On October 23, Unitree hit its first limit-up since listing. The shares closed at 624 yuan, up 20%, and the market cap went back above 250B yuan, a one-day gain of more than 40B, per Yicai. The stock had bottomed on October 15. Six sessions later it was up more than 54%.
Start with where the price came from. Unitree sold shares at 150.80 yuan, a valuation of about 61B yuan, as we covered on August 18. Day one closed at 845 yuan after touching 1,100, the round trip we wrote up on August 28. Today’s 624 is 4.1 times the offer price, and the market cap is 4.1 times the offer valuation.
The denominator
Per Eastmoney’s Oct 19 recap, Unitree’s first-half 2026 revenue was 1.152B yuan, up 48.54%. That beat the top of the 1.052B to 1.128B yuan range the company guided to, by about 2%. Net profit attributable to shareholders was 274M yuan against a loss of about 32M a year earlier. Net profit excluding one-offs was 244M yuan, down 19.34%.
Those last two lines point in opposite directions, and the coverage we found does not reconcile them. Quote both.
The valuation math is ours. 250B divided by 1.152B is 217 times half-year revenue. Double the half for a rough year and it is about 108 times sales. Double the 244M excluding one-offs and the price is around 512 times earnings. Seasonality makes the doubling crude. It sizes the gap, nothing more.
For scale: Unitree’s 2025 revenue was 1.70B yuan, of which humanoids were 868M on more than 5,500 units. Six months of 2026 already equal 68% of that year.
What Wang Xingxing said
At the October 22 World Physical AI Conference, which Eastmoney had billed for 3,000-plus guests and 40-plus exhibition zones, founder Wang Xingxing put the production number out loud. Unitree currently makes tens of thousands of robots a year, plus several hundred thousand joint motors and modules. His own verdict was that this is far short of mass adoption. If robots spread widely, he said, annual output would need to reach at least 10 million units.
Two programs came with it. A World Physical AI Competition with up to $10M a year in prize support. And the Yuchuang Program, which Unitree sizes at several billion yuan of collaboration over five years, with more than 200 companies to be incubated.
He also said internal tests of GPT-6 showed robot control without task-specific training, and that physical AI breakthroughs are one to three years out. That is Yicai’s relay of a conference remark. We found no test data behind it.
Against the cohort
Frost & Sullivan put global H1 humanoid sales at 36,000 units and $540M, and only 1,200 of them, 3.3%, were doing production work. The five Chinese vendors at the top, Unitree among them, hold 57.4% of the market. The MIIT’s own count for the half was 40,000 units. Unitree’s August filing said about 18,000 biped humanoids in its entire ten-year history.
On September 9 the CSRC raised the listing bar for humanoid companies, and the number that day was that under 10% of Unitree’s 2025 revenue came from industrial applications.
So a 250B yuan company reports 1.15B of half-year revenue, a small slice of it in factories, and makes tens of thousands of units a year. The price is not for any of that. It is for the year Wang described, in which output is not tens of thousands.
The career read
The market is pricing a labor-substitution story that the production data cannot yet support. The exposed work today is narrow: teleoperators and data collectors who train the machines, and the research and education buyers who are most of the demand. Factory floors come later, and the 200-company incubation plan is how they would arrive. Many of those robots will carry other brand names on Unitree actuators and models.
Three things to watch. Unit shipments broken out from revenue. Industrial share, which was under 10% in 2025. And the first deployments from the Yuchuang companies, which are the real test of whether this price is paying for factory work.
