The World Robot Conference had barely concluded when Unitree Technology, which had soared immediately after its listing, experienced another dramatic stock price collapse. On August 24, Unitree's shares plunged over 10% at the market open, dragging down the entire robotics sector. Despite multiple intraday recovery attempts, the stock ultimately closed down 10.31% at 603.08 yuan per share. Since its debut on August 19, Unitree's market value has evaporated by 100 billion yuan in just three trading days. Calculated from its peak price of 1,100 yuan, the stock has now retreated over 45%, and compared to its initial market value of 440 billion yuan at listing, it has lost 200 billion yuan.
Beyond the overall market decline, many believe a weekend article from Huxiu claiming that 80% of embodied intelligence orders might be fake directly spooked the secondary market. This article, which exposed the illusion of industry prosperity, was merely a trigger. Doubts about Unitree have persisted in the market, whether it's the criticism of "only performing, unable to work in factories," the fact that over 70% of orders come from scientific research and education, the meager R&D spending that is even less than MUYUAN, a pig farming company, or questions about whether Unitree's profitability and orders can be sustained. These are all lingering concerns surrounding Unitree. Reflecting on the listing day, Wang Xingxing's solemn expression may have signaled that he already foresaw this valuation collapse. Unitree Technology faced a stress test the moment it went public.
Are 80% of Orders Fake? A Huxiu Article Shakes Unitree
How can one article cause the robotics sector to collapse and Unitree to hit its daily limit down? It directly challenged the embodied intelligence industry, exposing its illusory prosperity: it claimed that 80% of embodied intelligence orders in the industry might be fake. By January 2026, domestic humanoid robot意向 orders had surpassed 120,000 units, but IDC statistics show that actual global humanoid robot shipments last year were only about 18,000 units—a gap of a full 100,000 units. Moreover, the vast majority of these orders are for performance, scientific research, and education scenarios, not for factory use. This data directly punctured the bubble of booming embodied intelligence robot orders. The second part of the data, "mostly flowing to performance, scientific research, and education," pointed directly at Unitree. The market has long questioned humanoid robots as "more flashy than practical," especially Unitree, which gained fame through several Spring Festival Gala performances. Previously, even TSMC's chairman criticized robots like Unitree, saying, "Always making robots jump and bounce around is useless; it's just for show." IPO prospectus data shows that in 2025, Unitree shipped 5,500 humanoid robots, ranking first globally. However, about 70% of Unitree's revenue comes from scientific research and education, with industrial scenarios accounting for only 9%. At the recently concluded robot conference, a subtle scene drew attention from both inside and outside the industry: Unitree, dancing, was surrounded by robot manufacturers who were "working." Against this backdrop, the Huxiu article, which seemed to question other non-headline companies in the industry, once again reignited external doubts about Unitree's commercial prospects.
Unitree Loses Top Shipping Spot to Agibot, Raising Questions About Next-Stage Growth
The core reason for Unitree's sustained stock price plunge is the market's disagreement over its next-stage growth potential: a Unitree that can only dance and box, cannot enter factories, and relies on scientific research for 70% of its orders—how can it continue to sell? Changes in sales volume have sparked market concerns. In the first half of this year, Agibot Robot surpassed Unitree with 8,400 units shipped, compared to Unitree's 5,900 units, clinching the top spot in global humanoid robot shipments. Although both are leading embodied intelligence companies, Agibot and Unitree have vastly different order structures: Unitree derives about 70% of its revenue from scientific research and education, with industrial scenarios at just 9%; Agibot, on the other hand, primarily targets industrial scenarios like automotive factories and 3C production lines. Generally, industrial orders are more replicable and sustainable, and shipment volumes tend to be higher than research orders. For example, in the first half of this year, Agibot's year-over-year growth rate was 562%, while Unitree's was 170%—a difference of more than three times. This is not just a gap in orders; it directly relates to differences in scenario data training and accumulation. Simply put, it's the ability to "evolve." Robots working in factories not only generate revenue but also allow robots to accumulate data in real-world scenarios, using a "data flywheel" to enhance their multi-scenario application capabilities. If this growth trend continues, the gap between the two will widen further. Unitree Technology founder Wang Xingxing also admitted in his WRC speech that he didn't have his own robots work in factories because the "brain" isn't smart enough yet. Now, Agibot, which is working in factories, has surpassed Unitree in shipments, while Unitree, still "performing," has yet to find its "brain" and broader industrial scenarios. In this race, Unitree is beginning to slow down.
Unitree Invests 2 Billion in R&D for the "Brain," But World Model Questioned as "Abstract" and "Money-Making Scheme"
In the Huxiu article, multiple investors questioned the world model concept as "too abstract" and "a money-making scheme": "90% of teams working on world models don't even believe in it themselves." The world model is the biggest trend in the embodied intelligence industry this year. It became a trend partly because it is "more cutting-edge than the 'embodied brain' and harder to disprove"—a perfect to-VC concept. Unitree's 2 billion yuan investment in R&D for the "brain" also includes the world model. Previously, on August 7, during Unitree's pre-IPO online roadshow, Wang Xingxing revealed that the company would concurrently develop the WMA World Model and VLA visual-language-action embodied large model. Of Unitree's fundraising proceeds, over 2 billion yuan will be used for R&D of embodied intelligence large models and core software and hardware. This expenditure far exceeds the total R&D spending in Unitree's entire history. It's worth noting that Unitree's R&D spending has always been low, far below its peers, to the point of being criticized as "R&D spending less than a pig farmer." In 2025, Unitree's R&D expenses were 145 million yuan, while MUYUAN's R&D expenses were 1.648 billion yuan during the same period. In terms of absolute amount, MUYUAN's R&D spending is 11 times that of Unitree. However, relative to overall company revenue, Unitree's R&D expense ratio of 8.53% is much higher than MUYUAN's. But compared to its peers, Unitree's 8.53% R&D expense ratio is still quite low, not only below listed peers like UBTech and Dobot, but also below the industry average of 27.92%. Unitree's cumulative R&D spending over the past three years was only 265 million yuan. Now, it's investing 2 billion yuan to build a "brain" and transform into a "software-hardware integrated platform," shifting from "selling bodies" to "building brains." However, shortly after announcing the parallel development of the world model, skepticism emerged in the industry about the world model being "abstract and a money-making scheme." This has also raised concerns about Unitree's "transformation" story in the capital markets.
Conclusion
There's no denying that Unitree is a good company. But a good company and a good stock are two different things. When the market begins to question the quality of orders across the entire track, when the "world model" is dismissed by investors as fundraising jargon, and when a robot company's R&D investment can't even match that of a pig farming enterprise—Unitree faces a full-fledged stress test. A robot that can dance must eventually enter factories to prove its value. A market value and stock price built on narratives must ultimately be supported by real orders. Unitree took three years to grow its revenue from 159 million yuan to 1.7 billion yuan. How long will it take to bring its industrial scenario orders, currently under 10%, to a level comparable with its scientific research and education orders? The market may not give it as much time as it thinks.
Comments