At a 36Kr WAVES 2026 panel, China's humanoid robot founders and investors split the sector's revenue into novelty event rentals versus durable 24/7 factory work, and conceded the second lane is still thin — at a Chinese tech industry conference in 2026.
The cleanest test for whether China's humanoid-robot industry has become a real business is not how much money it has raised. It is whether any of its machines are doing the same work, for the same paying customer, week after week, at a price the customer can explain to a finance director. At a 36Kr WAVES 2026 roundtable, six of the country's most visible embodied-AI founders and investors admitted, sometimes in the same breath, that the honest answer is mostly no.
The panel drew a sharp line between two kinds of money. The first is what Yunshi Capital's representative called "the short-term window": renting a humanoid robot for a few hours at a brand event, livestream, or staged combat show, the way a luxury brand rents a vintage Rolls-Royce for a product launch. Pricing can hit tens of thousands of yuan per appearance, a margin that looks excellent on a quarterly slide. The second is "real, durable revenue," where a robot does a specific industrial task on a paying customer's site for months at a time, and the customer can defend the spend to its own board on unit-economics grounds rather than novelty grounds. The panel spent more time describing the first lane than the second, and the panelists did not pretend that was an accident.
That gap matters because the sector's headline valuations and revenue targets are calibrated against the second lane. Agibot, one of China's best-funded humanoid-robot makers, has publicly targeted $1.4 billion in revenue by 2027 on an "ecosystem expansion" plan that includes a heavy push into industrial use cases (Yicai Global). The company has also begun moving down the stack, investing in Nexforce for core actuator research and development, the part of a robot that turns battery power into joint motion (Gasgoo Auto News). Vertical integration is consistent with a company trying to make the unit economics work, but it is also a tell that off-the-shelf parts have not yet closed the cost gap.
The deployment cases the panel kept returning to were mostly in the short-term window. Zhongqing, a Shenzhen-based humanoid maker, has built a visible brand around staged robot-combat events, a use case that pays the bills, generates viral footage, and produces exactly the kind of structured data the company's models need to learn from. A Zhongqing-aligned panelist framed combat as both a commercial product and a research-and-development flywheel. The category order he preferred, geek or "Maker" market first, then Pet IP, then combat, then enterprise R&D-side demand, inverts the usual industrial-robotics pyramid on its head. The base of his pyramid is entertainment, not manufacturing.
Qingtianzu, the robot-rental operator, sits squarely in the same window. According to secondary reporting cited on the panel, the company has raised "hundreds of millions of yuan" across A and A-plus rounds and is being valued at 70 billion yuan, or roughly 10 billion USD at current exchange rates (Sina Finance, chnfund). The model is to own the fleet and rent robots to short-tenor buyers: trade shows, weddings, mall openings, livestream studios. Investors reading that as a sustainable business are implicitly betting that the rental fleet keeps finding buyers after the novelty premium fades. The panelists did not argue that bet, but they did not defend it either.
What they did defend, repeatedly, was a small set of industrial niches where the unit economics already pencil out. Yunshi Capital's representative listed three: shipyards, where welding and grit-blasting robots can run in environments human workers increasingly refuse; underwater inspection, where a tethered robot replaces a dive team; and heavy-industry factories where the labor shortage is structural rather than cyclical. In each case the customer is not paying for a humanoid form factor. It is paying for a task that has become too expensive, too dangerous, or too hard to staff any other way.
The shape of the disagreement among panelists was as informative as the points they agreed on. Several founders pushed back on what they called the "physical AI" or "general humanoid" framing, arguing that the right near-term product is a single-purpose machine that does one industrial task well, with a humanoid body only if the customer's existing factory layout requires it. Others held out for the general-purpose bet on the grounds that data collected from staged combat and rental fleets will, over years, train the foundation models that eventually make general humanoids viable. Both sides were right about their own time horizon and arguably wrong about the other's. The panel host did not invite an outside skeptic to test either view.
The falsifier for the whole thesis is straightforward, and the panelists named it themselves: a high-profile humanoid-robot startup that raises at a unicorn valuation, signs a marquee industrial customer, and quietly lets that contract lapse when the renewal comes up. If that pattern starts to show up in 2026 results, the short-term window was the whole business. If instead the same companies start converting event-rental customers into 24/7 industrial buyers, the lane split closes. Until then, China's embodied-AI sector is, by its own insiders' admission, more Rolls-Royce rental than factory floor.