Jensen Huang argues that a useful robot requires three distinct machines: one to train the model, another to test it in simulation and a third to operate the robot in real time. Nvidia positions its DGX and Grace systems for training, its Omniverse platform for simulation and the Jetson Thor embedded computer for runtime. The architecture relies on digital twins, the Isaac GR00T robotics model, synthetic data generated by Cosmos and the Newton physics engine, developed with Google DeepMind and Disney. Nvidia claims around $10 billion in physical AI revenue during the first half of 2026 and forecasts a global shortage of 30 million to 50 million workers by 2030. For investors, the framework creates a revenue path spanning training hardware, simulation tools and embedded chips, with CUDA serving as the company’s competitive moat.
Source: Read the original article
