Unitree Technology (hereafter "Unitree") released a stock abnormal fluctuation announcement on the evening of October 26. The announcement stated that the company's shares had a cumulative closing price deviation exceeding 30% over three consecutive trading days on October 22, October 23, and October 26, 2026, constituting abnormal stock trading fluctuations.
Unitree was founded in 2016 and listed on the A-share STAR Market on August 19, 2026, at an issue price of 150.80 yuan per share, and is known as the "first humanoid robot stock" on the A-share market. On its first day of listing, it opened at 1,100 yuan per share, with a total market value of 444.9 billion yuan. It then continued to decline, hitting a low point on October 15, closing at 405.10 yuan per share that day, with its market value halved. After that, Unitree's stock price gradually rebounded, with a cumulative gain of 65.89% over seven trading days. On October 26, Unitree's intraday market value once returned to 280 billion yuan, then fell back to 271.8 billion yuan, with a daily turnover of 7.465 billion yuan and a turnover rate of 37.72%.
Unitree's announcement stated that as of October 26, the company disclosed a rolling price-to-earnings ratio of 465.22 times, higher than the average rolling price-to-earnings ratio of 38.54 times for the general equipment manufacturing industry over the most recent month. On October 25, investment bank JPMorgan upgraded Unitree's rating by two notches from "underweight" to "overweight," raising its target price from 300 yuan to 995 yuan. The advancement of embodied intelligence in the primary market and new stock IPOs has been slower than expected. JPMorgan believes that as the number of new comparable listed targets decreases in the short term, Unitree is regarded as the only leader in the A-share market and has become a scarce target in the embodied intelligence cycle. In this environment, Unitree's unique listing status, combined with its recent market value pullback, makes it one of the few investable representatives in this field, and its scarcity is still supporting its valuation.
In addition to Unitree, Shangwei New Materials, a subsidiary of embodied intelligence body manufacturer Zhiyuan Innovation, released a stock abnormal fluctuation announcement on the same day. It closed at 151.57 yuan per share on October 26, with a latest market value of 65.5 billion yuan. Zhiyuan acquired the A-share listed company Shangwei New Materials in July 2025. Robot data collection equipment and service providers have also been rising recently. From October 19 to 26, Fantuo Digital Creative, a concept stock in digital twins, embodied simulation, and industrial AI data platforms, rose by a cumulative 108.77%; Zhaowei Electromechanical, a supplier of robot end-effector data collection and actuators, rose by a cumulative 59.5%; Fulai New Materials, a supplier of tactile sensors and data collection gloves, rose by a cumulative 43.66%; and Duolun Technology, a data service provider for Zhongqing Robots, hit the daily limit for three days.
Performance Is the Foundation
Unitree's announcement mentioned that the humanoid robot industry is still in the early stage of commercialization, and there are uncertainties in technology maturity, downstream market demand cultivation, and industry chain supporting capabilities. At the same time, the large-scale implementation of physical AI technology from laboratory demonstrations to real-world scenarios still faces multiple challenges such as perception accuracy, motion control stability, data accumulation, and hardware-software coordination, and it will take time to verify the maturity of related technology paths and business models. Unitree's announcement also stated that if the progress of large-scale commercialization in the industry falls short of expectations, the company's future revenue expansion and performance release may be adversely affected, and there is an investment risk of valuation regression.
On October 22, at the 2026 World Physical AI Conference and the first Unitree Partner Ecosystem Conference, Wang Xingxing, founder, CEO, and CTO of Unitree, adjusted the expected time for physical AI to break through the critical point to one to two years or two to three years.
JPMorgan believes that advances in frontier AI large model capabilities have shifted the industry bottleneck of embodied intelligence from the "embodied brain" to deployment and interaction with the real world. This change enhances Unitree's value in short-term platform positioning, installed base, developer ecosystem, and self-evolution capabilities. JPMorgan believes that Unitree's next stage of growth will no longer be determined by complete machine sales volume, but more by its profitability quality, sustainability, and platform- and data-driven ecosystem.
According to data disclosed by Unitree, its actual shipments of humanoid robots delivered to end customers in 2025 exceeded 5,500 units, with more than 6,500 units rolled off the production line; as of July 2026, cumulative production of multiple bipedal humanoid robot models reached about 18,000 units. Based on data from multiple third-party institutions, global humanoid robot shipments in the first half of 2026 totaled between 22,000 and 25,000 units. Recently, Wang Xingxing disclosed at the Unitree Technology 2026 Ecosystem Partner Conference that as of now, Unitree's G1 humanoid robot shipments are between 10,000 and 20,000 units, ranking first globally for a single model.
Multiple embodied intelligence practitioners have told Caijing that the developer ecosystem is currently Unitree's "moat" ahead of other embodied companies. The high market share of Unitree's G1 humanoid robot among domestic and international research universities and geek communities enables rapid iteration of high-quality data feedback, hardware updates, and software adaptation, and in this process, a developer ecosystem is built around Unitree's body. However, in the view of some investors, the value of scientific research orders is not considered high. An investor mentioned that compared with industrial, household, and other scenarios, the repurchase rate of scientific research orders is not high, and the effect on subsequent revenue growth is not strong enough.
Policy is also actively promoting the commercialization of robots. On October 19, China's Ministry of Commerce and eight other departments issued the Opinions on Accelerating the Promotion of Technology-Driven Commerce, proposing to deeply implement the "AI + Manufacturing" special action, enrich the supply of smart goods such as smartphones, computers, home appliances, wearable devices, and robots, and clearly promote the implementation of embodied intelligence technology in the commerce and trade field, expanding application scenarios such as home services and commercial logistics.
Multiple Factors Stimulate the Market
Multiple fund analysts told Caijing that Wang Xingxing's judgment on the timing of physical AI breaking through the critical point was one of the catalysts for the recent physical AI theme rally. "The physical AI theme is just the original robot theme repackaged. Previously, the market focused on shipment volume, and later found that the key constraint on shipment volume lies in the robot's 'brain,'" a hedge fund industry analyst told Caijing. The logic behind this round of gains in robotics concept stocks is that the market has significantly raised expectations for the application of AI large models in the "embodied brain." The key variable is that GPT-6 Astra has demonstrated the ability to control robotic arms, "and what may be broken through in the future is not only embodied intelligence, but may involve a larger physical AI scope."
ChatGPT-6 Astra is the sixth-generation large language model released by U.S. large model company OpenAI in September. Multiple embodied intelligence companies have deployed the model in robot task tests and achieved good results. Wang Xingxing believes that with the current capabilities of this model, some robot planning or navigation tasks have already been solved. At the ecosystem conference, Wang Xingxing demonstrated the team's progress in deploying GPT-6 on the Unitree G1 robot, including cleaning brushes and paint palettes and changing paint during painting, playing Gomoku and chess, and fully autonomous navigation in unfamiliar indoor scenes. Wang Xingxing said the team only deployed it, wrote prompts, did zero training, and did not write a single line of code or complex program.
The core challenge of physical AI is the alignment between models and the physical world. Wang Xingxing said that current multimodal large models have relatively weak error correction ability for contact problems, and after deployment on real machines, input-output alignment errors will occur. The capabilities of GhatGPT-6 Astra are more reflected in the front end, but the last few millimeters of robot operation implementation cannot be solved. Wang Xingxing judged that at the fastest one year or at the slowest three years, "large models will 'break through' physical AI."
Multiple fund analysts told Caijing that recently, OpenAI founder and CEO Sam Altman's confirmation that OpenAI will develop humanoid robots has become another important driver of the market. A public fund machinery analyst told Caijing that before the National Day holiday, many institutions were looking for OpenAI experts to exchange views, and the meeting notes circulating in the market at that time showed that OpenAI had begun large-scale data collection. "OpenAI's public acknowledgment of this matter is different in significance, driving the collective rise of data collection targets in this round of market movement," the person said.
A brokerage research report showed that the market's trading focus has recently shifted from robot components (supply chain) to body manufacturers and data collection, with rapid gains in data collection equipment and services, sensors, and simulation targets. At present, the main data collection routes in the embodied intelligence industry can be divided into real-machine teleoperation, bodyless/UMI, first-person EGO, simulation synthesis, internet video, motion capture, and deployment feedback. Data collection equipment includes head-mounted EGO devices, data collection gloves, end-effector devices (grippers, dexterous hands), and corresponding tactile sensors. Multiple brokerage research reports show that after model capabilities are expected to break through the "embodied brain," data collection and training are considered a key link connecting large models with real-world robot deployment.
"But the potential market size and future growth rate of data collection are not clear," a fund industry analyst told Caijing. He believes that, on the one hand, large domestic embodied data collection service providers are not listed, and some of the stocks rising now have no direct connection with OpenAI itself, only a mapping relationship; on the other hand, the industry has not reached consensus on the amount of data needed to achieve embodied intelligence, and data collection equipment and technology routes are divergent, without stable demand having formed.
On October 21, Tesla founder and CEO Elon Musk disclosed progress on Optimus at Tesla's third-quarter earnings call. Musk said Tesla's Fremont factory has started early production of Optimus and expected issues such as AI, electromechanical design, reliability, and robot service life to be resolved by the end of 2027. Musk also said Tesla would release the latest Optimus demonstration video in about a month.
A public fund machinery analyst told Caijing that Tesla is not the main catalyst for this round of market movement. "The market itself has very low expectations for Tesla's Optimus, or rather no expectations." He believes the reason for no expectations is that Musk has repeatedly delayed the release timing and results demonstration of humanoid robot Optimus V3. "Compared with the second quarter, what Musk said at the third-quarter earnings call released some progress, but did not give information exceeding expectations." The aforementioned person said that in the past, the main focus was on the progress of Tesla Optimus's brain, but recently the ability of GPT-6 Astra applied to robots made him realize that breaking through the "embodied brain" may no longer require waiting for Tesla.
Comments