Nvidia's Humanoid Robot Push Depends on Chinese Hands
The company's new training systems utilize mechanical hands from two startups with deep roots in Shanghai, highlighting a critical supply chain link for U.S. robotics.
By Malik Salazar ·
Two. That is the number of distinct Chinese robot hand manufacturers currently integrated into Nvidia's humanoid robot training ecosystem. In a span of a few months, the chip giant has unveiled hardware packages that rely on components from Wuji and Sharpa, both of which trace their origins to Shanghai.
The first point of interest is the physical demonstration of these capabilities. On Oct. 21, Nvidia CEO Jensen Huang presented the Wuji Hand 2 on a slide in Berlin. The device was attached to a robot body built by Agile Robots. Huang described the resulting system, named Agile One SN, as a flywheel for accelerating robotics development. The system utilizes Nvidia's Isaac robotics software and the Jetson Thor computing platform. A video released last week showed the Wuji Hand 2 spinning a pen autonomously while facing interference.
The second point involves the specific companies powering these hands. Pan Yunzhe founded Wuji in 2019 and moved much of the team to Shenzhen. Backers include Ant Group, HSG, and Future Capital, according to PitchBook. The other key player is Sharpa, headquartered in Singapore. Sharpa was co-founded by David Li, who is also the CEO of Shanghai-based lidar firm Hesai. The company states its manufacturing research and development is located in Shanghai.
The third point is the timeline for academic access. Nvidia stated that research institutions, including Stanford and ETH Zurich, can likely begin using the Agile One SN system in early 2027. These institutions were also named in connection with Nvidia's first reference humanoid design, which was announced on May 31. That earlier system uses a body from Chinese humanoid company Unitree and hands from Sharpa. Deliveries for that first system were scheduled for late 2026. Nvidia did not immediately respond to a request for comment regarding whether deliveries have started or if researchers will test both reference robots.
Industry observers note that the hands represent a primary bottleneck in humanoid development. Keerthana Gopalakrishnan, a research lead at Google DeepMind, compared the two startups in a podcast last month. She stated that Sharpa Wave is a bit stronger than Wuji. She described the Wuji hand as closer in size to a typical human hand but possessing the strength of a ten-year-old. It was not clear if she was evaluating the Wuji Hand 2 or an earlier version.
Zhu Mingxuan, who works at humanoid startup JoyIn, provided a practical assessment of the hardware. She told CNBC that she found Wuji's mechanical hands very light and effective at completing tasks, though they tend to heat up. She said Sharpa's hands performed better but were far more expensive. Zhu left U.S. humanoid company Figure last year, where she worked on models to help robots mimic human actions. She argues that hardware, specifically hands, is limiting the broader development of humanoid robots.
China is actively pushing its humanoid industry to develop core technologies. Mechanical hands are one method companies use to differentiate themselves. Consumer electronics firm Xiaomi said last week that testing its robotic hand in its car factory revealed that most tasks require only a specific range of movement. Additionally, Meta has worked with Chinese startup Fourier on a humanoid prototype, according to The Information. Neither company immediately responded to a request for comment on that collaboration.
