A first-day scarcity event
Research cutoff: August 20, 2026.
Unitree Robotics began trading on Shanghai's STAR Market on August 19 at an offer price of RMB 150.80. The company raised about RMB 6.1 billion. Its shares reached an intraday gain of 629% and closed at RMB 845, up 460% from the offer price. The closing price valued Unitree at approximately RMB 341.8 billion, or about $50 billion.
Unitree entered the market with an offer valuation of RMB 61 billion and an already demanding price to earnings ratio of 219.23. One trading session added roughly RMB 281 billion of equity value, more than 165 times the company's 2025 revenue. The IPO had been expected for weeks, but few valuations implied anything close to the closing price.
Part of the move was mechanical. Only 10% of the post-offering shares were issued in the IPO, and strategic placements and lockups reduced the shares immediately available to trade. STAR Market IPOs have no daily price limit for their first five trading days. After that period, the normal daily limit is 20%. A thin float, intense retail demand, and the absence of an initial limit can produce a price that clears scarce supply without representing a settled view of long-run earnings.
The first-day market record shows how investors are combining a proven fact with a much larger forecast. Unitree can sell robots at unusual scale. Buyers are paying in advance for a future in which those machines perform useful work for many hours a day, something that has barely begun to appear in customer operations.
The company beneath the trading spectacle
Unitree is not a pre-revenue robotics laboratory. Its final prospectus reports 2025 revenue of RMB 1.699 billion, up from RMB 393 million in 2024. Humanoid robot revenue reached RMB 868 million and accounted for 51.78% of the main business. Quadruped robots contributed RMB 698 million, or 41.62%.
The volume and price data explain why Unitree matters. It sold 5,215 humanoids in 2025 at an average price of RMB 166,400, compared with 412 units at RMB 260,400 in 2024. Four-legged robot sales reached 23,037 units at an average price of RMB 30,300. Lower average prices did not damage revenue because unit volume rose much faster. That is the Chinese manufacturing advantage in its most concrete form: integrated motors and joints, fast product iteration, a broad supplier base, and hardware cheap enough for universities, developers, entertainment companies, and industrial customers to buy rather than merely test.
Profitability also separates Unitree from many robotics peers. Reported 2025 net income was about RMB 278 million. Profit excluding nonrecurring items was RMB 591 million because the reported figure absorbed a large noncash share-based compensation charge. Operating cash flow was positive. The company had moved beyond prototype economics before asking public investors for capital.
The latest operating direction is less explosive than the 2025 comparison. The prospectus estimated first-half 2026 revenue of RMB 1.052 billion to RMB 1.128 billion, an increase of 35.62% to 45.41%. It also estimated adjusted profit of RMB 236 million to RMB 283 million, down between 21.97% and 6.43% from the prior-year period as research, sales, and other operating expenses increased. Growth above 35% would be exceptional for an established machinery company. It is modest relative to what the first-day valuation now requires.
Unitree spent RMB 145 million on research and development in 2025, equal to 8.53% of revenue. The absolute amount doubled, but revenue grew faster and pulled the ratio down from 17.83% in 2024. The lower ratio does not by itself prove underinvestment. Unitree's manufacturing staff, suppliers, and production engineering also create technical value. Investors still need to see whether the newly raised capital produces a stronger autonomy stack or mainly adds hardware capacity.
The valuation requires productive hours
At RMB 845, Unitree traded at about 201 times 2025 sales. Depending on whether the statutory or adjusted profit figure is used, the price represented roughly 1,229 times reported earnings or 579 times adjusted earnings. These are not normal hardware multiples. They are claims on a possible future labor platform.
Reverse valuation makes the claim easier to see. If Unitree eventually traded at eight times sales, still a premium multiple for a scaled manufacturer, the current market value would require annual revenue of RMB 42.7 billion. That is 25 times the 2025 level. If investors instead valued a mature Unitree at 40 times earnings, the company would need RMB 8.5 billion in annual profit, about 14 times its 2025 adjusted profit.
The following scenarios are arithmetic, not price targets. They hold the starting revenue at the 2025 level and apply different five-year growth rates and terminal sales multiples.
| Scenario | Revenue growth, 2025 to 2030 | 2030 revenue | Illustrative sales multiple | Implied equity value |
|---|---|---|---|---|
| Strong industrial adoption | 45% a year | RMB 10.9 billion | 6x | RMB 65 billion |
| Broad commercial adoption | 70% a year | RMB 24.1 billion | 8x | RMB 193 billion |
| Exceptional platform outcome | 100% a year | RMB 54.4 billion | 8x | RMB 435 billion |
Even the middle case would leave the implied 2030 value below the first-day close. The current price becomes easier to defend only if revenue roughly doubles every year through 2030, if software and service revenue support a much higher multiple, or if investors continue discounting earnings beyond that date. None is impossible. All demand evidence that robot shipments are becoming paid, repeatable work.
A humanoid sold to a research lab counts as one unit. A humanoid operating two shifts in a factory also counts as one unit. The economic difference is enormous. The second machine generates task data, service revenue, renewal evidence, and a measurable return for the customer. Shipment leadership is useful, but productive hours will determine whether Unitree deserves a platform valuation.
Motion control is ahead of general autonomy
Unitree's strongest technical position is sometimes called the robot's "small brain": locomotion, balance, joint control, reinforcement learning for movement, and the electromechanical system that turns commands into stable action. Its videos are effective because the machines can run, recover, dance, fight, and absorb disturbances that would defeat many laboratory prototypes. This competence is real, and it lowers the cost of gathering physical data.
The prospectus is explicit about what remains unfinished. Unitree had not yet deployed its self-developed general embodied model at scale in sold products. Some models used third-party language models for conversation. Its own UnifoLM systems had been open-sourced or tested, and an industrial model had completed a pilot assembling joint motors in Unitree's factory. A pilot proves that the stack can perform a selected task under controlled conditions. It does not yet prove general autonomy across customers, objects, lighting, floor conditions, or work schedules.
Every major humanoid developer faces the same problem. Reliable walking is no longer the only bottleneck. Hands must tolerate thousands of grasps without losing calibration. Perception must recognize unfamiliar objects. The control system must recover safely from mistakes near people. A customer must be able to change a workflow without commissioning a robotics team for weeks. Maintenance, batteries, cybersecurity, and insurance have to fit the economics.
Unitree can close part of the gap through volume. More affordable hardware puts more sensors and actuators in the field, which can create a large data advantage. Data from remotely controlled demonstrations or choreographed performances has less commercial value than data from autonomous paid work. The useful loop begins with one deployed task, records the failures, improves the policy, and tests whether output per hour rises without more human supervision.
China and the West are testing different advantages
The leading Chinese and Western companies do not occupy one neat ranking. They are optimizing different bottlenecks.
| Company | Current proof point | Main advantage | Main unresolved risk |
|---|---|---|---|
| Unitree | More than 5,000 humanoids sold in 2025, plus a much larger quadruped business | Low hardware cost, motion control, supplier depth, rapid volume | General embodied AI is not yet deployed at scale; revenue still says little about autonomous productive hours |
| Tesla Optimus | First-generation production lines are under construction in California and Texas | AI compute, vertical integration, manufacturing capital, and an internal factory network | No disclosed external robot sales or sustained commercial deployment |
| Figure | More than 350 Figure 03 units delivered from BotQ, with a demonstrated rate of one robot per hour | Integrated vision-language-action model, fast iteration, and BMW factory experience | Small installed base relative to Unitree and no public evidence of profitable unit economics |
| Agility Digit | Commercial warehouse deployments, more than 100,000 totes moved at GXO, and orders tied to its next model | Narrow task focus, safety work, and a Robots-as-a-Service model | Lower production scale and dependence on a small number of enterprise workflows |
Tesla's second-quarter filing said the Model S and Model X lines in Fremont had been removed to make room for the first Optimus production line. The initial robots will enter an internal training program. Tesla has more capital, compute, battery experience, and factory space than a robotics startup, but construction is not output and internal use is not an external market.
Figure is making the strongest Western claim that AI and manufacturing can advance together. In April, the company said it had delivered more than 350 Figure 03 robots and demonstrated a one-unit-per-hour production rate. An earlier model accumulated 1,250 operating hours at BMW and handled more than 90,000 parts. Figure trails Unitree in shipment volume, but it is trying to connect production, real-world tasks, and training data inside one proprietary stack.
Agility has chosen a narrower route. Its Digit robots move totes and connect existing warehouse automation rather than trying to solve every human task. The company's public transaction materials describe a planned $2.5 billion public-market valuation, an order tied to 1,000 Digit V5 robots, and recurring service economics. Those orders remain subject to deployment milestones, but the model asks a useful question: can a customer buy productive capacity instead of robot hardware?
Geography will make the contest less direct. Unitree earned a large share of its earlier revenue abroad, but the United States has blocked authorization for new foreign-produced advanced robot models on national-security grounds. China offers the world's deepest robotics supply chain and a large domestic factory base. Western companies will have better access to U.S. industrial customers and some protected markets. The result may resemble electric vehicles: separate regional ecosystems, different cost structures, and less head-to-head competition than a global ranking implies.
Commercial adoption starts with narrow tasks
The International Federation of Robotics describes humanoids as promising but still limited in actual sales and deployment. Its assessment of vision and reality is a useful corrective to forecasts that jump straight from a demonstration to millions of household robots.
The first durable market is likely to be structured work with high labor turnover and simple output measurement. Moving containers between an autonomous mobile robot and a conveyor is boring, which is precisely why it is commercially useful. Automotive sequencing, machine tending, parts presentation, inspection, and hazardous-site work have similar properties. The environment is mapped, the objects are constrained, and a buyer can compare the robot with an existing labor or automation cost.
The decisive metric is cost per productive hour. Hardware price is only the numerator's first item. Financing, integration, service, spare parts, charging time, supervision, and downtime all belong there. Productive hours exclude rehearsals, remote-control recovery, and time spent waiting for a technician. A cheaper robot with low autonomy can cost more per completed task than an expensive robot that runs reliably through two shifts.
Humanoids will not absorb the whole robotics market. A fixed arm is better when the task never moves. An autonomous mobile robot is cheaper when goods only need transport. A wheeled base can be more stable and energy efficient than legs on a flat warehouse floor. Humanoid form earns its cost when customers need one machine to use stairs, tools, shelves, doors, carts, and workstations built for people. The related InvisibleHill report, The Next Robot Boom Will Be Broader Than Humanoids, explains that boundary in more detail.
Through 2028, most revenue should still come from research units, entertainment, logistics, factory pilots, and a small number of production deployments. The next stage requires hundreds of robots at one customer, high uptime, and renewals based on measured savings. Household adoption comes later. Homes combine children, pets, stairs, fragile objects, clutter, privacy, and liability in one unstructured environment. A factory can redesign a cell around a robot. A household will expect the robot to adapt to the house.
What can move the shares after the debut
The first five trading days will be dominated by supply, order flow, and the absence of a daily price limit. Once the 20% limit begins, price discovery may remain violent because buyers are comparing a small current business with a very large possible market. A precise short-term price forecast would imply knowledge the available evidence does not provide.
Several disclosures can narrow the range. The first is the actual first-half report. Investors need the final revenue, adjusted profit, gross margin, cash conversion, inventory, and receivables rather than the prospectus estimate. If revenue growth remains near 40% while expenses rise quickly, the stock will have to rely more heavily on distant expectations.
Product mix matters just as much. A higher count of low-priced education or entertainment units can lift shipments while weakening average revenue and doing little for autonomy. Industrial contracts should disclose fleet size, task success, uptime, paid operating hours, human interventions, and whether the customer expanded after a pilot. Software subscriptions and service revenue would support a platform multiple more convincingly than another acrobatics video.
The IPO proceeds create a third test. Unitree planned to spend on embodied models, robot bodies, new products, and manufacturing capacity. More factory space is useful only if demand and autonomy arrive together. A faster rise in model research, data operations, and customer deployment staff would show that management understands where the bottleneck has moved.
Finally, investors must watch overseas restrictions, domestic price competition, and future share unlocks. Unitree's low prices can expand the market, but they also invite competitors to cut prices before software revenue develops. The U.S. restriction removes an important future market for new models. Lockup expirations will eventually increase supply and reduce the scarcity that helped shape the first-day close.
The evidence threshold for the stock
Humanoid robotics has moved beyond a collection of science projects. Unitree has profitable hardware sales. Figure has accumulated factory hours. Agility has a commercial warehouse workflow, while Tesla is building production infrastructure. None yet demonstrates a general-purpose robot that customers can deploy across many tasks with little supervision.
Unitree has the clearest volume lead and one of the best cost positions. It can use the IPO capital to turn that hardware base into an autonomy and service business. If it does, the company may remain central to the global industry even as markets split along geopolitical lines.
At the first-day close, investors were already valuing many years of exceptional growth and a successful transition from machines sold to labor delivered. Future disclosures need to show the fleet size, autonomy rate, customer renewals, software revenue, and productive hours behind that transition. For now, the financial statements support the case for an unusually capable robotics manufacturer. The RMB 341.8 billion labor-platform valuation still rests on evidence that has not been reported.
Sources
Research cutoff: August 20, 2026. Valuation scenarios are InvisibleHill Research calculations, not price targets or investment advice. Company deployment figures are self-reported unless otherwise stated.
- Unitree Robotics, final STAR Market prospectus
- Associated Press, Unitree's August 19 trading debut and global shipment context
- Tesla, second-quarter 2026 shareholder update filed with the SEC
- Figure, Figure 03 manufacturing ramp and fleet data
- Agility Robotics and Churchill Capital XI, investor presentation
- International Federation of Robotics, Humanoid Robots: Vision and Reality