China Says It Built 40,000 Humanoids in Six Months. Read the Number Carefully.

On 19 October, at a State Council Information Office briefing on a new nine-ministry policy package on technology-driven commerce, Wei Wei — director-general of the science and technology department at China’s Ministry of Industry and Information Technology, the ministry that sets industrial policy for Chinese manufacturing and ICT — gave official production figures for the country’s humanoid robot sector for the first time.

Three of them matter:

• Over 400 complete-machine models in the first half of 2026.
• Over 40,000 units produced in the first half.
• Over 100,000 units expected for the full year.

Set against the same period, one of these numbers does not sit comfortably with the others.

The count that conflicts with the global one

Frost & Sullivan’s mid-year global humanoid report put worldwide shipments for H1 2026 at roughly 36,000 units. MIIT now says China alone produced more than 40,000.

Both can be true, and the reason they can is the interesting part. Production is not shipments. A unit that leaves the assembly line and sits in a warehouse, a showroom, a training facility, or an integrator’s demo fleet counts in MIIT’s figure and does not count in an analyst’s shipment number. Scope differs too: “complete-machine product” (整机产品) in Chinese industrial statistics is a broader bucket than the humanoid category most market researchers track, and it readily absorbs education, entertainment, exhibition and data-collection machines that analysts strip out.

So do not add the two together, and do not treat one as a correction of the other. There are two readings, and they are very different theses:

• Chinese production is running ahead of delivered global demand — in which case the story is an inventory build, and the second half is where it shows up as discounting or as a write-down.
• The two use definitions that barely overlap — in which case the global figure is a floor, and the humanoid market is materially larger than the widely quoted number.

Anyone writing a 2027 budget off these figures should decide which reading they believe before they commit capital, because they imply opposite supply risks.

400 models, 40,000 units: the average is 100

Divide one by the other and the sector’s shape becomes obvious. Four hundred model variants across forty thousand machines is about a hundred units per model, on average — and averages in a long tail are generous to the tail.

This is not an industry that has consolidated onto a handful of platforms. It is an industry that has proliferated: hundreds of differentiated SKUs, most of them built in volumes that would not justify a dedicated production line. For a buyer, that is the single most useful thing in the announcement. A model with a three-digit installed base has a different spare-parts reality, a different software-maintenance reality and a different failure-mode database than one that has shipped a hundred times.

The full-year target sharpens the picture further. Reaching 100,000 units requires roughly 60,000 in the second half — about 1.5 times the first. That is a steep ramp concentrated in six months, and it is the part component suppliers should actually plan against: if the target is to be met, the demand signal does not arrive evenly, it arrives in H2.

The four things MIIT says it will do next

Wei set out four priorities under the framing “innovation-driven, application-led, ecosystem-coordinated”:

• Attack the technology. Key common technologies will be put out to open competition (“揭榜挂帅”) against named application scenarios, with the aim of producing solutions matched to specific tasks rather than general-purpose capability.
• Expand scenarios. The real-environment training programme continues, digging into industrial, special-purpose and public-service scenarios, and — notably — encouraging commercial enterprises to open their sites for joint validation, with a batch of benchmark applications as the output.
• Build the ecosystem. MIIT will issue the National Humanoid Robot Industry Standards System Construction Guide (2026 Edition) and develop the standards that scenarios urgently need, alongside a high-level open-source community for humanoid robotics.
• Guarantee safety. An industry monitoring system, inspection and testing covering product quality plus network and data security, and ethics governance services.

For readers outside China, the third and fourth items are the ones with teeth, and they are easy to skim past.

Why the standards guide matters more than the production number

A standards system guide (体系建设指南) is not a standard. It is the map that decides which standards get written, in what order, and by whom — and in Chinese industrial policy that map has repeatedly been the leading indicator of where compliance requirements, procurement criteria and subsidy eligibility follow. The pattern is familiar from electric vehicles, power batteries and industrial internet.

Once the 2026 edition is published, two consequences follow fairly quickly. Standards tagged as urgently needed for scenario deployment are the ones that get adopted first, and they are the ones that subsequently appear as acceptance criteria in tenders. And because the fourth priority adds inspection and testing on product quality, network and data security, the guide effectively becomes a market-access document: it defines what has to be tested, and therefore what a machine — or a component inside it — has to demonstrate before it can be sold into a regulated deployment.

Non-Chinese suppliers should read this as a compliance horizon, not as domestic Chinese housekeeping. Non-Chinese buyers should read it as a checklist to request: which edition of which standard does this machine conform to, and is there a test report from a recognised body.

What to do with it

• Ask whether volume is shipped or built. When a Chinese counterpart quotes production capacity or output, ask what portion has been delivered to paying customers. The gap between the two is the whole story of the next four quarters.
• Track the guide, not the headlines. The production figures will be revised. The standards guide will set requirements that outlast them.
• If you supply components, qualify capacity for H2. The arithmetic of the full-year target puts the ramp in the second half.
• If you buy machines, ask about installed base, not model count. Four hundred models means choice; it does not mean maturity.

Related reading

• Humanoid robot market H1 2026: 36,000 units, and the top five by revenue are all Chinese — the global figure this announcement sits awkwardly against.
• Humanoid robot spec sheet: DOF, payload and torque, decoded — how to read a model sheet when there are 400 of them.
• Hollow-cup motor winding: the wire is the part — where the H2 ramp hits first.
• Cable-driven humanoids: a tendon rewrites the BOM — what changes when the standards move before the design settles.

Sources: figures, the three “accelerations” and the four priorities are from the 19 October 2026 State Council Information Office briefing, as reported by China Economic Net, New Beijing News, China Securities Journal and China News Service. The comparison with Frost & Sullivan’s H1 2026 global shipment figure, the units-per-model arithmetic, the two readings of the discrepancy and the buyer and supplier checklists are Latest Robot News editorial.