Unitree Robotics provides the clearest reason not to publish a single number for how many humanoid robots have been sold in 2026.
Its 2025 securities filing reports three different quantities for the same humanoid business: 5,716 robots produced, 5,511 shipped and 5,215 sold. Unitree explains that some robots had already left the company under customer orders or contracts but had not completed receipt, acceptance or other procedures required for revenue recognition by year-end. The difference was 296 robots.
That distinction matters because almost every headline number available for 2026 measures shipments, not recognized sales.
As of August 26, Humanoid Analytics could not verify a comparable global sold-unit total for January through August. The latest broadly comparable market estimates cover H1 2026, and they disagree substantially on how many humanoids were shipped.
A filing shows why shipment is not sale
Unitree’s filing is unusually useful because its unit table reconciles operating volume with financial reporting. The company reported RMB867.8 million of humanoid robot revenue in 2025 from 5,215 sales, with an average recognized sales price of RMB166,400 per unit.
The same filing defines its 5,500-plus 2025 shipment figure as “pure humanoid” robots and explicitly excludes wheeled dual-arm robots.
UBTECH provides another strong company-level disclosure, but it should not be added mechanically to Unitree’s total. UBTECH reported 1,079 units of sales volume in 2025 for its full-size embodied intelligent humanoid robot products and services category, which generated RMB820.6 million of revenue.
The scopes are different. Unitree’s count covers its pure humanoid range across multiple sizes while excluding wheeled dual-arm machines. UBTECH’s figure covers its full-size humanoid category. Presenting 5,215 and 1,079 side by side is useful. Adding them into a supposedly comparable market subtotal would import the same definition problem that complicates the 2026 research estimates.
Both companies’ disclosures sit within securities filings associated with audited financial statements. That makes them substantially stronger than an unsupported shipment claim, but it does not mean the individual operating unit counts received separate audit opinions.
The practical lesson is narrower: company filings can distinguish production, shipment and recognized sale in a way that market shipment tables usually cannot.
The 2026 trackers agree on the leaders, not the denominator
Two independent research providers agree surprisingly well on who leads the market.
Smart Analytics Global estimates approximately 19,100 global humanoid shipments in H1 2026, including about 8,400 from AGIBOT and 5,900 from Unitree. Counterpoint Research estimates more than 22,000 globally, including about 9,700 from AGIBOT and more than 7,000 from Unitree. Both rank the same five companies in the same order: AGIBOT, Unitree, Galbot, UBTECH and Leju.
| Source | H1 2026 volume | AGIBOT | Unitree | Important scope point |
|---|---|---|---|---|
| Smart Analytics Global | About 19,100 globally | About 8,400 | About 5,900 | Includes full-size and smaller bipeds, wheeled humanoids and highly biomimetic humanoids |
| Counterpoint Research | More than 22,000 globally | About 9,700 | More than 7,000 | Same top-five vendor order as SAG; application mix differs materially |
| Humanoid Robot Scene Application Alliance | More than 30,000 in China | Not disclosed in public summary | Not disclosed in public summary | Industry-body estimate includes more than 18,000 bipeds and more than 12,000 wheeled humanoids |
| China Humanoid Robotics and Embodied Intelligence Committee of 100 | More than 40,000 in China | Not publicly reconciled | Not publicly reconciled | Public reporting does not disclose a vendor-level reconciliation or counting methodology |
The agreement on vendor order is useful evidence. The disagreement on magnitude is just as important.
SAG says Chinese vendors accounted for more than 97% of its 19,100 global estimate, while China itself represented more than 85% of demand. On SAG’s basis, Chinese-vendor shipments therefore amount to roughly 18,500 or more units.
Two Chinese industry sources describe a much larger base. The Humanoid Robot Scene Application Alliance estimated more than 30,000 H1 shipments in China, including more than 18,000 bipedal and more than 12,000 wheeled humanoids. A separate report from the China Humanoid Robotics and Embodied Intelligence Committee of 100 put China’s H1 volume above 40,000 and described it as 97% of the global total.
The important conflict is therefore not simply whether China’s share is 97%. It is the unit base underneath that percentage.
SAG has a roughly 19,100-unit global market and more than 97% Chinese-vendor share. The two domestic industry sources describe China volumes above 30,000 and 40,000. Their public materials do not establish that vendor origin, shipment geography, robot definition and recognition point are being measured identically.
Backing a global total out of the 97% figure would create precision the underlying sources do not support.
A sale still does not establish productive deployment
Unitree’s filing also provides unusually strong evidence about what those sales were actually for.
In its response to the Shanghai Stock Exchange’s second-round inquiry, Unitree disclosed that 73.60% of humanoid robot revenue during January through September 2025 came from research and education. Commercial and consumer applications represented 17.39%. Industry applications represented 9.01%.
“Research and education” is broader than university laboratories. Unitree says the category also includes research institutions, technology companies and individual developers purchasing robots for research, technical development or secondary development.
The industry category also requires care. Unitree said enterprise reception and guidance represented roughly 50% to 70% of humanoid industry-application revenue. It identified RMB15.702 million of revenue from clearer operating scenarios including intelligent manufacturing, intelligent inspection and logistics delivery. That represented 29.29% of the industry-application subtotal. The company also said it did not fully know the final use of robots where downstream customers conducted their own development.
This is stronger evidence for the commercial-readiness question than shipment volume alone. Unitree had already achieved thousands of humanoid sales, but most disclosed humanoid revenue was still associated with research and development rather than industrial labor.
Independent analysis points in the same direction, although with lower evidentiary weight than the company filing. SemiAnalysis estimated in June that Unitree may have shipped roughly up to 250 humanoids into productive industry pilots or deployments during 2025, excluding research and hobbyist sales. It also cautioned that many current deployments remain software-limited and often rely heavily on teleoperation.
That estimate is not a company-confirmed deployment count. It does, however, illustrate why several thousand recognized robot sales should not automatically be interpreted as several thousand robots performing productive labor.
The 2026 application estimates remain conflicted as well. SAG says industrial and commercial uses represented more than 70% of H1 shipments. Counterpoint says entertainment and performance plus data production and research remained above 60%, with intelligent manufacturing at 13% and warehousing and logistics at 5%.
Those classifications cannot both be used as if they describe an identical application taxonomy.
Public markets are now pricing the disclosure gap
The measurement question became more consequential in August because Unitree itself entered the public market.
Unitree listed on Shanghai’s STAR Market on August 19. Its offer of approximately 40.45 million shares at RMB150.80 each represented 10% of post-offering equity, implying an issue valuation of about RMB61 billion, roughly $9 billion at the time.
On its first trading day, the shares rose more than 600% intraday and closed about 460% above the IPO price, giving Unitree a market capitalization of roughly $51 billion at the close.
That stock-market performance is not deployment evidence and does not make Unitree’s robots more commercially mature. It does make disclosure quality more financially consequential.
Investors now have a public valuation attached to a business whose prospectus allows them to distinguish 5,716 units produced from 5,511 shipped and 5,215 sold, and to see that most recent disclosed humanoid revenue was still generated in research and education.
That is substantially more decision-useful than a market-share percentage without the underlying unit definitions.
What can actually be said about 2026
The available evidence supports several conclusions, but not a single global sold-unit number.
Humanoid robot production and shipments have clearly accelerated. SAG and Counterpoint independently put H1 global shipments above 19,000 and agree on the same five leading vendors. Chinese companies dominate the units counted by both.
The exact size of that market remains unresolved. Public H1 estimates range from approximately 19,100 globally to more than 30,000 or 40,000 in China under broader or differently disclosed industry methodologies.
More importantly, shipments cannot simply be relabeled as sales. Unitree’s own filing demonstrates a 296-unit difference between shipped and sold robots in 2025 because customer receipt and acceptance affected revenue recognition.
And sales cannot simply be relabeled as productive deployment. Unitree’s application disclosures show that research and education generated nearly three-quarters of its humanoid revenue during the first nine months of 2025, while clearly identified manufacturing, inspection and logistics applications represented a much smaller base.
A defensible 2026 market count will require vendors to disclose units produced, shipped, delivered, accepted and sold separately. Researchers also need to state whether they include wheeled humanoids, smaller research platforms and highly biomimetic systems, and whether “China shipments” refers to Chinese vendors or Chinese customers.
Until those fields can be reconciled, Humanoid Analytics would not publish one number for humanoids “sold in 2026.”
The most useful number in the sector may still be 296: the gap between Unitree robots shipped and sales recognized in one year. It is a small difference in absolute terms, but it demonstrates why the market’s much larger 2026 shipment totals cannot yet answer a seemingly simple sales question.
Sources:
- Unitree Robotics, “宇树科技股份有限公司首次公开发行股票并在科创板上市招股意向书”
Source type: Tier 1, direct or official evidence, securities filing, company-controlled disclosure
https://static.cninfo.com.cn/finalpage/2026-07-31/1225449783.PDF - UBTECH Robotics Corp Ltd, “Annual Report 2025”
Source type: Tier 1, direct or official evidence, exchange-filed company disclosure
https://www.hkexnews.hk/listedco/listconews/sehk/2026/0414/2026041401168.pdf - Smart Analytics Global, “SAG: Global Humanoid Robot Shipments Surged 272% YoY to 19.1K Units in 1H 2026; AGIBOT Overtook Unitree for No. 1 Position”
Source type: Tier 2, strong independent evidence, market-research estimate
https://smartanalyticsglobal.com/global-humanoid-robot-shipments-2026-agibot-unitree/ - Counterpoint Research, “Global Humanoid Robot Shipments Soar Nearly 300% YoY in H1 2026, Driven by Commercial Deployments”
Source type: Tier 2, strong independent evidence, market-research estimate
https://counterpointresearch.com/en/insights/global-humanoid-robot-shipments-soar-nearly-300-percent-yoy-in-h1-2026 - Humanoid Robot Scene Application Alliance / CMRA, “China’s Humanoid Robot Shipments Exceed 30,000 Units in H1 2026”
Source type: Tier 3, detailed industry-body estimate
https://cnmra.com/chinas-humanoid-robot-shipments-exceed-30000-units-in-h1-2026/ - China Economic Net, “China’s humanoid robot shipments top 40,000 in H1 amid factory push”
Source type: Tier 2, strong independent reporting of an industry-body claim
https://en.ce.cn/main/latest/202608/t20260821_3162430.shtml - Unitree Robotics, “关于宇树科技股份有限公司首次公开发行股票并在科创板上市预先审阅申请文件的第二轮问询函的回复”
Source type: Tier 1, direct or official evidence, securities filing, company-controlled response
https://static.sse.com.cn/stock/disclosure/announcement/c/202603/002178_20260320_BLK4.pdf - SemiAnalysis, “Unitree’s Impossible Trajectory Is Still Overlooked”
Source type: Tier 2, strong independent evidence, analytical estimate
https://newsletter.semianalysis.com/p/chinas-unitree-will-dominate-global - China Economic Net / Xinhua, “China’s robot maker Unitree opens IPO subscription after strong preliminary demand”
Source type: Tier 2, strong independent reporting based on securities issuance disclosures
https://en.ce.cn/main/latest/202608/t20260810_3138774.shtml - Financial Times, “Chinese humanoid robot maker surges 600% in trading debut”
Source type: Tier 2, strong independent evidence
https://www.ft.com/content/14cd8246-7fb1-4f8f-81b4-8de11ced79e5