Jensen Huang Says Physical AI Could Create Three Layers of $NVDA Compute Demand

Jensen Huang says physical AI requires three separate layers of computing.
The first computer trains the AI model.
The second simulates environments and tests skills before deployment.
The third runs inside the robot itself, handling real-time inference and control.
$NVDA wants to power all three.
That expands the robotics opportunity far beyond simply selling chips inside humanoid robots.
Even before humanoids reach mass production, companies may need large amounts of GPU capacity to train models and simulate millions of movements, tasks and edge cases.
That means the robotics boom could begin driving substantial AI infrastructure demand well before unit shipments become meaningful.
For $NVDA, the opportunity is not just the robot.
It is the entire development stack surrounding physical AI, from data-center training to simulation to edge compute.