China’s Humanoid-Robot Boom Has a Deployment Problem

China’s 2026 humanoid-robot story is best understood as a manufacturing scale-up colliding with a commercialization reality check. The Ministry of Industry and Information Technology (MIIT) expects annual output to exceed 100,000 units, roughly five times the approximately 20,000 units reported for 2025. But China has also set a separate target of putting more than 10,000 humanoid robots into commercial use by the end of 2026. 671016

That difference matters. Production, shipments, paid orders, pilot projects and repeatable customer use are not interchangeable measures of market maturity.

The 100,000-unit forecast signals industrial momentum

MIIT’s projection points to rapid expansion in factories, suppliers and production lines. As of the first half of 2026, one industry account reported that output had already passed 40,000 units, while Unitree and AgiBot had each reached significant cumulative production milestones. 916

AgiBot reported producing its 15,000th embodied-intelligence robot in June. Unitree had produced approximately 11,000 bipedal humanoid robots by early June, according to reporting carried by News18. 39

Those numbers help explain why the two companies are increasingly treated as the center of gravity in China’s humanoid market. Yet they still describe what manufacturers built, not necessarily what customers received and used in routine work.

AgiBot and Unitree lead shipments, but the market is still small

Independent shipment estimates show strong growth without demonstrating mass adoption. Counterpoint Research said global humanoid shipments exceeded 22,000 units in the first half of 2026, nearly 300% higher year over year. AgiBot ranked first with 9,700 shipments, and the top five vendors represented 86% of the global total. 26

Other estimates are lower: CNBC reported that Morgan Stanley expected 50,000 humanoid robots to ship in China across all of 2026, excluding prototypes, preorder trials and internal-use units. 14 The variation is a reminder that market numbers can depend on whether a source is counting production, external shipments, or a broader category of embodied-intelligence machines.

That definitional gap is central to the story. A robot assembled in a factory is not automatically a robot deployed on a production line. A shipment is not necessarily a robot operating without intensive supervision. And a preorder is not the same as recurring demand.

Research demand remains an important part of the market

Unitree’s growth illustrates both the opportunity and the uncertainty. The company shipped more than 5,500 humanoid units in 2025, according to reporting from Gasgoo and China.org.cn, while industry reports described plans to expand production and capacity. 1227

But other reporting has cautioned that many sales so far have been one-off purchases and that relatively few robots are being used in sustained commercial operations. Reuters also reported that most of the humanoids sold in the previous year went to scientific and educational research and testing rather than commercial or industrial applications. 2022

This does not make research and education sales unimportant. They can provide revenue, hardware feedback and training data. They simply do not prove that humanoids are already delivering reliable, lower-cost labor substitution at scale.

Why production and deployment can diverge

The industry is moving from impressive demonstrations toward practical tasks such as parcel sorting, mobile-phone packing, logistics and factory work. Reuters described these efforts as part of a broader search for useful commercial applications rather than evidence that a settled mass market already exists. 11718

The obstacles are operational as much as technical:

• Reliability: Customers need robots to perform the same task repeatedly, not only during a controlled demonstration.
• Autonomy: A system that requires constant human intervention may not deliver the expected labor savings.
• Integration: Humanoids must fit existing factory, warehouse or service workflows and safety procedures.
• Service economics: Maintenance, downtime, supervision and software updates all affect the real cost of ownership.
• Task selection: A humanoid must outperform, or offer enough flexibility to justify replacing, human labor or a simpler specialized machine.

TrendForce described deployments in automotive, 3C electronics, aerospace, logistics and energy as tests of commercial viability. That framing is more cautious than treating every installation as proof of broad adoption. 3

The most useful metric is repeatable customer value

China appears to have demonstrated a powerful ability to manufacture humanoid hardware quickly and concentrate shipments among a small number of leading vendors. Chinese manufacturers accounted for more than 97% of global first-half shipments in one industry estimate, underscoring the country’s supply-chain and domestic-market advantage. 25

The harder question is what happens after the robot leaves the factory. The strongest evidence of commercialization would be multi-year customer contracts, repeated orders for the same productive task, documented uptime and output, and service costs that remain acceptable outside a showcase environment.

So the 2026 verdict is mixed but significant: China is crossing a manufacturing threshold, while deployment is still being validated. The 100,000-unit forecast shows how quickly supply can expand. The 10,000-unit commercial-use target shows that routine, economically sustainable operation remains the scarcer achievement. 67

Until production, shipments and durable customer results converge, China’s humanoid-robot boom should be read as evidence of industrial capacity—not yet as proof of mass commercial adoption.