Summary of Key Points
Qianxun Intelligence is a “dark horse” in the field of embodied intelligence (physical AI), founded by Han Fengtao, a veteran in industrial robotics, and Gao Yang, a leading expert in AI. In just over two years since its establishment, it has raised over 4.5 billion yuan in funding and is valued at nearly 20 billion yuan, setting a new record for the speed of financing in this industry. The company focuses on developing the “brain” of embodied intelligence—universal models—and uses an unconventional training approach with “dirty data.” However, it faces challenges such as technical controversies (being removed from the RoboArena rankings), key researchers leaving the company, and slow commercialization, which highlight the contradiction between the current capital frenzy in physical AI and the realities of the industry.
I. Financing at Light Speed: Why Is Capital Flocking to Invest in Qianxun?
Qianxun’s financing pace is truly remarkable: it secured nearly 200 million yuan in seed and angel rounds within seven months of its founding, 1.1 billion yuan in a Pre-A round in 2025, and another 4.5 billion yuan in the first half of 2026 (equivalent to one-third of the total funding raised by more than 200 competitors). Its valuation has grown from zero to nearly 20 billion yuan in just over two years (compared to nine years for Yushu Technology). There are three main reasons for this capital rush:
1. Industry Trend: Physical AI is considered the next wave of AI innovation (referred to by Huang Renxun as the “growth engine after AIGC”), with the potential to enable robots to perform tasks in the real world like humans, creating a market worth trillions of yuan.
2. Strong Team: Han Fengtao has extensive experience in industrial robotics (having built 20,000 robots), and Gao Yang is a disciple of a pioneer in embodied intelligence (a Ph.D. from Berkeley and co-founder of Physical Intelligence). The combination of industry expertise and AI technology reassures investors.
3. Top-Tier Presence: The industry’s financing is highly concentrated; the top 5 companies account for 37% of all funds. Investors want to be among the first to invest in these leading players. Han Fengtao emphasizes that “embodied intelligence in 2026 is like large-scale models in 2023—without significant investment, you can’t make it to the forefront.”
The list of investors includes prominent figures such as Lei Jun from Shunwei Capital, Jack Ma from Yunfeng Capital, Liu Qiangdong, and Zhou Hongyi, as well as Saudi Aramco and Sequoia Capital, all seeking to invest in what they see as the next billion-dollar opportunity.
II. How Does Qianxun Train Its “Robot Brain”? The Unconventional Use of “Dirty Data”
Physical AI differs from digital AI like ChatGPT, which processes text and images in a virtual world. Physical AI aims to make robots capable of performing real-world tasks like picking up cups or folding clothes. The key lies in the model’s ability to understand physical laws and adapt to uncertainties. Qianxun’s training approach is unique:
1. Data Is Crucial: Data on how to perform tasks successfully (e.g., handling failed attempts to pick up a cup) is not readily available online, so Qianxun has to collect it itself. The company aims to gather 1 million hours of data by 2026.
2. Hierarchical Data Structure: The training uses a three-tiered approach: large amounts of human video data (low cost, wide range of scenarios), interaction data from remote control and wearable devices, and high-precision data from actual robot experiments.
3. Unconventional Data Strategy: While other companies use “clean data” with perfect actions, Qianxun includes failed and chaotic attempts because the real world is imperfect. This approach helps robots behave more like humans and handle unexpected situations.
Han Fengtao notes that the current models are only as intelligent as a two- or three-year-old child and needs to be improved significantly before they can be widely deployed (expected in the second half of 2027 to 2028).
III. Rising and Falling from the Rankings: Minor Setbacks or Major Concerns?
Qianxun has recently faced two issues:
1. Removal from RoboArena: In June, Qianxun’s Spirit v1.6 model topped the global embodied intelligence rankings, but it was removed three days later due to having only 25 valid evaluations (below the required 100). Although there were technical flaws in the ranking system, this raised doubts about its capabilities.
2. Key Researcher Departure: Guo Junliang, the head of the embodied intelligence team (hired from ByteDance), left the company after a year due to differences in research direction. The new leader is former Microsoft researcher Guo Junliang, but it’s uncertain whether this will affect the progress of the models.
These issues have not shaken investor confidence for now, but if technical progress slows or the company deviates from its goals, it could impact future funding and competitiveness.
IV. Robots Are Still in the Learning Stage: How Far Is Qianxun From Mass Production?
Despite its rapid financing, Qianxun’s commercialization is still in its early stages:
- Industry Comparison: Yushu Technology sold 5,500 robots last year and aims for 10,000 to 20,000 this year; Zhiyuan Robotics has received over 10,000 orders.
- Qianxun’s Progress: It has partnerships with companies like Bosch, JD Pharmacy, and CATL, but it hasn’t yet started large-scale shipments. Its robots are still working on basic tasks like folding clothes and making coffee demonstrations.
- Han Fengtao’s Goal: He aims to sell 100,000 robots per year, but acknowledges that the models need further improvement before reaching that goal.
The industry commonly faces the challenge of products remaining in demonstration stages. Investors want to see practical applications, but the technology is not yet ready. If Yushu and Zhiyuan gain a competitive edge in scale, Qianxun will face greater commercial pressure.
V. Is Physical AI the Next Big Thing? The Logic Behind the Capital Frenzy?
The potential of physical AI is widely recognized: it could replace humans in repetitive and dangerous tasks (e.g., factory logistics, hospital deliveries), with a market value of trillions of yuan. However, it’s currently at the “Robot GPT-1” stage (similar to large-scale models in 2023)—the technology is not yet mature. Nevertheless, investors are willing to bet early because no one wants to miss out.
Qianxun represents this trend: it has seized the opportunity and secured funding, but it still needs to overcome challenges in model development, data collection, and commercialization. In the next few years, it will either become a leader in physical AI or be eliminated in the fierce competition. This reflects the harsh reality of the tech industry.
In Conclusion: Qianxun’s story illustrates the current state of the AI industry, where capital enthusiasm temporarily masks gaps between technology and commercialization. Only real achievements will determine its success. The future of physical AI is promising, but the road ahead is long. Whether Qianxun can transform from a “dark horse” into a “giant” depends on its ability to develop robust models and successfully market its products.