Waymo Opens Up About the Computer Inside Its Robotaxis

Waymo has, for the first time, publicly disclosed details about the high-performance compute system housed in the trunks of its robotaxis, including chip architecture, processor specs, and internal component details. The disclosure came in a blog post published on August 20, 2026, authored by Satish Jeyachandran, VP of Engineering, and Daniel Rosenband, Compute Lead. The post also named the hardware suppliers Waymo relies on to build the computers powering its driverless fleet. (The Verge)
Waymo describes the onboard computer as the "brain" of the Waymo Driver. It combines server-grade CPUs and GPUs, processing information from dozens of sensors — cameras, lidar, and radar — mounted on the vehicle. Waymo pairs its machine learning (ML) technologies with CPUs, GPUs, and accelerators to manage critical non-ML tasks while maximizing the time available for ML computation. In practical terms, the system must juggle the heavy lifting of recognizing objects and predicting their movements with the more routine work of running vehicle systems, all on the same hardware. (Waymo, The Verge)
The company's compute design follows three core principles it calls non-negotiable: responsive, ruggedized, and redundant. Executives say the system must run nearly silent and occupy minimal trunk space so as not to compromise the rider experience. "Ruggedized" means the hardware can withstand vibration, temperature swings, and the wear of daily driving; "redundant" means critical systems have backups so a single failure does not compromise safety. Waymo builds its compute stack in-house but relies on a network of third-party suppliers for components it cannot manufacture itself. (The Verge)
The scale of that compute has grown dramatically. Waymo says it has scaled its on-vehicle compute power 20 times over the past eight years, a trajectory that tracks the increasing sophistication of the perception and planning models the system must run in real time. Perception models handle the work of identifying and classifying what the car's sensors see; planning models decide what the vehicle should do next. Both have become more complex as the system takes on denser traffic and more varied city layouts. (The Verge)
The disclosure lands at a moment of significant operational scale. Waymo operates roughly 4,000 vehicles across more than 10 cities, conducting approximately 500,000 paid trips per week. The fleet now drives more than 4 million miles every week. Its latest safety analysis covers more than 220 million fully autonomous miles through the end of March 2026, spanning five operating geographies including Atlanta for the first time. (The Verge, Waymo)
The safety data is notable. Compared to human drivers in the same areas over the same period, the Waymo Driver was involved in 94% fewer crashes causing serious or fatal injuries, 82% fewer crashes in which an airbag deployed, and 82% fewer crashes involving any reported injury. Vulnerable road user figures are equally strong: 93% fewer injury-causing crashes involving pedestrians, 84% fewer involving cyclists, and 84% fewer involving motorcyclists. In Atlanta alone, across more than 5.4 million autonomous miles, the Waymo Driver recorded 94% fewer airbag-deployment crashes and 86% fewer injury-involving crashes than the human benchmark, both statistically significant. (Waymo)
Translated into weekly terms, Waymo estimates its performance equates to one fewer serious-injury-or-worse crash every eight days, roughly six fewer airbag-deployment crashes per week, and approximately thirteen fewer injury crashes of any kind per week. Over its lifetime of operations, the company estimates 47 fewer serious-or-fatal-injury crashes, 305 fewer airbag-deployment crashes, and 707 fewer injury crashes of any kind than would be expected had those miles been driven by humans in the same locations. (Waymo)
Carol Flannagan, a Research Professor at the University of Michigan Transportation Research Institute (UMTRI), noted that Waymo now drives enough miles to make direct comparisons to human drivers on crash rates and praised the consistency of results across locations. The full data and methodology are published at waymo.com/safety/impact. (Waymo)
The compute disclosure also sharpens the contrast with Tesla's approach to autonomous driving. Tesla CEO Elon Musk has called lidar a "crutch" and "a fool's errand," arguing that sensor fusion among cameras, radar, and lidar introduces "sensor contention" — a situation where disagreeing sensors create conflicting signals that the system must reconcile. Waymo's architecture, by contrast, is built around fusing all three sensor modalities and requires substantial onboard compute to do so in real time. (The Verge)
The cost picture remains substantial. Waymo's current-generation robotaxi vehicles cost over $120,000, a figure Reuters reported could fall to $85,000. The company took many years to build a 1,500-vehicle robotaxi fleet and has also been conducting tests in Japan. (Reuters, Reuters)
The broader context here is that the decision to publish chip architecture, processor specs, and a supplier list is a departure for a company that has historically kept its hardware stack under tight wraps. For competitors and suppliers alike, the blog post provides the first concrete look at what is inside the trunk of a Waymo vehicle, and the supply chain behind it. Whether that transparency reflects confidence in a durable hardware lead, a response to regulatory or partner pressure, or an effort to normalize the technology with the public, the post does not say.
What it does make clear is that Waymo's compute problem is enormous, well-defined, and being solved at a scale few competitors can match. A 20x compute scaling over eight years, paired with a sensor-fusion architecture that processes dozens of inputs in real time across a 4,000-vehicle fleet logging 4 million miles weekly, puts Waymo in a category that is difficult to replicate without comparable investment and operational discipline. The safety data, now covering 220 million miles across five geographies with statistically significant results, adds a layer of empirical grounding that the autonomous driving field has lacked for most of its history.
The architecture choices Waymo has disclosed, the safety record it has published, and the scale at which it now operates are not separate stories. They are the same story: a vertically integrated system where the compute stack, the sensor suite, the vehicle platform, and the operational deployment are designed in concert. The trunk-mounted "brain" is the connective tissue, and for the first time, we can see what it is made of.


