 19, at a State Council Information Office briefing on commerce and technology, Wei Wei, director of the technology department at China’s Ministry of Industry and Information Technology, gave three numbers. More than 400 humanoid robot models in the first half of 2026. Output above 40,000 units. A full-year forecast above 100,000. 21st Century Business Herald and Sina Tech, relaying ITHome, carried the same figures.
None of the three is new in itself. Chinese brokers and trade outlets had been repeating “40,000 produced, 100,000 for the year” since July, and on August 20 we took apart a conference report that filed the same 40,000 under shipments. What is new is the speaker. The ministry that sets industrial policy has now said the number itself, on a State Council stage, under the label output.
Which count the ministry chose
Wei said production. He did not say shipped, and he did not say working.
That matters because four counts of the same half-year are now on the table. The August conference report said China shipped 40,000-plus. A tracker published ten days earlier counted 19,100 worldwide. On September 28, Frost & Sullivan put global sales at 36,000 and production-work units at 1,200. A day later IDC counted about 25,000 shipped worldwide, 19,000-plus of them in China.
The ministry’s 40,000 sits 15,000 above IDC’s global shipment count. The definitions differ, so that gap is not inventory, and we are not claiming it is. But the stages that separate a unit leaving the line from a unit being accepted by a customer are exactly the ones a production count skips: warehousing, delivery, commissioning, sign-off.
So the October 19 figure settles who is on record. It does not settle how many of the 40,000 reached a buyer.
The back half has to run 50% faster
A 100,000 year after a 40,000 half means 60,000 or more in the second half, a pace 1.5 times the first. The forecast is not new either, but coming from the ministry it reads differently: it is a target inside policy language, not a research house extrapolating.
Against that, the other numbers we have covered look small. Shanghai’s October 13 plan wants 8,000 humanoids in factories by 2028, which is 8% of what the ministry expects China to build this year alone. The Leju and Zhaofeng cell we wrote up the same day is 15 robots on one auto-parts line, about 1.25% of Frost’s global production-work count. Output is running far ahead of the number of workstations anyone can name.
The category leader gives a second scale. Unitree’s listing filing said it had delivered about 18,000 bipedal humanoids across its entire ten-year history, as we covered on August 18. The ministry’s figure for six months is more than twice that. For both to be right, a large share of the 40,000 has to be coming from makers whose lifetime totals are smaller than Unitree’s, or has to be sitting in inventory waiting for a customer. Neither is a reading the ministry offered.
Spread 40,000 over 400 models and the average is 100 units a model. The real distribution will be steeply skewed to a handful of makers, with a long tail of prototypes. The model count is not growing news either: MIIT chief engineer Wang Weiming gave “more than 400” at another State Council briefing on July 20, when he also said Chinese models were more than half of the world’s. In three months the model count stayed put and the production figure arrived.
The half of the briefing without numbers
Wei also described where the robots are going, and that part has no figures but more bearing on work.
Products are being validated and deployed in more than 300 key scenarios, he said, naming warehousing and logistics, food service and retail, and home services. Industrial, special-purpose and “livelihood” settings are what the ministry will dig into next. A “real-scene training” campaign, run by the ministry, asks commercial and retail companies to open their stores and warehouses for joint validation, with the aim of producing a batch of benchmark applications.
None of the versions we read gives units per scenario or a year-on-year comparison. “More than 300 scenarios” could be 300 robots or 30,000.
Two more items from the same answer. A 2026 edition of the national humanoid industry standards-system guidelines has been issued, with more standards promised for application scenarios. And Wei said the ministry will build an industry monitoring system and strengthen quality, network and data-security testing.
What it means for jobs
Start with who is being asked to host the trials. Retail and logistics companies were named as the place where validation happens, which puts warehouse picking and sorting, restaurant running, shelf replenishment and floor assistance first in line to share a floor with humanoids over the next 12 to 24 months. Factory roles come later in the ministry’s own sequence. That fits Frost’s 1,200 production-work units and IDC’s finding that 69% of first-half shipments went to research, education, performance and government data centers: volume is rising, and most of it is still consumed by demonstration.
Then the monitoring system. At present the gap between 19,000 and 40,000 is mostly a gap in definitions. If the ministry defines production, delivery and deployment in one framework and publishes against it, next year’s October numbers will be comparable to this year’s. Job-risk assessment can then move from what the industry says it built to how many units are doing paid work in a given kind of site.
Two things to track. First, whether the monitoring system’s definitions are published, and whether they count delivery and deployment or only output. Second, whether any scenario category discloses an installed base. A retail or warehouse category reporting units in the dozens or hundreds would pull that category’s timeline forward.
