Nvidia Bets on Robotics Physical AI, China Emerges as Key Customer
nashnova research
Nvidia's physical AI business already generates roughly $10 billion in annual revenue, and CEO Jensen Huang sees a tenfold increase over the next decade; China, the world's largest industrial-robot market, is both the core engine of that bet and its biggest policy risk.
What is physical AI, and why is Nvidia betting on it?
Physical AI — using AI to control real-world objects like robots, cars, and drones, rather than just generating text or images — is Nvidia's new growth vector beyond data centers.
Current annual revenue sits at roughly $10 billion; Huang projects a 10× increase in about a decade. This means → Nvidia treats physical AI as a CUDA-scale long-term bet, not a short-term concept play.
Radiant Intel CEO Michael Frank's read: physical AI opens Nvidia's path into vehicles, factories, warehouses, and robots — a ceiling far higher than data centers alone.
What role does China play in this strategy?
China is the world's largest industrial-robot market. In the first half of this year, roughly 90% of global humanoid-robot shipments came from China. In plain terms = if you need to test physical AI at scale, China has the most robots and the densest manufacturing environments.
Counterpoint Research analyst Brady Wang states plainly: "China is critical to Nvidia's physical AI roadmap." Chinese firms' edge in manufacturing and commercialization gives Nvidia a wealth of real-world application data.
Nvidia already uses humanoid robots from Chinese makers — Unitree, Agibot, and Fourier — for data collection in labs in both the U.S. and China. Unitree and others are running Nvidia's latest Thor chip.
The lock-in playbook — how does Nvidia keep customers inside?
Nvidia bundles dedicated physical AI chips with developer software tools. The deeper a customer goes, the harder it is to switch. This reflects the exact same logic behind CUDA nearly two decades ago — attract developers with hardware, then lock them in with the software ecosystem.
The CUDA precedent: hardware + software binding ultimately pushed Nvidia to a trillion-dollar market cap. Put simply = the physical AI bundle is the "CUDA playbook" replicated in robotics.
Nvidia has also released open-weight models such as GR00T and Cosmos, free for developers to download and modify. The official line: "Physical AI will be built by a broad ecosystem, not any single company." But those free models run on Nvidia hardware — the more open the ecosystem, the tighter the hardware lock-in.
How much of a family affair is this?
Huang's daughter Madison heads marketing for the physical AI division; his son Spencer serves as director of product management.
This August, Madison visited LG Electronics' robot data factory in Seoul, then flew to Beijing the next day for the World Robot Conference, touring booths of multiple Chinese companies using Nvidia technology. This signals the strategic weight physical AI carries inside Nvidia — with family members directly advancing China-market relationships.
Where is the biggest uncertainty?
The U.S. has banned Nvidia from selling its most advanced AI chips to China, but robotics and automotive chips remain unrestricted for now, allowing Nvidia to maintain its China physical AI business within the current trade-policy framework.
Some Chinese robotics firms are already testing domestic alternatives in case the U.S. tightens export controls further. This means → Nvidia's share of China's physical AI market ultimately depends on whether the regulatory boundary stays where it is.
In plain terms = the growth story doesn't hinge on technology — it hinges on policy. That is the single most important uncertainty in the entire narrative.
市场有风险，内容仅供研究参考，不构成投资建议。