Nvidia integrates Chinese robotic hands into humanoid training systems

The hardware bottleneck

Developing humanoid robots capable of performing human tasks requires more than just advanced software. One of the most significant technical hurdles in the industry is the creation of mechanical hands that can mimic the dexterity of a human.

Because hands are so complex, they have become a primary pain point for developers. Companies are currently struggling to create hardware that is both strong enough to perform work and precise enough to handle delicate objects.

Nvidia is addressing this by showcasing two different humanoid robot training packages. These packages feature mechanical hands from two different firms with ties to Shanghai: Sharpa and Wuji.

Nvidia's training systems

Nvidia CEO Jensen Huang recently introduced a new humanoid hardware system designed to train robots on Nvidia's software. The hardware, known as Agile One SN, is built using the Nvidia Isaac robotics software and the Jetson Thor computing platform.

The Agile One SN system is intended to act as a 'flywheel' to speed up the development of robotics. By providing a standardized body and hand configuration, Nvidia aims to help researchers accelerate the training of complex AI models.

The system features a robot body from Agile Robots, a company with offices in Germany, China, and the United States. This collaboration highlights how multi-national hardware and software layers are combining to create a unified training environment.

The role of Chinese firms

Despite broader trade tensions, Chinese companies continue to underpin many U.S. tech endeavors. Sharpa, which is headquartered in Singapore, was co-founded by David Li of the Shanghai-based lidar company Hesai. The company keeps its manufacturing research and development in Shanghai.

Wuji is another key player in this space, founded by Pan Yunzhe in 2019. While the founder moved much of the team to Shenzhen, the company has drawn significant backing from Chinese entities like Ant Group, HSG, and Future Capital.

The Wuji Hand 2 was recently demonstrated in a video where it autonomously spun a pen even when faced with interference. This level of performance is helping these Chinese startups gain prominence in the global robotics race.

Research and delivery

Nvidia has identified several prestigious institutions that will likely use these systems. Research institutions such as Stanford and ETH Zurich are expected to start using the Agile One SN system in early 2027.

These same schools were also named in connection with Nvidia's first reference humanoid design, which was announced in May. That earlier design utilized a body from the Chinese humanoid company Unitree and mechanical hands from Sharpa.

While Nvidia has not confirmed if deliveries for the first design have begun, the 2027 timeline for the new system provides a clear roadmap. This suggests a long-term strategy to integrate Chinese hardware into the backbone of Western AI research.

Market competition

The competition for robotics dominance is intensifying as companies look for ways to stand out. For example, Xiaomi recently tested its own robotic hand in a car factory and found that most tasks only required a specific range of movement.

Other major players are also engaging with Chinese hardware. Meta has reportedly worked with the Chinese startup Fourier to develop a humanoid prototype, according to reports from The Information.

As China pushes its humanoid industry to develop core technologies, mechanical hands remain a primary way for companies to differentiate themselves. The industry is moving toward a reality where the best software must be paired with the most capable hardware.

Broader economic shifts in China

While robotics technology advances, China's broader industrial landscape is showing mixed signals. Official data indicates that profits for large industrial firms rose by 14.4% in September compared to the previous year, marking the strongest growth seen since June. However, the automotive sector is facing a difficult period, with car salesmen reporting that the "golden September and silver October" season is the slowest they have seen in over two decades.

Retail and consumer trends are also shifting significantly across the region. For instance, the company Pop Mart reported a major drop in third-quarter revenue, with sales falling by 45% to 50% year-on-year. The steepest declines in these sales occurred in the Americas and Asia-Pacific regions, reflecting a cooling period for some major consumer brands.

Upcoming global AI events

The tech landscape is also preparing for several high-profile milestones in the coming weeks. In early November, the Global Financial Leaders' Investment Summit will take place in Hong Kong to discuss international finance. Additionally, the AI sector will see a significant release from Moonshot AI, which plans to provide the full weights for its new K3.1 Pro model on November 4.