SHANGHAI, 18 September 2026. The past week, one of the most debated topics in finance circles has been whether IPO review for humanoid-robot startups is tightening. The Information reported that Chinese securities regulators are tightening approvals for humanoid robot startups seeking listings, prompted by Unitree’s first-day volatility and widespread copycat behaviour in the industry, raising the bar. The market then circulated a more specific claim: a company must show the ability to generate recurring revenue, or be on a track of narrowing losses, or present genuine technological innovation before its application is even considered.
Yet several brokerage bankers told Yicai the review is not really tightening, it has always held this standard, just stricter on IPO quality. This is no sudden new policy, only an old review logic wrapped in the “humanoid” buzzword that instantly ignited attention. So why the focus now? The reason is direct. On 19 August Unitree listed on the STAR Market, opening at RMB 1,100 with a market cap above RMB 440 billion, closing at RMB 845. By 14 September it had first fallen below RMB 500, touching a low of RMB 465.58, down 57.27 per cent from its peak. Over half the market cap evaporated in little more than a month, making regulators more cautious on the queue behind it.
The queued companies differ sharply in fundamentals. Already listed Unitree posted 2025 revenue of RMB 1.699 billion, net profit of RMB 591 million, over 5,500 humanoids shipped and about 32 per cent global share. Revenue grew from RMB 159 million in 2023 to RMB 1.699 billion in 2025, roughly 11 times. But 2026 first-quarter net profit fell 52.55 per cent year on year, and the first half is guided down 6.43 to 21.97 per cent. Growth is shifting gears, a fact.
UBTECH, the “first humanoid stock”, posted 2025 revenue of RMB 2.001 billion, up 53.29 per cent, with a net loss of RMB 703 million narrowed from RMB 1.124 billion, a 37.4 per cent improvement, and gross margin up from 28.7 to 37.7 per cent. Revenue up, loss narrowing, at least the direction is right. Deep Robotics sits in a trickier spot: 2025 revenue RMB 337 million, up 227 per cent, turned profitable, but the SSE asked a pointed question, its DR-series humanoids sold only four units combined in 2024 and 2025, so why list as a “humanoid” company on the STAR Market? Those four units swung margin from minus 31.32 to 68.74 per cent, raising how to verify single-unit revenue.
Lejv’s problem is more visible. 2025 revenue RMB 258 million, up 365.2 per cent, but net loss widened from RMB 41.12 million in 2023 to RMB 71.25 million in 2025. Revenue rockets while losses widen, two curves running opposite ways. Dobot posted 2025 revenue RMB 492 million, up 31.7 per cent, net loss RMB 83.535 million, still unprofitable, but its collaborative robots have shipped over 100,000 units and formed a relatively mature commercial loop in industry.
Why is commercialisation so hard? The deep contradiction of the industry. Global humanoid shipments in 2025 were about 13,000 units, all top six from China, yet the sector is still early, with most products serving as “embodied carriers” and “data-generation platforms”. Wang Xingxing put it bluntly at WRC: robots have not mass-entered factories because efficiency lags people and generalisation is weak. A perfect demo can be debugged, but move to another store or line and success collapses. The technology route has not converged, with the VLA versus WAM debate exposing clear shortcomings in real physics. Shen Yujun of Ant Lingbo noted the route war was really about everyone wanting a shortcut. Per MIIT, China had over 140 humanoid robot makers and over 330 models by 2025, many homogeneous, few with recurring revenue.
Primary-market heat has not cooled. In the first half of 2026 domestic embodied-AI financing totalled about RMB 93.5 billion, roughly five times year on year, with 22 firms above RMB 10 billion valuation. But the secondary robotics sector saw a max drawdown near 30 per cent this year with continuous net outflows. Primary market floods in, secondary market cools and retreats. The split itself shows capital has far from reached consensus.
The stricter IPO review is not bad for the industry. Three clear standards: revenue, narrowing loss, or real innovation, meeting one advances, not all three. Unitree and Mech-Mind faced no challenge queries because the fundamentals were there. Deep Robotics qualified on recurring revenue plus profit, still in inquiry not rejected. Lejv, flawed on the first two, bets on the third, core technological innovation. This is where divergence begins. For investors, the long-term direction is most likely sound, China’s edge in hardware manufacturing, supply-chain coordination and cost control is clear. China’s humanoid market is projected at about RMB 15 billion in 2026 with 2027 growth above 60 per cent. But a correct direction does not mean every company is investable. As IPO review scrutinises revenue quality and commercial loops, demo-and-PPT firms will struggle, which is exactly the mark of a maturing market. After the bubble is squeezed, what remains are the companies that can actually run the commercial game.
Editor’s note: This is an adapted translation of the original OFweek Robotics report. It has been trimmed and restructured for readability for an international business audience.