Two numbers landed in the same week, and they do not agree with each other.
Unitree told the world, in a social-media post, that it had cumulatively produced and delivered about 18,000 bipedal humanoid robots across all its models as of July.
Then its stock closed up 629 percent on its Shanghai debut.
So which future did the market just buy? Because it plainly wasn't the one where a company has shipped 18,000 walking machines in its entire history.
Last week we covered the listing itself — the roughly $9 billion valuation, the question of whether Unitree's manufacturing lead could ever convert into a defensible "brain" premium. That thread is now closed. The debut priced, popped, and settled the listing. What's left is harder: the multiple has to be justified by the models, not the machines.
Start with the arithmetic, because the arithmetic is the whole story.
Unitree priced its IPO at 150.8 yuan a share. It opened trading and ran to 1,100 yuan — a 629 percent gain — pinning the company at a valuation north of 61 billion yuan.
At the offer price, before the pop, the prospectus already put Unitree at 35.89 times sales. Hong Kong-listed peers like UBTech and Dobot trade around 20 times.
So the market started the day paying nearly double the sector's going rate for revenue, then bid the equity up another 629 percent on top of that. Reports peg the earnings multiple above 200 times.
Now hold that against production. Roughly 18,000 bipeds, ever. Unitree shipped more than 5,500 of them last year, enough to rank first in the world.
Let that sit. The global sales leader in humanoid robots is a company whose lifetime output would fit inside a mid-sized office park, and its debut valuation implies a mass-adoption endgame that its own shipment numbers are nowhere near.
The global sales leader in humanoid robots is a company whose lifetime output would fit inside a mid-sized office park.
The counterintuitive part is that this isn't a story about lying. Unitree's revenue is real and growing fast — it more than quadrupled to 1.7 billion yuan in 2025, with gross margin above 60 percent. Those are genuine numbers.
The problem is the tell buried in the first-quarter 2026 filing, and it's the tell that froth is designed to obscure.
Q1 revenue rose to 423 million yuan. But the growth rate collapsed — from 332.64 percent a year earlier to 68.49 percent.
And adjusted net profit, stripped of non-recurring items, fell 52.55 percent to 40.25 million yuan, down from about 84.8 million a year before.
The stated reason is plain in the document: R&D and sales expenses jumped sharply. Revenue is climbing, profit is shrinking. That is the signature of growth being bought, not banked.
This matters because a 200-times-earnings multiple is a bet on operating leverage — the assumption that at scale, costs stop rising faster than revenue and the margin snaps back. Unitree's most recent quarter shows the opposite motion.
The company's own half-year guidance concedes the point while trying to soften it: adjusted profit for the first six months is expected to fall somewhere between roughly 7 and 22 percent year-on-year — a narrower decline than Q1, but still a decline.
Then, days before the listing, came the demo.
Unitree unveiled a machine it called "Superman," which it said it built in "just over three months." It clears a two-meter standing jump and hits a top speed of 12.66 meters per second — faster than the 12.42 Usain Bolt recorded during his 2009 world-record 100 meters.
It is a genuinely astonishing piece of engineering, and it dropped to coincide with both the IPO and the opening of the World Robot Conference in Beijing. The timing was not an accident.
Read the fine print, though, and Unitree says Superman "remained a work in progress," with "significant room for further improvement in the coming months."
Now set that beside the other admission in the prospectus — the one that got far less airtime. Unitree concedes that large-scale commercial adoption could come more slowly than expected, because robotic hands still lack the precision and durability needed for sustained work.
So the same filing that markets a robot faster than the fastest human on Earth also states that its robots can't yet reliably do sustained manual labor. A humanoid that can out-sprint Bolt but can't hold a wrench through a shift is a demo reel, not a workforce.
That is the split screen. The stock is pricing the jump. The business runs on the hands. And the hands, per the company, aren't there.
Which forces the real question: what is actually driving the 629 percent?
The honest answer is scarcity. Unitree is the first humanoid maker to list on mainland China's STAR Market — the only pure-play ticket in a category everyone wants exposure to. The retail tranche was oversubscribed by more than 5,000 times, an allocation rate of 0.018 percent.
When demand for a theme outruns the supply of stock by that margin, the price stops describing the business and starts describing the shortage.
And the shortage is temporary. Caixin reports Unitree's debut is expected to set the valuation benchmark for 30 to 50 peers preparing listings in Hong Kong. AgiBot, Leju, Deep Robotics, Robotera — the pipeline is already forming.
Every one of those IPOs adds supply to the scarcity trade currently holding Unitree's multiple aloft. The benchmark Unitree just set is the same benchmark that dilutes it.
Which brings us to the one thing that could turn a scarcity premium into a durable one: the brain.
Unitree earmarked about 4.2 billion yuan of the proceeds for embodied-AI model development, research, new products and manufacturing. And it locked in DeepSeek as a strategic investor — a 2.31 percent stake with a three-year lockup, plus a collaboration to build embodied-intelligence models. Tencent-linked money came in on similar terms.
The logic is coherent. As one investor put it, the long-term value depends on whether Unitree can translate its hardware lead into AI capabilities; the money is chasing the "brain," not the body.
The three-year DeepSeek lock is the tell that someone in the deal understands the timeline. Locking a marquee AI partner in place for three years is an admission that the brain premium doesn't exist yet — it has to be built, and built before the peer flood arrives.
So the wager underneath the 629 percent is precise, even if the price isn't. It is a bet that Unitree's manufacturing lead buys enough time and data for a DeepSeek-fueled software moat to form, and that the moat closes before 30-to-50 competitors turn the scarcity trade into a commodity trade.
Nothing in the current numbers proves that bet wins. The financials show growth being spent, not compounded. The flagship demo is faster than a human and, by the company's own word, still can't do the work.
The listing is resolved. The manufacturing lead is real. But a lead in building machines is not the same as a lead in the models that make them useful — and it is the models, now, that the price demands.
For one week, the market got to buy the jump. Eventually it has to buy the hands. Those are very different numbers, and only one of them is on the tape.