While U.S. export controls have severely restricted Nvidia’s ability to sell its most advanced artificial intelligence training chips to Chinese customers, the Silicon Valley giant is quietly but aggressively expanding its footprint in another critical sector of the Chinese tech ecosystem: robotics.
According to job postings published on the company’s official WeChat account and reported by the South China Morning Post, Nvidia is currently recruiting for more than a dozen specialized roles for its China robotics team. The positions are spread across the country’s major tech hubs (Beijing, Shanghai, and Shenzhen) and signal a deep, long-term commitment to what the industry is calling “physical AI”: the convergence of advanced AI models with robotic hardware that can perceive, reason, and interact with the physical world.
Targeting Embodied Intelligence
The recruitment drive is highly targeted, focusing on four key domains: embodied intelligence, simulation, implementation, and solutions. Specific open roles include engineers specializing in dexterous manipulation, whole-body control, robot learning, and AI systems optimization specifically tailored for humanoid robots.
Nvidia’s stated goal for this expansion is to “build a leading robotics platform and ecosystem to help developers and companies create autonomous machines.” The company aims to accelerate the transition of advanced robotics from research laboratories into real-world, commercial deployment.
To achieve this, the new hires will be working directly with Nvidia’s cutting-edge robotics software stack. This includes the Project GR00T humanoid robot foundation model, the Cosmos physical simulation world model, and the company’s suite of GPU-accelerated computing platforms designed for edge processing in autonomous machines. The Cosmos platform, in particular, is a critical tool for generating the vast quantities of synthetic training data required to teach robots how to navigate and manipulate objects in the physical world—a process that would be prohibitively expensive and slow if conducted entirely in real environments.
This hiring surge is part of Nvidia’s broader expansion in China. Despite geopolitical friction over semiconductor sales, Nvidia’s headcount in China grew significantly, expanding from approximately 3,000 to roughly 4,000 by the end of 2024, according to industry tracking data.
The Physical AI Loophole
Nvidia’s aggressive push into Chinese robotics highlights a fascinating dynamic in the current U.S.-China technology conflict. While Washington has placed strict embargoes on the export of high-end data center GPUs (like the H100 and B200) used to train massive large language models, the hardware and software required for robotics and embodied AI have not faced the same level of restriction.
Robotics relies heavily on edge computing, processing data locally on the machine rather than in a massive server farm. The chips required for this (such as Nvidia’s Jetson platform) are powerful, but they generally fall below the performance thresholds that trigger U.S. export bans. Furthermore, the software frameworks and simulation environments Nvidia provides are largely open or commercially available tools, not restricted military technologies.
This creates a unique environment in which Nvidia can continue to partner with and supply China’s booming robotics sector. And that sector is moving incredibly fast. In early June, Spirit AI, a Chinese robotics company, led the global RoboArena leaderboard with its v1.6 model, demonstrating the rapid pace of innovation from Chinese embodied AI developers.
Partnering with China’s Robotics Leaders
Nvidia is not just hiring in China; it is actively partnering with the country’s top hardware manufacturers. On June 1, Nvidia announced a major collaboration with Unitree Robotics, a leading Chinese developer of quadruped and humanoid robots. The partnership aims to build a humanoid robot platform that integrates Nvidia’s next-generation Blackwell chips, designed specifically for global researchers at institutions like Stanford University and ETH Zurich.
Unitree, which is currently preparing for an IPO on the Shanghai Stock Exchange, represents exactly the kind of hardware partner Nvidia needs to deploy its software ecosystem at scale. By providing the underlying simulation environments (Cosmos) and foundation models (GR00T), Nvidia is positioning itself as the indispensable “operating system” for the next generation of autonomous machines, regardless of where the physical hardware is manufactured.
Nvidia’s strategic logic is compelling. As EastFrontier has covered, China is aggressively pursuing the industrialization of humanoid robots, with government mandates pushing for mass deployment in manufacturing by 2030. Barclays projects that China will deploy 11 million humanoid robots by 2035, compared to just 2 million worldwide. By embedding its software and simulation platforms deeply into the Chinese robotics ecosystem now, Nvidia is ensuring it will capture a significant share of the enormous revenue opportunity this deployment wave will generate—even if the physical chips inside those robots are eventually made by domestic Chinese manufacturers.