Physical AI is Breaking the Hyperscale Model

AI’s biggest constraint is no longer compute but where compute can physically run. In her CTech op-ed, NGV Partner Tali Rosenwaks argues that limited grid capacity and Physical AI’s need for intelligence close to machines are pushing compute toward wherever energy is available. Hyperscale campuses will still handle model training, while distributed, energy-powered sites run real-time AI in the physical world. The winners will be the operators who can coordinate energy, location and compute together.