19,100, 22,000+ or 40,000+? Humanoid market trackers do not agree

The humanoid market is scaling quickly, but there is currently no single public number that can be presented without qualification as the definitive size of the H1 2026 market. Smart Analytics Global (SAG) estimates global shipments at about 19,100 units. Counterpoint Research now reports more than 22,000 global shipments. A China Humanoid Robotics and Embodied Intelligence Committee of 100 report presented during the World Robot Conference, meanwhile, says China alone shipped more than 40,000 humanoids and represented 97% of global volume.

Those figures cannot describe exactly the same product universe and the same point in the channel. The most useful question is therefore not simply “which number is right?” but how each tracker defines a humanoid, a shipment, an installation and an application.

Counterpoint: more than 22,000 global shipments

Counterpoint Research’s public release dated 19/20 August 2026 says global humanoid shipments exceeded 22,000 units in the first half, up nearly 300% year over year. AGIBOT leads the tracker with about 9,700 units and 43.1% share, followed by Unitree with more than 7,000 and 31.1%, Galbot at 5.0%, UBTECH at 4.4% and Leju at 2.9%. The top five vendors account for 86% of shipments. All five are Chinese companies, but 86% is not the official market share of China as a country.

The application mix matters just as much. Counterpoint’s chart assigns 33.6% to entertainment & performance, 27.0% to data production & research, 19.0% to service & guidance, 12.8% to intelligent manufacturing, 4.9% to warehousing & logistics and 2.7% to other uses. In other words, more than 60% of volume still sits in entertainment/performance and data production/research. That is a strong signal for hardware scale, but it is not evidence that 22,000 robots are doing regular productive factory work.

Unitree's filing shows what was actually generating humanoid revenue

An additional benchmark comes from Unitree's official response filed with the Shanghai Stock Exchange. This is not an H1 2026 market tracker; it is Unitree's own humanoid revenue split by application for January–September 2025. The figures therefore cannot be combined directly with Counterpoint, SAG or Committee of 100 shipment volumes. They are still useful because they show how early the application market remains.

Unitree reported RMB 595.19 million in humanoid revenue for the period. Research and education contributed RMB 438.07 million, or 73.60%; commercial and consumer applications contributed RMB 103.52 million, or 17.39%; and industry applications contributed RMB 53.60 million, or 9.01%. Research and education is a broad category in the filing: it includes not only universities and research institutes but also technology companies and individual developers buying robots for research, development and secondary integration.

The most important qualification sits inside that 9.01%. Unitree says roughly 50–70% of revenue classified as industry application was associated with enterprise tours, reception and guided-showcase use. Sales tied to clearly identified scenarios such as intelligent manufacturing, inspection and logistics delivery totalled RMB 15.70 million — 29.29% of industry-application revenue and about 2.64% of total Unitree humanoid revenue for the period. Those are real sales and real deployments, but they do not support the claim that most humanoids being sold are already performing productive factory work.

SAG: 19,100 and a similar vendor hierarchy

SAG published its own H1 2026 tracker on 9/10 August. It estimates about 19,100 global shipments, up 272% year over year. AGIBOT is at roughly 8,400 units and 44% share, while Unitree is at around 5,900 and 31%. Galbot, UBTECH and Leju follow.

The gap between 19,100 and more than 22,000 is material but not implausible in a young category where researchers may use different vendor lists, functional-humanoid definitions, shipment recognition points and rules for wheeled platforms. Notably, both datasets still show the same broad ranking and very similar share positions for AGIBOT and Unitree despite the difference in absolute volume.

Committee of 100: more than 40,000 in China

On the evening of 20 August, Global Times reported figures from the China Humanoid Robotics and Embodied Intelligence Committee of 100 report presented at WRC. According to that account, China shipped more than 40,000 humanoids in H1 and represented 97% of global shipments. If the two figures share the same denominator, the implied global market would be above 41,000 units — almost twice Counterpoint’s estimate.

There is no defensible reason to average these datasets. Until the Committee of 100 methodology can be compared in detail with the commercial trackers, its figure should be treated as a separate dataset rather than a correction to Counterpoint or SAG.

Where can such a large gap come from?

• Product scope: one tracker may count only functional humanoids meeting a specific morphology definition, while another may include a broader embodied-AI platform universe, wheeled variants or education systems.
• Shipment vs installation: a unit leaving a factory, arriving at an integrator and being installed at an end customer may fall into different reporting periods.
• Research and data production: a robot purchased by a data center or lab is a real hardware shipment, but it is not the same as a productive production-line deployment.
• Vendor coverage: China has a long tail of young local manufacturers whose output may be difficult to track consistently from outside the market.
• Unit of account: public summaries do not always reveal whether only complete systems are counted or whether additional platform configurations enter the dataset.

Shipment ≠ deployment ≠ productive work

RoboMorrow separates three stages. A shipment means hardware has left the manufacturer or entered a channel. A deployment indicates that a system has been put into operation at a specific site. Productive work requires another layer of evidence: autonomous operating time, success rate, human interventions, MTBF and measurable process value.

Shipment growth is still important. Scale can lower component costs, produce more field data and accelerate iteration. But it should not be translated directly into “jobs replaced” or “market maturity”.

What can be said with higher confidence

Across the public datasets, three conclusions are more robust than the argument over one headline number. First, humanoid hardware has entered a rapid scaling phase. Second, AGIBOT and Unitree are clearly the leading vendors in the commercial trackers. Third, a large share of current volume is still connected with entertainment, research, data production, education and pilots — activities that help the ecosystem mature but are not equivalent to stable full-shift industrial work.

Counterpoint projects more than 50,000 global shipments for full-year 2026, while SAG expects around 60,000. Both are research-firm forecasts rather than audited production plans. The key metric over the next few quarters will therefore be how much volume moves from research and demonstrations into regular industrial and service operations.

What this means for Europe

The methodology gap does not change Europe’s strategic problem: China is scaling humanoid hardware faster than Europe is scaling finished humanoid platforms. Europe still has deep strengths in automation, safety, drives, industrial integration and manufacturing. If Chinese hardware costs continue to fall with volume, European competitiveness will depend on components, software and integration that deliver measurable operational value rather than simply matching the number of humanoid brands.

Sources and verification

This article is based on Counterpoint Research — H1 2026 humanoid shipments, Smart Analytics Global — H1 2026 and Global Times — Committee of 100 report. Figures are estimates from the individual sources and are not mutually audited. RoboMorrow verification: 21 August 2026.