From a spectacular debut to a real-world question

In eight trading days, the share price fell from RMB 1,100 to RMB 585. This was not the collapse of an obscure speculative stock. It was Unitree Robotics, listed under the halo of being the first humanoid-robot stock.

On its first day, Unitree opened at RMB 1,100 and its market value passed RMB 444.9 billion. A successful IPO allocation was briefly worth RMB 475,000 in unrealized profit. With 85.28% turnover that day, much of the stock quickly moved from IPO winners to buyers entering at the peak. Eight days later, the price had nearly halved and more than RMB 200 billion in market value had disappeared.

The question is not whether Unitree's robots can do a backflip. It is when they will learn to work.

01 / Why strong results did not prevent the fall

In 2025, Unitree shipped 5,511 humanoid robots—first globally—and reported RMB 1.699 billion in revenue, up 332%. Those numbers helped support the opening-day valuation.

By the first half of 2026, however, growth had slowed from 332% to 48.5%, while adjusted net profit fell 19.34% year on year. The high-growth narrative was losing momentum.

The market's message was straightforward: it was no longer willing to pay an unlimited premium for a distant story. Nomura's target price of RMB 370 sat far below a stock still trading at more than 400 times forward earnings. The space between those numbers represented the thickness of the bubble.

02 / The arena and the factory use different scorecards

Unitree's hardware is impressive. In an arena, its robots can perform backflips and boxing combinations with striking fluidity. In a factory, reliably tightening a single screw remains a different class of problem.

An arena rewards peak motion capability. A factory requires long-duration, stable and precise repetition. Gemini Robotics 2 reported sobering results in fine manipulation: a 36% success rate for screwing in a light bulb and 44% for tying a trash bag.

The deeper gap is data. The field has roughly 500,000 hours of high-quality training data, while broadly capable physical intelligence may require closer to 100 million hours—a two-hundred-fold gap. Unitree founder Wang Xingxing has said household deployment could take two to three years at the fastest, or five to ten years at the slower end. If the builders themselves are patient, why should the market price certainty today?

03 / The warning behind the demonstration

Stanford professor Fei-Fei Li warned that publicity often treats prototype demonstrations as if they were mature products and presents an overly optimistic commercialization timeline.

The next frontier of AI extends beyond language into spatial and physical intelligence. But unstructured environments combined with a general-purpose body remain among the hardest problems. For a particular task, a humanoid form may not even be the best answer; a more specialized body can be better suited to narrower work.

Unitree's own prospectus illustrates the point. Universities and research institutes accounted for 73.6% of humanoid-robot revenue, commercial consumer use for 17.39%, and industrial scenarios for only 9.01%. Today, the product is still more often a research instrument than a factory tool.

04 / RobotBay judgment: learn to work before claiming generality

Unitree's sell-off is not the end of the robotics story. It is the first time the story has been forced through a financial statement—from doing backflips to accounting for cost.

Galbot did not insist on a fully humanoid body. Its wheeled base and humanoid upper body achieved more than 95% picking success in pharmacies, with a pick completed in about one minute. Astribot's T1, starting at RMB 89,900, similarly targets more focused scenarios.

China shipped more than 40,000 humanoid robots in the first half of 2026, representing 97% of global volume. Scale is real, but production is not the same as commercialization, and shipment is not the same as profit. The eventual winner may not be the robot that runs fastest, but the company that first turns reliability, cost and a closed-loop use case into a standard product.

Closing / Who pays for the bubble?

The problem is neither Unitree nor the direction of humanoid robotics. It is pricing a story that will take years to validate at more than 400 times earnings in a single day.

Wang Xingxing reportedly responded to the opening-day market frenzy with two words: “I'm sorry.” It may have been the most candid line in the entire episode.

If you work on the front line—in a factory, warehouse or laboratory—watching a robot learn how to do a real job, RobotBay wants to hear from you. Bubbles burst. The small, measurable improvements happening in the real world are what deserve a durable record.