Unitree Robotics' debut on Shanghai's STAR Market priced in, all at once, the expectation that humanoid robots will transform the factories of the future. On August 19, the stock closed at 845 yuan, 460% above its IPO price of 150.80 yuan. At its intraday peak on listing day, it was reportedly up 629%. The company issued 40.45 million new shares and raised about 6.1 billion yuan. Based on the closing price and roughly 404.5 million shares outstanding after the listing, its market capitalization comes to about 341.8 billion yuan.
But the demand the share price anticipates is not the same as the demand already showing up in revenue. Set Unitree's listing documents beside those of Leju Robotics, which has applied to list in the same Chinese market, and the current buyer base looks broader than factory automation. Universities, research institutes, developers, and data-collection centers set up with policy support are the ones buying actual robots. Rather than measuring how much on-site labor robots have replaced, the numbers show that hardware and data infrastructure for training embodied AI are expanding first.
That does not mean the sales are not real. Research, education, and data collection are paid uses, and real-world robot data is a necessary input for developing models that operate in the physical world. At the same time, this revenue does not confirm sustained factory use, repeat purchases, or a return on investment. In reading the first-day valuation, it is necessary to separate which demand supports today's revenue and which is still a future hypothesis.
A 341.8 Billion Yuan First-Day Valuation vs. 2025 Revenue
Unitree issued 40.45 million shares at 150.80 yuan each. That equals 10% of the post-offering share count, for gross proceeds of about 6.1 billion yuan and expected net proceeds of 5.92 billion yuan. According to the Associated Press, the stock ended its first day at 845 yuan. The market value of about 341.8 billion yuan, obtained by multiplying roughly 404.5 million shares by the closing price, is based solely on the August 19, 2026 close and does not reflect later prices.
For comparison, 2025 revenue was 1,699,269,300 yuan. The company disclosed a gross margin of 60.13%, net profit of 278,210,500 yuan, and adjusted net profit of 590,752,800 yuan. Dividing the first-day market capitalization by that revenue gives roughly 201.1 times. This is not a multiple based on forecast earnings, but it does show that the market priced in growth well beyond current revenue.
Meanwhile, first-quarter 2026 revenue was 422,840,500 yuan, up 68.49% year on year. Adjusted net profit was 40,253,600 yuan, down 52.55%, which the company attributed to higher R&D and selling expenses. Revenue and profit are not growing at the same pace. That is why supporting the valuation requires looking not only at rising shipment volumes but also at where profitable uses are spreading.
73.60% of Unitree's Humanoid Revenue Comes From Research and Education
In its responses to regulatory inquiries, Unitree reported humanoid robot revenue of 595,187,900 yuan for January to September 2025. Of that, research and education accounted for 438,067,400 yuan (73.60%), commercial and consumer uses for 103,516,900 yuan (17.39%), and industrial uses for 53,603,600 yuan (9.01%). At present, the core of sales is not factory automation but customers who use robots for research and development.
The company defines its research and education customers broadly. They include universities, research institutes, and technology companies, as well as individual developers who buy for research, technology development, or secondary development. The 73.60% figure therefore cannot be read as sales to government data-collection centers. Still, this category shows that there is a large market of buyers who purchase robots as tools for learning and development, separate from the industrial uses that directly measure labor replacement.
The breakdown of industrial uses also matters. Revenue explicitly classified as manufacturing, inspection, and logistics was 15,702,000 yuan, or 29.29% of industrial revenue. Calculated against total humanoid revenue for the same January–September 2025 period, that equals 2.638%. The remainder is understood to be mainly corporate reception and guide roles. Confusing the 9.01% industrial figure with the share of revenue that actually went to manufacturing, inspection, and logistics makes factory use look larger than it is.
At Leju, Data Collection Became the Largest Buyer
Leju Robotics disclosed in its listing application that 2025 revenue from Kuavo was 177,782,600 yuan. The largest category was data collection at 79,904,100 yuan, or 44.94%. Research and education accounted for 57,771,700 yuan (32.50%), commercial services for 33,213,200 yuan (18.68%), and industrial manufacturing for only 6,893,600 yuan (3.88%). Because the disclosure categories at Unitree and Leju are not identical, the percentages cannot be compared one-to-one. Even so, the two companies' filings agree that uses outside factories make up most of revenue.
Leju sold 577 Kuavo units in 2025, at an average price of 308,115.40 yuan per unit. In 2024 it sold 32 units at an average of 413,936.39 yuan, so unit volume rose while the average price fell 25.56%. The company's filing says it won bids for multiple regional data-collection centers. The disclosure indicates that procurement for data-gathering sites supported part of the sales, rather than showing that more ordinary factory customers adopted the robots.
These centers are often set up under local government leadership in cooperation with robot makers, we are told. They replicate real work environments one-to-one and record motion and environment data from teleoperation and demonstrations. The formatted, standardized datasets are sold to robot makers, embodied-model companies, and research institutes, and training facilities may also be rented out. Deploying robots generates more data, and models trained on that data are then fed back into the robots. This cycle functions as infrastructure for technology development.
At Least 90 Sites Are Creating Demand for Training Hardware
According to Interact Analysis, as of the end of April 2026 there were at least 90 humanoid data-collection and training centers in China that were operating, planned, or under construction. Of these, 64 were operating, and 28 opened in 2026. At least 13 of the operating centers house 100 or more robots. Efforts have moved beyond lab-scale experiments to prepare fleets of robots across the country for data acquisition.
Policy is also pushing in this direction. The "AI+ Manufacturing" action plan announced by eight Chinese government departments calls for establishing humanoid robot test sites, training grounds, and benchmark production lines, and for advancing early adoption in representative manufacturing scenarios. Public procurement and site development increase the number of installed robots and, at the same time, the supply of data. This could speed up model training and encourage cost reductions.
However, purchases by data-collection sites should be counted separately from demand from independent end users. Center purchases have a clear role: producing motion data in controlled environments. Getting from there to replacing staff on a factory line requires confirming that tasks can be reproduced, that operation is safe, and that maintenance and post-deployment profitability hold up. Unitree's filings also do not show what share it sold to regional data-collection centers.
Factory Use Can't Yet Be Gauged From the Revenue Mix Alone
What investors should check next is whether sales in industrial manufacturing, inspection, and logistics repeat beyond one-off demonstrations. At Unitree, revenue explicitly classified as manufacturing, inspection, and logistics was 2.638% of humanoid revenue for January to September 2025. At Leju, industrial manufacturing was 3.88% in 2025. The two companies' accounting categories differ, but neither shows a ratio that would let permanent factory use serve as the center of growth.
Another question is how far motions improved with collected data can transfer beyond tasks replicated at the sites. Training facilities recreate real environments, which makes them useful for validating technology and acquiring data. But operating there is not the same as working continuously across diverse factories and service settings. What robot makers need to show is not the number of sites where robots are deployed but evidence that the same customers keep buying for other sites and other tasks.
The August 19 closing price of 845 yuan reflects the moment when the market placed large expectations on China's humanoid robot industry. Whether those expectations last depends on whether the cycle of hardware and data built through research, education, and data collection can move, outside policy-backed sites, into repeatable work in factories and service settings.