Nvidia’s next growth story may not be confined to data centers. The company is positioning its chips, simulation software, and AI models as the infrastructure layer for machines that can perceive, reason, and act in the physical world—including robots, autonomous vehicles, and industrial systems.
That business is estimated to generate about $10 billion in annual revenue today. Jensen Huang has said it could grow tenfold over the next decade, implying a potential $100 billion business. But the figure is an ambition built around future adoption, not a standalone revenue line Nvidia reports in its financial statements.
What “physical AI” means for Nvidia
Physical AI refers to systems that operate in the real world rather than only generating text, images, or software. For Nvidia, the opportunity spans several layers:
- Training: Large data-center GPUs train the models that control robots, autonomous vehicles, and other machines.
- Simulation: Developers can test behavior in virtual environments before deploying systems in the real world.
- Inference at the edge: Processors inside cars and robots interpret sensor data and make decisions in real time.
- Models and software: Open models and development tools help manufacturers build, test, and deploy physical-AI applications.
This combination is strategically important because it gives Nvidia more ways to participate in each machine’s lifecycle. It can sell the computing used to develop a robot, the software used to simulate it, and the processor installed inside the finished product.
Why Huang sees a 10-fold opportunity
Huang’s thesis is that AI is moving from the digital world into factories, vehicles, warehouses, and other physical environments. Manufacturing is particularly attractive because tasks are often repetitive, measurable, and structured—conditions that make automation easier to test and scale than many consumer or open-world applications.
Huang has described robotics as a potential $50 trillion industry and said its “ChatGPT moment” has arrived. That figure describes a broad potential market for manufacturing and related robotics, not Nvidia’s expected revenue. The company would capture only a portion of that value through processors, software, simulation, and infrastructure.
The key question is therefore not whether the market is large. It is whether autonomous machines can move from demonstrations and limited factory deployments to dependable commercial use. They must become safe, affordable, productive, and maintainable across much larger operating environments.
Nvidia is selling a stack, not just a robot chip
Nvidia’s strategy is designed to make its platform useful before and after a robot leaves the factory floor.
Its Isaac robotics tools and simulation frameworks are intended to help developers create and test robot systems in virtual environments. Nvidia says its broader platform is built for designing, training, testing, and deploying physical AI.
The company is also developing Cosmos models for physical-AI applications. Nvidia describes Cosmos Transfer and Predict as open world models that support physically based synthetic-data generation and robot-policy evaluation in simulation. Cosmos Reason is designed to help intelligent machines interpret and act on the physical world.
At the hardware level, Nvidia’s edge processors are intended for onboard AI. Its Unitree research system, for example, combines a Unitree humanoid body with Nvidia’s Jetson Thor hardware and Nvidia robotics models and simulation software.
The business logic is similar to Nvidia’s broader platform strategy: make the tools, hardware, and developer ecosystem work together closely enough that customers continue using Nvidia throughout development and deployment.
Why China is central to the robotics strategy
China currently provides an unusually large manufacturing base and domestic market for humanoid robots. Chinese companies accounted for more than 97% of global humanoid-robot shipments in the first half of 2026, when worldwide shipments reached roughly 19,100 units—up from about 5,100 in the same period a year earlier. 
1
The shipment figures show early industrial momentum, but they do not by themselves prove that humanoid robots are already broadly profitable or capable of replacing human labor. Still, China’s supply chains, production capacity, and factory demand give its robotics companies a strong environment for moving from prototypes toward higher-volume deployments.
That makes Chinese robot makers valuable partners for Nvidia. The company’s collaboration with Unitree connects Nvidia’s software and edge-computing platform to a prominent Chinese humanoid-robot manufacturer. The announced system is aimed initially at research institutions, including Stanford and ETH Zurich, rather than representing proof of mass consumer adoption.
For Nvidia, the strategic goal is broader than selling processors into one robot. It is to establish a common development and computing architecture as manufacturers scale.
H200 sales show the limits of the China opportunity
Nvidia’s China strategy also remains constrained by export controls. The H200 is a data-center accelerator used for demanding AI workloads, including model training and inference. U.S. authorities have permitted some sales, but access remains limited and politically sensitive.
Nvidia said it sold a small number of H200 chips to Chinese customers in its most recent quarter, with those sales representing less than 1% of data-center revenue for the period. The company also said it had not shipped the full amount permitted under its U.S. license.
Reports said ByteDance and Tencent each received batches of about 10,000 H200 processors, illustrating that some demand can still reach Nvidia even under restrictions. But these transactions should not be treated as a return to unrestricted China sales. Export approvals, Chinese policy, supply decisions, and local alternatives can all change the size and timing of the opportunity.
The distinction between Nvidia’s products also matters. Data-center GPUs such as the H200 support the training and operation of AI models, while processors such as Orin and Thor are designed for real-time computing inside vehicles and robots. A customer may therefore use Nvidia infrastructure during development while relying on Nvidia edge hardware when a machine is deployed.
How physical AI could diversify Nvidia
Nvidia’s financial performance is dominated by data-center demand. Physical AI offers a way to extend the company’s reach into automotive computing, industrial automation, logistics, robotics, simulation, and edge inference.
The business could also create a reinforcing cycle:
- More machines are deployed in the real world.
- Those machines produce operational data and expose new edge cases.
- Better data and simulation improve physical-AI models.
- More capable models make additional automation economically viable.
- New deployments increase demand for training, simulation, and inference hardware.
That flywheel is the central attraction of Nvidia’s strategy. If the company becomes the standard platform for physical AI, revenue could come from many thousands or millions of machines rather than only from a smaller number of hyperscale data centers.
The biggest risk: shipments are not adoption
The strongest evidence for a robotics boom today is growth in shipments and the rapid expansion of the surrounding software and hardware ecosystem. But shipment volume is only an early indicator.
Nvidia’s long-term case depends on autonomous systems working reliably beyond controlled environments. Robots must perform useful tasks in factories, warehouses, construction, farms, stores, roads, and eventually homes. They must also meet safety requirements, operate at an acceptable cost, and deliver enough productivity to justify deployment and maintenance.
If that transition takes longer than expected, Huang’s $50 trillion industry framing will remain a description of long-term potential rather than an immediate revenue pool. Nvidia’s physical-AI business has real products, customers, and early revenue—but the leap from a roughly $10 billion business to $100 billion depends on the real economy adopting autonomous machines at scale.
