Physical AI

Overview

Physical AI refers to AI systems that perceive and act in the physical world — spanning autonomous vehicles, humanoid robots, autonomous mobile robots (AMRs), drones, and industrial systems. Where Agentic AI operates in the digital domain across tokens and tools, physical AI must close the loop through sensors, actuators, and real-time control, subject to hard latency constraints, safety requirements, and the inherent unpredictability of the physical environment.

The central engineering challenge is Moravec's paradox: tasks trivially easy for humans — balance, grip correction, catching a ball — are the hardest unsolved problems in robotics, while tasks hard for humans (chess, theorem proving) are straightforward for AI. Physical AI requires distributed, hierarchically organized compute that spans millisecond-scale reflexes through sub-100 ms coordination to 300+ ms high-level reasoning, all running simultaneously. The market opportunity is substantial: Intel cited a $25 trillion projected physical AI market by 2050, while NVIDIA framed the physical world as the next frontier after digital AI, with robotaxis built on DRIVE Hyperion already representing manufacturers covering ~80% of global car production.