Nvidia Wants To Run The World's Robots. China Is An Eager Customer

Nvidia's chips aren't just for training chatbots. The company is also betting big on artificial intelligence for real-world objects like robots, cars and drones-and China is emerging as a major customer.

The business of physical artificial intelligence already generates some $10 billion in annual revenue at Nvidia. While that is still a fraction of the company's overall revenue, which totaled $303 billion in the four quarters ended in July, Chief Executive Jensen Huang says it will grow 10-fold within the next decade.

"If I can generate video of a finger moving, if I could generate video of a hand picking up a glass, why can't I cause a robot to do the same?" Huang recalled thinking, speaking onstage in San Francisco in July.

Nvidia pairs specialized chips for physical AI with software tools used by developers of robots and robotaxis so that customers grow committed to an Nvidia suite of products. The company has released open-weight models, which robot developers can freely download and modify, with names such as GR00T and Cosmos.

"Physical AI gives Nvidia a path from data centers into the much larger physical economy-vehicles, factories, warehouses and robots," said Michael Frank, chief executive of intelligence startup Radiant Intel.

Huang has brought in his children to help run the effort. His daughter, Madison, oversees marketing for Nvidia's physical AI division while his son, Spencer, is a director of product management.

In August, Madison Huang visited LG Electronics' robotics data factory in Seoul after the South Korean company announced plans for a humanoid robot using Nvidia's robotic model. The following day, she toured a robot exhibition in Beijing, drawing onlookers as her father often does on his Asia trips. She watched robots perform flying kicks and stopped by the booths of Chinese companies that have used Nvidia's technology.

The U.S. has barred Nvidia from selling its most powerful AI chips to China, but trade remains open for its robot and car chips.

China is the world's largest industrial-robot market, and in the first half of this year, the country shipped around 90% of the world's humanoid robots.

"China is critical to Nvidia's physical AI road map," said Brady Wang, an analyst at Counterpoint Research. Chinese companies are strong in manufacturing and commercializing new technology, so working with them "provides many real use cases" for Nvidia, he said.

While China has undertaken a national effort to develop AI chips that can match Nvidia's, Chinese robot makers today rely on Nvidia silicon and software tools to train and operate their machines, people in the industry said.

In its labs in the U.S. and China, Nvidia has used humanoid robots from different vendors including China's Unitree, Agibot and Fourier for collecting data and conducting experiments, the people said.

Nvidia has been beefing up its robotics and self-driving teams in China, according to company webpages and people familiar with the recruiting. Nvidia's published research papers reflect contributions from China-based researchers including interns studying at local universities. Chinese robot makers such as Unitree have used Nvidia's latest Thor chips, people familiar with the matter said.

"Nvidia works with companies around the world consistent with U.S. laws," an Nvidia representative said.

Nearly two decades ago, Nvidia released a tool called compute unified device architecture, or CUDA, which allows users to write code directly for its chips. By pairing hardware and software, Nvidia locked in customers and grew to be a company valued at trillions of dollars.

Now it is applying that same playbook to win over robotics developers.

The Nvidia representative said the company has supplied open models and open-source materials "because physical AI will be built by a broad ecosystem-not by any one company."

Jensen Huang said in July that Nvidia had set its sights on self-driving cars as the first robotics application that could have a large enough market and real economic value. Among its customers are Alphabet's robotaxi unit Waymo and Tesla as well as Chinese auto and robotaxi companies.

Its lineup for developers includes specialized computers that they can use to program and operate their devices. The latest of these computers, an entry-level model called Jetson Orin Nano 2, was released Aug. 25.

Nvidia's competitors in physical AI chips and software tools include Intel and Qualcomm. Some in China are testing products from Chinese vendors such as Huawei, Black Sesame and Horizon Robotics, anticipating that Washington could extend export restrictions.

"Using homegrown alternatives isn't top of mind right now, but I can see a day coming eventually as the technology gets sensitive," said Harry Qin, a former Horizon researcher who co-founded a startup using Nvidia's chips and software to develop robotic systems.