Humanoid robotics has an honesty problem. For years the sector's most famous moments were videos: a backflip, a parkour run, a dance routine. What was hard to find was a production number.
On 12 August 2026, Unitree Robotics published one. Cumulative production of its bipedal humanoids had passed 18,000 units.
The figure is deliberately narrow. It counts only pure bipedal humanoid machines — the H1, G1 and H2 — and excludes quadruped robot dogs, robot arms and wheeled-base humanoids. The company listed on the STAR Market on 19 August 2026, a week later.
Key takeaways
- Cumulative production: more than 18,000 bipedal humanoids (as of the August 2026 disclosure), against 6,500 produced across 2025.
- Shipments: over 5,500 units in 2025, ranked first globally; roughly 5,900 in the first half of 2026, reported as about 31% of the global market.
- Finances: revenue grew from 159 million yuan in 2023 to 1.708 billion yuan in 2025, with net profit turning positive at 288 million yuan in 2025 after a loss in 2023.
- Ramp speed: the compact G1 passed 11,000 units produced by May 2026, and 7,000 more were built across June and July alone.
- Capacity: current humanoid capacity of about 12,000 units a year, planned to expand to 50,000 with the new IPO-funded base, and total all-category robot capacity of 190,000 including the Ningbo facility.
- Listing: STAR Market IPO application accepted 20 March 2026, listed 19 August 2026, raising about 4.2 billion yuan.
Why the number matters
A production count is the only statistic in robotics that cannot be faked with a rendering. It settles an argument that has run since the first viral humanoid videos: is this an industry, or is it a demo reel?
Eighteen thousand units is not a consumer product. It is, however, unambiguously a manufacturing operation — and the customer mix explains where the value currently sits. More than 70% of Unitree's humanoid customers are universities, research institutes and cultural-tourism or performance venues. The G1 education variant, priced between roughly 145,000 and 414,000 yuan, is the volume driver. Deployment on industrial production lines remains at small-scale pilot stage.
That is a realistic picture of 2026: humanoids are being bought as research platforms and public-facing attractions, with factories as the next hill to take.
The two-brain strategy
Unitree's own assessment is that hardware is the solved part and the deficit is intelligence. The company is pursuing two model lines in parallel — world models and vision-language-action (VLA) models — and has directed more than 2 billion yuan of IPO proceeds into intelligent robot model research and development.
It also signed a strategic memorandum with DeepSeek to work jointly on large models and embodied intelligence. The logic is straightforward: China's open models are cheap enough to run at the edge and strong enough to control a body, and a robot company that owns the actuators should not have to rent the brain forever.
Capability, demonstrated in public
Unitree has used spectacle as a development forcing function, and the results have been genuinely novel:
- 2025 CCTV Spring Festival Gala: the "YanBot" performance.
- 2026 Spring Festival Gala: "WuBOT", reported as the world's first fully autonomous humanoid cluster martial arts performance — multiple robots coordinating with no human teleoperation.
- August 2025, World Humanoid Robot Games: four gold medals on the H1 and G1 platforms.
- 16 April 2026, Beijing Yizhuang half-marathon qualifier: an H1 autonomously completed a 1.9 km multi-turn course in 4 minutes 13 seconds — a pace that, scaled, would beat the human 1500 m world record.
None of these are factory productivity. All of them are evidence that locomotion, balance and multi-agent coordination have crossed from fragile to reliable — which is the precondition for anything else.
The honest caveats
Reports on the company note a gap between units produced and units delivered, with inventory going to education and exhibition orders that run over long horizons. Analysts also point out that the competitive emphasis has shifted decisively from prototype research to stable mass production, cost control and scenario deployment — which is precisely where a company with 18,000 units of experience has an advantage over one with a better video.
A rival, AgiBot, has reported roughly 15,000 cumulative units, and global humanoid shipments in the first half of 2026 have been estimated at about 19,100 units. This is now a manufacturing race, not a science fair.
Production, shipment and financial figures are as disclosed by the company and reported by Chinese financial media in 2026; market-share estimates are third-party.