OFICIAL Waymo Blog

A look under our trunk: what’s in our compute

What happened
Based on Waymo Blog · Aug 20, 2026

Waymo details its custom compute system for autonomous driving, featuring a 5nm ASIC and partnerships with industry leaders to process sensor data in real time with low latency and redundancy.

Key points
·
Compute is the brain of the Waymo Driver, translating raw sensor data into real-time driving commands.
·
Operating demonstrably safe, physical AI on the road demands a fundamental shift towards a system engineered for deterministic, low-latency performance.
·
Over the past decade, we have co-designed its hardware, sensors, and algorithms side-by-side to solve the unique constraints of real-world edge compute.
·
Waymo is offering the first look under its trunk to share its approach to compute, its custom silicon, and how we collaborate with industry leaders to build the most capable computing system on the road.
Key numbers
·
The company has scaled its raw compute power by 20 times over eight years while optimizing software to manage the demands of real-world edge computing.

Waymo’s compute system serves as the central processing unit for its autonomous driving technology, translating sensor inputs into real-time driving decisions. Unlike traditional driver-assist systems, Waymo’s approach operates entirely onboard without human intervention, requiring ultra-low latency and high reliability. The company has scaled its raw compute power by 20 times over eight years while optimizing software to manage the demands of real-world edge computing. This system is designed to handle diverse workloads, including advanced machine learning models that process high-fidelity data from multiple sensors simultaneously.

The compute hardware is engineered to withstand extreme conditions, including constant vibration, shock, and temperatures ranging from freezing winters to extreme heat. By integrating with the vehicle’s liquid cooling system, Waymo ensures peak performance in all environments. Redundancy is built into the design, with two independent compute engines operating in parallel. If one fails, the other seamlessly takes over, maintaining safety and minimizing latency in critical decision-making scenarios.

Waymo’s transition from off-the-shelf components to custom silicon has led to a highly integrated system that balances processing power with efficiency. The company’s latest 5nm ASIC is a specialized machine learning accelerator designed to process raw sensor data—including lidar, radar, and camera streams—in real time. This chip, along with other custom components, enables advanced neural networks to run at minimal latency while managing non-ML tasks such as data orchestration and logging. The result is a heterogeneous system optimized for both performance and efficiency.

Collaboration with industry leaders like AMD, Micron, NVIDIA, Samsung, Sandisk, Socionext, and TSMC has been critical to scaling Waymo’s compute capabilities. The company’s custom silicon and sensor co-design efforts have improved sensor fidelity, bandwidth efficiency, and model execution, including support for sparse convolutions and dense transformers. Waymo continues to explore new use cases for its AI stack, with plans to discuss its compute approach at Hot Chips and expand its team to further advance autonomous driving technology.

Original source → Deals on Clipraptor.com →