The clearest signal that an industry has left the demonstration phase is when it has to publish quarterly numbers.
In August 2026, that happened twice over in China's AI and robotics sector.
Unitree Robotics listed on the STAR Market on 19 August 2026, roughly a week after disclosing cumulative production of more than 18,000 bipedal humanoids. Its IPO application had been accepted on 20 March 2026 and the company raised about 4.2 billion yuan. The prospectus revealed a business most observers had assumed did not yet exist: revenue rising from 159 million yuan in 2023 to 1.708 billion yuan in 2025, with net profit turning positive at 288 million yuan in 2025.
Kling, Kuaishou's AI video business, was reported to have opened a pre-IPO round at a pre-money valuation of about US$18 billion, with the company working toward a Hong Kong listing filing in early 2027. Kuaishou confirmed in a Hong Kong stock exchange filing on 12 May 2026 that its board was evaluating a restructuring of the Kling business, potentially involving external financing.
Key takeaways
- Unitree (STAR Market, 19 Aug 2026): raised about 4.2 billion yuan; 2025 revenue 1.708 billion yuan, net profit 288 million yuan, versus a loss in 2023; cumulative humanoid production past 18,000 units.
- Kling (pre-IPO): reported valuation around US$18 billion; Q2 2026 revenue above 850 million yuan, up more than 200% year on year.
- Sector scale: more than 6,600 AI companies in China as of June 2026, about 15% of the global total.
- Hardware momentum: 400+ domestic humanoid models, over half the world's total, with Chinese humanoid and quadruped robots at roughly 80% of global sales by volume.
- Conference signal: at WAIC 2026, 177 international procurement delegations with intended purchase value of about 20.36 billion yuan, up roughly 25% year on year.
What a profitable robotics company proves
Humanoid robotics has been funded for years on narrative. A listed company has to explain itself differently: units produced, units delivered, gross margin, and what customers actually do with the machines.
Unitree's filing answered the distribution question with an honest and slightly deflating detail: more than 70% of its humanoid customers are universities, research institutes and cultural-tourism or performance venues. The volume driver is the G1 education variant, priced roughly between 145,000 and 414,000 yuan. Factory deployment on production lines remains at pilot scale.
That is a real business, just not the one in the promotional videos. It is also the classic shape of an early platform market: research buyers fund the volume that drives the cost curve down, until the price reaches the point where industrial buyers enter.
The second thing the filing revealed is intent. More than 2 billion yuan of IPO proceeds is directed at intelligent robot model research, pursuing world models and vision-language-action models in parallel. Hardware is treated as solved; the deficit is intelligence.
The software side has better economics
Kling's numbers are healthier still, because software economics are different: no actuators, no inventory, marginal cost approaching zero. Revenue growing more than 200% year on year at a quarterly run-rate above 850 million yuan, with analysts projecting annualised revenue beyond US$1 billion by end-2026, is a growth profile most listed software companies would envy.
Goldman Sachs, maintaining a buy rating on Kuaishou in June 2026, described Kling's competitive position as a flywheel: model capability drives revenue, revenue funds compute, compute improves the model. That is the same argument made for frontier labs, with one difference — Kling's is already generating cash.
What could go wrong
Three risks are visible and worth stating:
Inventory versus delivery. Reports on Unitree note a gap between units produced and units delivered, with inventory absorbed by education and exhibition orders that run over long horizons. Production numbers are not revenue.
Competition is domestic. In video models, ByteDance's Seedance 2.0 is reported to hold a substantially larger share of the Chinese market than Kling. In robotics, AgiBot has reported roughly 15,000 cumulative units. These companies are competing with each other, not coasting.
Valuation versus delivery. An US$18 billion pre-money valuation on a business with a sub-billion-dollar annualised revenue is a bet on the next three years, not the last one. Analysts' sensitivity work using 10–25× price-to-ARR produces a very wide range for a reason.
The encouraging part is that all three risks are now discussable in public, with numbers attached. That is what markets are for.
Figures as disclosed in company filings, exchange announcements and 2026 financial reporting.