As reported by Caixin Global, Forbes and others.
Caixin reports that Unitree Robotics is pricing a Shanghai STAR Market IPO that will, for the first time, give public markets a profitable, high-volume humanoid maker to value — and it is Chinese.
What Happened
According to Caixin, Unitree Robotics — the Hangzhou company that ships more humanoids than any other manufacturer on earth — is in active book-building for a Shanghai STAR Market IPO, with public subscription opening August 10. The offering raises approximately 4.2 billion yuan (~$623 million) by selling roughly 10% of the company, implying a valuation floor near 42 billion yuan (~$6.2 billion). Banks are modeling a potential re-rate toward ~$15 billion, though that ceiling is speculative: the IPO is being priced, not closed, and a range that wide means the tape will decide.
The financials are the anomaly. Unitree posted approximately $235 million in 2025 revenue at roughly 60% gross margins and shipped around 5,500 humanoid units in the year — making it profitable at a moment when every other high-profile player in the space is burning cash against a promise. The caveats are real and material: most of those 5,500 units are lower-priced research, education, and quadruped robots (average selling price around $23,000), not the general-purpose labor humanoids that underwrite the industry’s largest valuations. A $16,000 research humanoid and a $1,600 robot dog are genuine products. They are not the same bet as a warehouse floor worker.
Beijing fast-tracked regulatory approval in a record 104 days, treating Unitree as a strategic champion. Its most-cited American rival, Figure AI, carries a reported ~$39 billion valuation with essentially no revenue. Those two numbers sit in the same conversation only because both companies call their products humanoid robots. In practice, they are pricing different futures — which is the structural point worth sitting with.
The key insight: For the first time, public markets have a profitable, high-volume humanoid manufacturer to price. The comparison with Figure AI is not a paradox — it is two different bets on two different time horizons. Unitree is priced for what robots sell for today. Figure is priced for what robots might replace in ten years. Both could be right. Both could be wrong. The structural shift is that one of them is already profitable and Chinese.
The Structural Read
The Unitree IPO does something no funding round or product launch has done before: it forces a public valuation on the question of who leads in humanoid robotics. And the answer it gives is uncomfortable for the narrative that American frontier AI labs are running this race.
Two philosophies are now priced in the same market. The first — call it priced for profit — is Unitree’s model: cheap hardware, high volume, real margins, built on the same manufacturing base that has made China the cost leader in commodity DRAM and AI server assembly. The second — priced for promise — is Figure’s model: a moonshot valuation underwritten by a bet on general-purpose labor humanoids that are years from volume deployment in a market that remains unproven.
This is the commoditization barbell that has defined this entire AI cycle — cheap-and-profitable at the floor, speculative-and-frontier at the top — now expressing itself in embodiment rather than models or memory. DeepSeek versus GPT-4o was the barbell at the model layer. CXMT versus Micron is the barbell at the memory layer. Unitree versus Figure is the barbell at the physical layer. The pattern is consistent enough to be a structural feature, not a coincidence.
The Commoditization Barbell — Embodiment Layer
Cheap-and-profitable vs. speculative-and-frontier
At the model layer: DeepSeek vs. GPT-4. At the memory layer: CXMT vs. Micron. At the physical layer: Unitree vs. Figure. In each case, a Chinese manufacturer captures the high-volume, lower-ASP, profitable floor while American players hold the high-valuation, low-revenue frontier. The question the barbell asks is always the same: which end of it scales faster?
The strategic subtext is the part that deserves the most attention. Beijing’s 104-day fast-track is not an administrative footnote. It signals that the Chinese government is treating robot manufacturers as strategic champions in the same category as chip fabs and memory producers. The logic is coherent: if the next layer of AI is physical — if the value of foundation models ultimately expresses itself through embodied agents doing physical labor — then the country that can manufacture robots cheaply, at volume, and at a profit holds a different kind of structural advantage than the one that can raise the most venture capital. As explored in The Map of AI Redrawn, the race for AI supremacy is not only a software race. It is increasingly a manufacturing race.
One important caveat holds throughout: the STAR Market listing is retail-driven, volatile, and largely closed to foreign capital. The price Unitree discovers on August 10 is a Chinese price, not a global comparable. The $6.2 billion floor and the ~$15 billion bank ceiling will not trade in the same pool as a Nasdaq listing. What it sets is a precedent and a reference point — not a universally arbitrageable benchmark. That matters for how seriously to take the Figure comparison as a direct valuation gap. It is more useful as a philosophical gap than a financial one.
Beyond NVIDIA’s Moat — Business Engineer
“The companies that define the physical layer of AI may not be the ones that win the intelligence race. They may be the ones that win the manufacturing race — and those two competitions have different leaders.”
Three Implications
IMPLICATION 1 — The Benchmark Problem for Western Robotics Valuations
Even as a Chinese-market price, Unitree’s listing injects a real data point into a conversation that has been running entirely on private valuations and press releases. Figure AI’s ~$39 billion is a bet on a far larger and later market than anything Unitree sells today — that comparison is not clean — but any investor underwriting a frontier humanoid valuation now has to explain, at minimum, why the profitable, high-volume floor is not a ceiling on the near-term market. That is a harder conversation than it was before August 10.
IMPLICATION 2 — China’s Physical-AI Supply Chain Is Now Three Layers Deep
CXMT in commodity DRAM, Foxconn and Flex in AI server assembly, and now Unitree in humanoid hardware. China is building a vertically integrated cost-leadership position across the physical infrastructure of AI — compute memory, server chassis, and embodied agents. Each layer reinforces the others. A robotics manufacturer that can source cheap actuators, cheap chips, and cheap assembly from domestic supply chains is structurally harder to displace than one competing on software alone. Beijing’s fast-track approval is the policy layer of a strategy that already has the manufacturing layer in place.
IMPLICATION 3 — The Humanoid Labor Market Is Still Unproven; Both Bets Could Lose
The framing of Unitree as the disciplined realist and Figure as the reckless moonshot flatters the contrast. Unitree’s profitability rests on a $235 million revenue base built on research units and robot dogs — not on displacing factory workers at scale. Figure’s valuation rests on a general-purpose labor humanoid market that does not yet exist at commercial volume. Both bets carry existential risk. The durable takeaway is narrower than a winner-picks narrative: for the first time, you can see what a profitable humanoid business looks like, and it is smaller and more hardware-specific than the labor-replacement thesis requires. That is useful information regardless of who ultimately wins.
