⁠Inside Unitree’s $9B IPO: Wang Xingxing’s one-of-a-kind hardware vision

In August, Unitree Robotics became China’s first publicly traded humanoid robot maker. It priced its Shanghai STAR Market listing at RMB 150.80 ($22.34) a share, valuing the company at about $9 billion, well above its earlier $6.2 billion valuation target. The order book was oversubscribed more than 8,000 times — a record for the Chinese market — and regulators waved the deal through in 104 days, signaling just how badly the state wants this industry to exist. That said, the listing documents were unusually blunt about what Unitree robots can do. They cannot yet perform genuinely useful work reliably. That is not a critic’s verdict but the company’s own disclosure. The founder has been saying a version of this statement for years. Wang Xingxing, 36, is the most successful robot maker on earth by volume, and he keeps telling audiences that the intelligence needed to make robots truly useful does not yet exist. Why he says that and what he does instead explains more about the robotics business than any demo video. Making robotics cheap Wang grew up in Ningbo and built his first legged robot, XDog, as a master’s student at Shanghai University on a budget that would embarrass a research lab. It won a competition prize and circulated among Chinese engineers, who noticed the detail that would define his career: Wang had built a capable robot for almost nothing. Wang spent a brief period at drone maker DJI before founding Unitree in Hangzhou in 2016. His cost-first strategy has not changed: where competitors bought proprietary actuators from specialist suppliers, Unitree designed its own; where competitors sold to institutions with unlimited budgets, Unitree sold to everyone else. The numbers explain the attention. Boston Dynamics’ Spot robot has a base price of around $75,000; Unitree’s comparable Go1 launched at $2,700, and its R1 humanoid now sells for roughly $5,900, less than a decent gaming computer. In 2025, Unitree shipped about 5,500 humanoids, more than any other company in the world, and revenue quadrupled to RMB 1.7 billion ($253 million) at gross margins near 60%. Cheap did something more interesting than win customers. It put capable hardware within the reach of almost everybody. Universities that could never justify a Spot robot bought Unitree machines by the dozen; startups that could not afford to experiment suddenly could; and rival robot companies began running their own software on Unitree hardware. Wang turned his product into the industry’s default laboratory bench, meaning much of the world's robotics research now runs on a Chinese platform. The bet nobody else is making This is where Wang separates from the industry. The consensus in robotics is that hardware is close to solved, and the prize belongs to whoever builds the brain. Silicon Valley is spending enormous sums on exactly that. So are China’s tech giants. Wang has declined to join the race on those terms. He has argued the AI for embodied intelligence remains inadequate, that assuming a ChatGPT-style model can be pointed at a robot is wrong, and that Unitree cannot match the scale of dedicated AI labs and will not pretend otherwise. Robots, he says, are the foundation. He has gone as far as telling customers they may replace Unitree’s software and keep its body. Almost no founder in any industry says that. He is openly ceding the layer everyone calls the future and defending the one everyone calls a commodity. The logic is defensible. Nobody knows which brain will win, and every winning brain still needs a body. Unitree has hedged too, open-sourcing its own world-model framework, UnifoLM, followed by a vision-language-action model: an ecosystem play, but also an admission that no single company can generate enough physical training data alone. The risk is just as clear, and Chinese manufacturers know it from the smartphone era: whoever owns the operating system takes the profit, while the hardware maker competes on price forever. Wang is betting robots break that pattern. That bet is what investors just bought. What the market is actually paying for The financials show a company at a hinge point. Revenue quadrupled in 2025, and Unitree is targeting 10,000 to 20,000 shipments this year. However, first-half growth cooled sharply from the previous year’s pace. A 60% gross margin is exceptional, but it exists partly because Unitree’s buyers are still research labs, developers, and showrooms rather than factories putting robots to work at scale. So what did the oversubscription buy? Not earnings. It bought an option on the body layer of an industry that may or may not arrive. If humanoid robots become ordinary, the company already making them at scale and at the lowest cost would be positioned to supply the world. The candor of the prospectus and the frenzy of the order book describe the same gap from opposite ends. Why China keeps winning the hardware half Unitree is not the accident of one clever founder. And the usual explanation, that China has a good supply chain, is too vague to be useful. Four factors are actually doing the work. The first is inheritance. Humanoid robots need precision motors, reducers, batteries, and sensors. China built all of these at scale for electric vehicles and drones. Unitree buys from an ecosystem the EV boom paid for, and its suppliers accept small experimental orders that Western component makers often refuse. The second is design order. Chinese teams routinely fix the price first and engineer down to it, while Western robotics companies build the best possible machine and price it afterwards. Only one of those habits produces a $5,900 humanoid. The third is deployment as data collection. Every robot sold is a sensor package running in a real environment, and China has more of them in the field than anywhere else. If the central bottleneck is a shortage of real-world physical data, cheap hardware is really a data strategy disguised as a pricing strategy. The fourth is the state. The 104-day review, municipal robot training centers, and a capital market primed to absorb the listing are all deliberate. Embodied AI is national policy, and Unitree is its showcase. The same attention cuts both ways abroad: Washington added Unitree to a defense-linked watchlist in June 2026, and its supply chain still depends on Western technology it does not control. What comes next, for Unitree and for robots Unitree’s roadmap for 2026 and 2027 points toward greater dexterity and, ultimately, the home: household help and elderly care, the market everyone wants but nobody has cracked. The listing proceeds will fund model research, new products, and a manufacturing base. The wider industry is running three bets at once. The first is the brain: world models trained on generated environments, where Nvidia, the American labs, and China’s AgiBot are concentrated. The second is the business model: robots rented by the hour rather than bought, so customers avoid owning technology that ages in two years. The third is form: whether the human shape is right at all, or whether wheels and specialized arms will do the real work while humanoids remain the thing that gets filmed. Wang is positioned to survive all three, which is the point of his strategy. It is also why the next two years will test whether being indispensable to an industry is the same as capturing its value. What his story teaches Most companies chase whatever is fashionable and end up mediocre at several things. Wang picked the body, said so publicly, and told customers to bring their own software. Clarity about what you are not doing is also a strategy. Price is a decision made before engineering, not after. The $5,900 humanoid exists because the number came first. Companies that treat cost as an outcome of design never get there. Arming your competitors can be the strongest position available. Selling cheap to universities, startups, and rivals made Unitree the standard platform, and standards outlive products. Credibility compounds when you state what your technology cannot do. Wang’s insistence that robot intelligence is not ready costs him nothing, while competitors who overclaim will spend years being measured against their demo videos. The bottom line The market just put roughly $9 billion on a company whose founder keeps saying the technology is not ready and whose prospectus agrees with him. That is not a contradiction but a wager: that the useful robot is coming, that nobody knows which intelligence will drive it, and that when it arrives it will be wearing a body from Hangzhou. Wang has spent a decade making the answer to one question boring and settled. Whoever finally builds a robot that can do your laundry will probably buy the legs from him. Whether that is the most valuable position in robotics or merely the busiest is the thing to watch.

⁠Inside Unitree’s $9B IPO: Wang Xingxing’s one-of-a-kind hardware vision

How 2026 welcomed ‘Too Cool’ back

Upgrade to Pro

Sectors

Insights

Company

⁠Inside Unitree’s $9B IPO: Wang Xingxing’s one-of-a-kind hardware vision

How 2026 welcomed ‘Too Cool’ back