Ex-Qianxun Embodied AI Lead Jie Junyuan Launches gi labs in North America
An article by ZhiKe and republished by TMTPost with authorization reports that Jie Junyuan, former head of Qianxun Intelligence's embodied intelligence unit, has founded gi labs, short for General Intelligence Lab, in North America. The startup is focused on building embodied data-collection equipment, including motion-capture gloves, multi-camera systems and robotic grippers.
The company's integrated data-collection headset, EGO1GS, is equipped with 1080p binocular cameras and offers a field of view of 157H by 84D. Its sampling frequency reaches about 400Hz, twice that of the previous generation. The design is aimed at capturing wider, higher-frequency visual data for embodied AI model training.
People familiar with the matter said gi labs has set up an overseas office in San Francisco, while its main domestic office is in Suzhou. The team is composed of AI researchers, roboticists and engineers. The company is currently recruiting mechanical engineers and developing products such as robotic grippers, multi-camera systems and motion-capture gloves.
On the product side, gi labs is polishing a new-generation data-collection device. The product is still in the prototype stage and is expected to enter mass production in early November. Pricing starts at 3,555 yuan per set, with a price of 3,150 yuan per set for orders of 1,000 units or more. Compared with the previous generation, the core differences are a frame rate increased to 60fps and an improved field of view.
Public information shows that Jie graduated from the University of Science and Technology of China as an undergraduate. In his third year, he published a first-author paper at NeurIPS on image denoising and restoration, which has been cited more than 1,900 times. He then went to the University of Washington to pursue a doctorate. His career covers Amazon, ByteDance and Qianxun Intelligence. He led research and development of the MXNet system, was responsible for ByteDance's recommendation system and federated-learning platform, and from 2025 to 2026 took full charge of advancing Qianxun Intelligence's Spirit VLA large model, promoting its deployment in scenarios including CATL and JD.com.
In February 2026, Qianxun Intelligence completed two financing rounds totaling nearly 2 billion yuan, with a valuation exceeding 10 billion yuan. Investors included Sequoia China, Yunfeng Fund, Chaos Investment and industrial capital such as CATL. Jie previously held a 15% stake in controlled subsidiary Wanjing Qianxun. He exited that stake in March this year and formally left Qianxun in April 2026.
The supply-demand gap for embodied data is also substantial. The China Academy of Information and Communications Technology estimates that for embodied models to achieve a breakthrough comparable to ChatGPT, at least tens of millions of hours of effective data would be needed, while currently only million-hour-level high-quality data is available globally. For Jie and gi labs, credentials and capital are only an entry ticket. Mass-production delivery in November and repeat purchases from the first batch of customers will provide the first hard answer.