Waymo executive explains how self-driving cars make split-second AI decisions

In a talk at The Next Endeavor 2026, Waymo's Sabareesh Ravikumar discusses the unique demands of running AI in self-driving cars, where decisions must happen in milliseconds using live sensor data.

He contrasts this 'physical AI' with typical data-center AI, explaining how Waymo combines foundation models, sensor fusion, and custom-built silicon to achieve fast, reliable performance on the road. The talk covers how Waymo designs resilient computing systems and why actual real-world latency matters more than lab benchmarks.

Ravikumar also touches on the company's approach to building world models and preparing for increasingly powerful AI systems that must operate within the physical and power constraints of a moving vehicle.