On August 24, Unitree Robotics, often dubbed the "first humanoid robot stock," saw its share price continue its steep decline, with intraday losses at one point exceeding 10%. By the market close, the drop had settled at 10.31%, with the stock price at 603 yuan per share. While this still represents a gain of roughly 303% above the initial public offering price of 150.80 yuan, it marks a cumulative retreat of more than 44% from the first-day opening price of 1,100 yuan. The company's total market capitalization has now fallen below the 250 billion yuan mark, having evaporated nearly 100 billion yuan compared to its closing value on listing day.
The stock made its debut on the STAR Market of the A-share market on August 19, opening at 1,100 yuan per share, a 629.44% surge from the IPO price, pushing the company's valuation to 444.9 billion yuan. According to the prospectus, Wang Xingxing, the company's chairman, general manager, and chief technology officer, directly holds 86,714,964 shares, representing 21.4395% of the total share capital post-issuance. Prior to the offering, he also indirectly held a 9.5367% stake through the equity incentive platform Shanghai Yuyi. His combined direct and indirect holdings amount to approximately 30% of the company, with a market value exceeding 100 billion yuan, making him the new richest person born in the 1990s.
Retail investors who were fortunate enough to secure an allotment and sold on the first day would have netted a profit of 475,000 yuan per lot, based on 500 shares per lot at the 150.8 yuan issue price. However, that paper gain has now been nearly cut in half.
Share Price Inflated Beyond Fundamentals
On the evening of August 17, Unitree Robotics updated its interim results for the first half of 2026 in its listing announcement on the STAR Market. During the January to June period, the company generated revenue of 1.152 billion yuan, a 48.54% year-on-year increase. Net profit attributable to shareholders reached 274 million yuan, a turnaround from the loss of 32.0245 million yuan recorded in the same period last year. However, non-GAAP net profit came in at 244 million yuan, down 19.34% year-on-year. The company attributed this "revenue growth without profit growth" to a significant year-on-year increase in research and development expenses and selling expenses during the period.
More concerning is the deceleration in revenue growth. Unitree's revenue growth rate for the full year of 2025 was 332%, which slowed to 68.49% in the first quarter of 2026 and further to 48.54% in the first half. The company acknowledged that as its revenue base has expanded significantly, industry enthusiasm has cooled, and market competition has intensified, the year-on-year growth rate has declined compared to its compound annual growth rate.
At the issue price of 150.80 yuan per share, Unitree's post-offering market capitalization was approximately 60.993 billion yuan, corresponding to a price-to-earnings ratio of 219.23 times, far exceeding the average of 38 times for the general equipment industry. An analyst from Guoxin Securities noted in a research report that this valuation is not pricing in 2026 earnings alone, but rather a pre-emptive discounting of the long-term potential of the humanoid robot market, which is expected to reach trillions of yuan. The 219x P/E ratio implies a core assumption that Unitree must grow at a pace far exceeding the industry average for years to come to justify its current valuation.
On the same day the interim results were released, Unitree published a video showcasing its "Superman" humanoid robot. The robot, with 0.85-meter legs, can jump 2 meters in place and reach a top speed of 12.66 meters per second, surpassing human records. The company stated that the new machine was developed in just over three months and that significant room for improvement remains in the coming months. The release was widely interpreted as a "technology flex" ahead of the listing, further fueling market enthusiasm.
On the evening of August 10, the company announced that the final online allotment rate was 0.01809759%, less than 2 in 10,000, significantly below market expectations and making it one of the hardest STAR Market new stocks to secure an allotment in 2026. For context, the average allotment rate for STAR Market new stocks typically ranges from 0.03% to 0.08%, and even the high-priced new stock Pinzhun Laser had a rate of 0.0201%. The storage chip darling Changxin Technology had an allotment rate of 0.47%, meaning Unitree was 26 times harder to secure.
Despite this initial frenzy, market enthusiasm lasted just one day before the stock entered a three-day plunge. Pan Helin, a member of the Expert Committee on Information and Communication Economics under the Ministry of Industry and Information Technology, was blunt in his assessment: "The sharp decline in Unitree's stock is easy to understand: the share price was hyped too high and has become detached from fundamentals. In fact, the IPO price itself was already too high."
Pan further explained to China Newsweek: "The underperformance in applications is merely a symptom; what truly falls short of expectations is the development of the humanoid robot's 'brain.' Currently, humanoid robots have made decent progress in mechanical capabilities, able to run, spin, and jump, but these abilities do not meet market needs." He argued that the cost of achieving these capabilities is prohibitively high, and even when achieved, the capabilities do not necessarily translate into usefulness. The market requires humanoid robots to possess autonomy, which depends on the development of the robot's brain, a process that is inherently a long and slow industrial evolution. He cited autonomous driving as the simplest form of embodied intelligence, noting that even it has not fully adapted to driving environments, let alone humanoid robots facing complex operational conditions. "Without a brain, current humanoid robots cannot compete with industrial robots in factories, nor can they beat smart home appliances in households."
In his view, the humanoid robot industry currently faces two paths: either reduce the cost of humanoid robots without sacrificing performance, or enhance their autonomy, which requires coordinated advances in computing power, algorithms, and data, likely entailing a long and costly R&D journey.
Global Concerns on AI and Robotics
"Intelligence has always been a key focus for us. In recent years, I have spent the most time on robot AI and AI model training. People around the world are worried about this," Wang Xingxing admitted in a recent media interview. "Not just our company, but even major AI companies and large corporations are concerned about falling behind in the rapidly advancing global AI landscape." He said the question worth deep consideration is what kind of robot model can truly enable robots to perform general tasks across various scenarios, calling it "likely the most important and core issue for the coming years."
Explaining why robots still have a low success rate in performing tasks, Wang pointed to the fundamental mismatch between AI model outputs and the real world. "This is the biggest bottleneck globally right now. Once we break through, the ChatGPT moment for embodied intelligence will essentially arrive." He also emphasized that data collection and training methods must be better aligned with the model, noting that "for robots, the biggest problem is that the training model is not sufficiently aligned with the real world."
Behind Unitree, a large queue of robotics companies is preparing for public listings. According to incomplete statistics, more than 20 embodied intelligence companies, including CloudMinds and Leju Robotics, have announced IPO plans. Additionally, Agibot, Galaxy General, Fourier Intelligence, Zhongqing Robotics, Xinghitu, and Songyan Power have completed share reform, with the latest funding rounds of Galaxy General and Xinghitu widely viewed in the industry as pre-IPO financing. Magic Atom is also accelerating its timeline, with co-founder Gu Shitao revealing that the company could "potentially hit the secondary market as early as 2026."
Guoxin Securities analysts pointed out that Unitree's listing has established a valuation anchor for the industry. Currently, leading unlisted embodied intelligence companies are valued at 20 billion to 30 billion yuan, and Unitree's performance in the secondary market will serve as a key reference for their future fundraising and IPO pricing. On the other hand, industry consolidation is also accelerating. Xu Guangtan, chief machinery analyst at China Securities, previously told China Newsweek that while the industry is still in its early stages, relying solely on capacity expansion narratives no longer works. Capital will place greater emphasis on a company's commercialization capabilities, measured by whether customers are willing to pay, whether products offer value for money, and whether the company can achieve profitability. As a wave of companies completes IPOs, the embodied intelligence sector will face higher-dimensional competition.
Yu Yiran, managing director at CIC Consulting, told the media: "The market's investment logic will require companies to have rigid demand scenarios. Companies with breakthrough capabilities and clear commercial paths will receive further resource support; those lacking verifiable commercial closed loops or relying on purely conceptual narratives will face valuation corrections or even accelerated elimination due to high operating and hardware depreciation costs."
On August 20, Wang Xingxing appeared at the 2026 World Robot Conference, the day after the company's listing. He acknowledged that "insufficient generalization capability" is the biggest bottleneck facing the global robotics industry today. While current robot models can execute tasks such as moving, assembling items, and transporting goods without major issues, success rates drop noticeably whenever the objects are changed or the environment is slightly altered. This is because the "output precision" of AI models cannot align with the "execution error" of the physical world. Every input and output from a robot generates deviations and losses, and it is this accumulation of errors that prevents the robot from effectively correcting tactile errors during the final milliseconds of fine-tuning, leading to task failure.
Looking ahead, Wang expressed cautious optimism: "Whether it's model architecture, data collection methods, or data training approaches, things are much clearer this year than they were a few years ago. The level of attention and resource investment from everyone is also significantly greater than before."