China's humanoid boom gets its first market test: Unitree gives back ~45% as XPeng's robot unit raises $900 million.

The first market-set price for a pure-play humanoid maker is already deflating. After Unitree opened its August 19 Shanghai debut up 629% and closed up 460% — briefly worth about 445 billion yuan, or roughly $66 billion — its shares fell 18.7% the next day to 687 yuan, and kept sliding to 603.08 yuan by August 24, down about 45% from the debut peak and erasing more than 200 billion yuan of market value in four sessions. Even after the drop the stock still trades at roughly four times its 150.80-yuan IPO price. The reversal has less to do with the business than with the price it was handed: Unitree's 2025 revenue quadrupled to 1.7 billion yuan and the company is profitable, but its adjusted first-quarter 2026 profit had already fallen about 53% as costs rose, and quarterly revenue growth has slowed sharply from a year earlier.

What made the sell-off notable is that Unitree's own founder supplied the reality check. Addressing the World Robot Conference the day after the surge, Wang Xingxing — poker-faced at his own bell-ringing ceremony, an image that circulated widely online — acknowledged that humanoids are not yet as efficient as humans and still struggle to generalize across tasks, the industry's central bottleneck: success rates that approach 100% in a fixed setting fall sharply the moment the environment changes. He framed the field's 'ChatGPT moment' — a robot dropped into an unfamiliar home that completes about 80% of tasks with no setup — as perhaps two to three years away, a more measured horizon than the humanoid marketing cycle implies. HSBC had warned before the listing that the recent surge in humanoid shipments could be hard to sustain without real gains in the AI models that run the machines, and one government-tied expert, Pan Helin, told the Global Times the true boom could be a year off or a decade off, adding that 'the real test has just begun.'

Yet private money kept pouring in. On August 24, XPeng said its robotics business had raised more than $900 million at a post-money valuation of over $6.3 billion — the largest single-round private financing yet in China's embodied-AI industry — led by IDG Capital, with Tencent and Alibaba as strategic investors. The round gives the EV maker's humanoid unit its own market price for the first time, and the pitch is explicitly about manufacturing: XPeng says it will bring 'automotive-grade quality standards and large-scale mass production' to its IRON humanoid — 76 degrees of freedom in the body and 21 per hand, with three in-house Turing chips rated up to 2,250 TOPS running its physical-AI model on-device — and push it into mass production by the end of 2026, starting in its own stores and campuses before customer deliveries in China and overseas in 2027. Notably, XPeng claims IRON performs complex tasks 'without remote operation,' a pointed contrast with rivals whose demos are often teleoperated — a claim worth checking once the robot is in customers' hands rather than on a stage.

For a buyer, the week's two headlines cancel out in a useful way. A stock that quadrupled off its IPO and a $6.3 billion private round both measure investor appetite, not product readiness — and the most instructive data point was not a valuation at all, but Unitree's founder naming generalization as unsolved and putting the payoff years out. XPeng's IRON is a target, not something you can order today; Unitree's shipped robots still go mostly to labs and classrooms; and both, as Chinese-made machines, are blocked from new US sale by the FCC's July import ban. The signal to act on is the repricing itself: the market is shifting its attention from spectacle to execution, margins and real deployment, and so should any procurement team. Judge these companies by unit economics and dated, in-facility results — not by a first-day pop or a record funding round.