Unitree Hit ¥1.7B Revenue With Only 480 Employees

In February 2025, Wang Xingxing sat in the front row at a private symposium hosted by Xi Jinping — the only attendee born after 1990, flanked by Ren Zhengfei of Huawei, Jack Ma, and BYD's Wang Chuanfu. He gave a speech. Eleven months earlier, a London research firm had barely registered his company in global humanoid rankings. Last month, he hosted German Chancellor Friedrich Merz at his factory in Hangzhou. This week, Wang filed to take Unitree Robotics, the company he founded in a 50-square-meter office with roughly $12,000 in competition prize money, public on Shanghai's STAR Board at a ¥4.2 billion ($630 million) target raise.

The prospectus, filed March 20 with the Shanghai Stock Exchange and underwritten by CITIC Securities, contains numbers that don't look like a typical pre-IPO pitch deck. They look like a mature industrial company. Revenue hit ¥1.71 billion ($257 million) in 2025, up 335% year over year. Net income reached ¥600 million ($90 million), up 674%. Gross margin came in at 60.27% — a figure that, as Recode China AI's analysis noted, beats Apple's 47.3% and approaches Louis Vuitton territory. Operating cash flow tripled. And all of this with 480 employees — revenue per head of roughly ¥3.56 million ($534,000).

The headline claim is that Unitree shipped 5,500 humanoid robots in 2025, asserting 32.4% global market share and the #1 position worldwide. For context, Reuters and CNBC both reported the figure against the backdrop of Western competitors: Figure AI, Agility Robotics, and Tesla Optimus each shipped approximately 150 units in the same period, according to London-based consultancy Omdia. Tesla's Optimus missed its own 5,000-unit internal target for 2025. Elon Musk has said external sales won't begin until the end of 2027. The gap between China and the West in humanoid production — right now, today — is not a rounding error.

The #1 Claim Is Messier Than It Looks

But the "global #1" assertion deserves scrutiny before it gets laundered into received wisdom. In January 2026, Omdia published a report putting Shanghai-based AgiBot (also known as Zhiyuan Robotics) first, with 5,100+ units shipped and 39% global share — making Unitree second. Unitree pushed back publicly, issuing a statement claiming its 5,500-unit figure and challenging AgiBot's methodology. Caixin covered the public dispute directly: "Unitree Defends Robot Sales as Rival Claims Market Crown." By March, when the prospectus was filed, Unitree was officially claiming #1 again, citing "industry data" — but that data is not independently audited, and the discrepancy with Omdia's figures turns on definitional questions: what counts as "shipped," does it include test units, partner demos, internal use? Both companies likely moved between 4,200 and 5,500 humanoids in 2025. They're neck-and-neck. The prospectus's market share claim should carry an asterisk.

What's not contested: even if Unitree is second to AgiBot, both Chinese companies are operating in a different universe from everyone else. UBTECH, the Shenzhen-based humanoid maker, shipped around 1,000 units. The rest of the field — EngineAI, Figure AI, Agility, Tesla — are each in the low hundreds. The frontier of humanoid production is a Chinese civil war right now, and the US competitors are watching from a different altitude.

The Buried Fact About What These Robots Are Actually Doing

Here is the number that essentially no English-language coverage has run properly. Buried in the prospectus's breakdown of humanoid "industry application" revenue: enterprise reception and tour-guide use accounts for 50 to 70% of commercial humanoid deployments. Not factory automation. Not manufacturing. Lobbies. Corporate reception areas. Visitor centers where a G1 humanoid stands and greets guests.

The South China Morning Post cited the filing's revenue breakdown, and Recode China AI was among the few outlets to flag the enterprise tour-guide dominance as the real story. The robot that's supposed to usher in the era of embodied AI is, in most of its current commercial deployments, doing expensive receptionist work.

This isn't a fatal critique. First-generation commercial humanoid use cases have always been performance-adjacent — showing visitors that the technology exists, creating novelty in corporate environments, softening the public's relationship to these machines. But it matters enormously for anyone reading the 5,500-unit number as evidence that humanoid robots are already doing real industrial labor at scale. They're not. Not yet. The research and education market — universities, labs, developers — accounts for the bulk of the remainder, which is how Unitree got this far. It sold robots to people building with them, not to factories deploying them.

Unitree's own response to this tension: in February 2026, the company published a video of its G1 humanoid robots assembling robot parts at its own Hangzhou facility, using a new foundation model called UnifoLM-X1-0. HumanoidsDaly noted the video played at 2x speed and took place in Unitree's own controlled environment — not an "unfamiliar" industrial setting. Wang Xingxing has defined the "ChatGPT moment" for robotics as the point when a robot can complete 80% of tasks in 80% of unfamiliar environments. He has said they're not there yet. The factory video, impressive as it is, is a proof of concept filmed in a known environment. The gap to general deployment remains real.

The Price War and What It Actually Costs

Unitree's pricing strategy is philosophically different from every Western competitor and practically important for anyone thinking about the market. The G1 humanoid starts at $12,750 — a price point that generated enormous coverage when Wang announced it last year. But that locked-down base unit has no SDK and can't be reprogrammed. It exists for display and entertainment. The version developers can actually build on — the G1 EDU model — starts at $30,000 and runs to $60,000.

Compare that to Boston Dynamics' Spot, which retails around $74,500, or Agility Robotics' Digit, priced above $150,000 for enterprise contracts. Even the G1 EDU is dramatically cheaper than US competitors. The quadruped story is more extreme: Unitree's Go2 starts around $1,600 against Boston Dynamics' comparable Spot at $74,500 — a 46x price differential.

This is how the gross margin story and the pricing story coexist. Unitree is not competing on premium; it's competing on volume. Revenue per humanoid fell from ¥593,400 in 2023 (when five units sold) to ¥167,600 for the first nine months of 2025. A 72% price decline while gross margins expanded to 60.27%. That combination — achieved through vertical integration and supply chain density that China's manufacturing ecosystem makes possible — is the core competitive moat.

The EV Playbook

The comparison to China's electric vehicle industry is almost too obvious to make, which is precisely why it matters. China's EV sector went from minimal global presence to controlling a substantial portion of global output through a combination of aggressive domestic subsidy, supply chain control, price competition, and manufacturing scale-up — in a period that left Western incumbents scrambling. The pattern in humanoid robotics is tracking the same arc at roughly half the timeline.

China's 14th Five-Year Plan designated humanoid robots as a key strategic industry. Beijing has listed embodied AI alongside quantum computing, 6G, nuclear fusion, and brain-computer interfaces. Nearly 90% of all humanoid robots sold globally in 2025 were Chinese, according to Rest of World. Unitree, AgiBot, UBTECH, Leju Robotics, Deep Robotics, and Galbot are all at various stages of seeking public listings. The playbook is unified even if the companies are competing hard against each other.

The difference from EVs: China's humanoid advantage right now is heaviest in the research-and-education segment, not yet in industrial deployment. The export ceiling for humanoid robots is also unclear — US trade restrictions and geopolitical dynamics that have hit other Chinese hardware categories could theoretically apply here, and the prospectus is conspicuously silent on that risk.

The Founder Equation

Wang Xingxing is 35. He built his first quadruped robot — the XDog — as a master's thesis project at Shanghai University. It went viral online. An investor named Yin Fangming, then co-president of Roobo, gave him approximately $300,000 in angel funding. Wang spent two months at DJI in 2016, quit, and founded Unitree in Hangzhou with his winnings from robotics competitions and that first check. The company's registered capital at founding was ¥100,000 — about $13,800.

By June 2025, a private funding round — backed by Meituan, HongShan Capital, and Matrix Partners China — valued Unitree at ¥12.7 billion. The IPO seeks ¥4.2 billion in new capital, implying a market cap target in the neighborhood of ¥42 billion. That's a roughly 3.3x increase from the June valuation in nine months, a number the market will have opinions about.

Wang still serves as both CEO and chief technologist. He is personally involved in engineering decisions. He gave a speech to Xi Jinping about the future of robotics. He showed Friedrich Merz around the factory floor. He has 68.8% voting rights in the company, as disclosed in the prospectus — the IPO will not change who is making decisions.

What the Numbers Don't Say

R&D as a percentage of revenue has fallen from 31.4% in 2023 to 17.8% in 2024 to 7.7% in 2025. Bulls will note absolute R&D spending kept climbing: roughly ¥30 million, then ¥70 million, then approximately ¥90 million annualized. Bears will ask whether a 7.7% R&D ratio is sustainable for a company claiming to be at the frontier of embodied AI, especially as competitors close the gap and AI training costs rise. This is the number worth tracking in quarterly reports after the IPO.

The STAR Board acceptance — after preliminary review and two rounds of inquiry from the SSE — gives the filing legitimacy, but it doesn't guarantee final approval. Caixin reported that multiple Chinese robotics firms are queued for public listings, and regulatory approval can move slowly or stall. Leju Robotics and Deep Robotics are both seeking A-share IPOs; Galbot and AgiBot are eyeing Hong Kong. The prospectus is not a prospectus for a company that has already gone public.

The Lobby Problem and What Comes Next

The honest read of Unitree's moment: the company is real, the numbers are real, the manufacturing lead is real, and the ambition is real. The prospectus was prepared by Unitree to attract investment, and every figure in it served that purpose — which is not a reason to dismiss it, but is a reason to read it alongside Omdia's data and Caixin's reporting rather than in isolation.

What the filing establishes, at minimum: one Chinese company has figured out how to manufacture humanoid robots at a scale Western competitors are not approaching. It has done so while remaining profitable, maintaining exceptional margins, and growing faster than almost any hardware company in recent memory. The gap between 5,500 units and 190,000 annual capacity — the production target the IPO proceeds are partly meant to fund — is enormous. Closing it requires customers who want robots not to greet visitors, but to do something harder.

That transition, from novelty deployment to industrial utility, is where the real story lives. The lobby robots are a stepping stone. The question — the same question that applied to early EVs, early industrial robots, early smartphones — is how long the stepping stone takes to disappear underfoot.

Portions of this article draw on Chinese-language financial reporting translated and summarized by Recode China AI, Caixin Global, and the South China Morning Post. The Unitree prospectus PDF is available on the SSE disclosure portal.