Waymo Deploys Custom 5nm Silicon to Power Next-Gen Robotaxi Fleet

Waymo has revealed the architecture of its onboard computer, centered on a custom-designed 5-nanometer (5nm) application-specific integrated circuit (ASIC). The move signals a shift toward bespoke silicon to optimize the critical path between sensor data acquisition and vehicle actuation.

The custom ASIC serves as a front-end processor, handling raw data streams from lidar, radar, and cameras before the information reaches the core machine learning brain. According to company specifications, the chip delivers over 1,000 TOPS (tera-operations per second) of ML performance dedicated specifically to this front-end processing. A key technical capability of the silicon is temporal denoising, which allows the system to extract clearer information from raw sensor streams to improve perception in low-light environments.

Hardware Redundancy and Supply Chain

To ensure safety and reliability, Waymo designed the compute system with two independent engines. This hardware redundancy ensures that if one engine fails, the second can take over immediately. In a rare disclosure of its supply chain, Waymo listed its hardware partners, which include TSMC, Socionext, Nvidia, AMD, Micron, Samsung, and Sandisk.

This silicon is being integrated into the "Ojai," a purpose-built four-seater robotaxi manufactured by Zeekr, a subsidiary of Geely. By pairing a third-party vehicle body with in-house "brains," Waymo is positioning itself to scale its hardware while navigating the complex geopolitical tensions surrounding the use of Chinese hardware in U.S. infrastructure.

Scaling Amidst Operational Hurdles

Waymo is currently scaling its operations rapidly, managing a fleet of approximately 4,000 vehicles across more than 10 cities. The service is now completing roughly 500,000 paid trips per week. This transition to custom silicon mirrors a broader trend within Alphabet to reduce AI infrastructure costs and optimize performance for specific, high-intensity workloads.

However, the hardware reveal comes during a challenging period for the company. Waymo recently issued its sixth recall, which was specifically related to vehicles entering highway construction zones, highlighting the ongoing difficulty of solving "edge case" scenarios in autonomous navigation.

Industry Implications

By designing its own front-end silicon, Waymo reduces "pixels-to-actuation" latency, a critical metric for safety in high-speed autonomous driving. The choice of a mature 5nm node suggests the company is prioritizing reliability and manufacturing yield over the risks associated with bleeding-edge miniaturization.

Industry observers will now watch how this custom integration affects the fleet's real-world performance and whether the reduction in latency leads to a decrease in the types of operational errors that triggered recent recalls. While the hardware specifications are now public, the long-term impact on the cost-per-mile of the Ojai fleet remains to be seen.