Waymo's Secret Weapon: A Custom 5nm Chip Pushing 1,000+ TOPS Inside Its Next-Gen Robotaxi

Waymo gets attention for the visible parts of its business: the robotaxis arriving in a new city almost every week, most recently the public opening of its next-generation Ojai vehicle to all riders in Los Angeles, Phoenix, and San Francisco. But the most consequential announcement the company made this week was about a component most riders will never see — a custom silicon chip that Waymo designed itself, sitting at the front of the data pipeline of every sixth-generation Driver it ships.

As reported by TechCrunch Mobility’s Kirsten Korosec on August 23, 2026, the chip is a 5-nanometer ASIC built to handle the massive influx of raw sensor data before it ever reaches the core “brain” of the self-driving system. Waymo says it delivers more than 1,000 TOPS — trillions of operations per second — of computing performance, which puts it roughly in the same performance class as Nvidia’s latest DRIVE AGX Thor automotive processor, one of the most powerful computers ever aimed at automated driving.

Why a robotaxi company needs its own silicon

The numbers explain the problem. The Waymo Ojai carries 13 high-fidelity cameras, each built around a next-generation 17-megapixel imager — a resolution Waymo describes as “a generation ahead” of standard automotive cameras. Add four lidars, six imaging radars, and a set of external audio receivers that can localize a siren before the vehicle sees it, and the raw data stream flowing into the compute stack is enormous. Every frame from those cameras must be processed, cleaned, and fused with radar and lidar returns fast enough for the vehicle to react to a pedestrian stepping off a curb in dense urban traffic.

That is precisely the job of the new ASIC. By pushing pre-processing complexity into custom silicon rather than relying on arrays of commodity hardware components, Waymo converts an expensive, power-hungry general-purpose computing problem into a fixed-function one that runs with, in the company’s words, “unmatched efficiency and performance.” The payoff is measured in fractions of a second of latency — and in dollars per vehicle.

The chip is also the clearest evidence yet of how vertically integrated Waymo has become. The company is not just a software developer bolting perception models onto someone else’s driving computer. It designs its own imagers, its own lidar optics with core components built in California, its own in-house radar algorithms — and now its own pre-processor, from the silicon up. TechCrunch notes that the chip revelation made clear just “how hands-on” the company has gotten as it chases the cost structure that robotaxi profitability requires.

The cost story is the real story

Context matters here. The sixth-generation Driver, which debuted in the Ojai, was designed from the outset around a brutal constraint: the vehicle had to be cheaper to build, operate, and maintain than the fifth-generation Jaguar I-Pace fleet it replaces. Waymo achieved part of that by radical sensor simplification — cutting from 29 cameras on the Gen 5 vehicle to 13, while actually improving perception, because each 17MP imager sees more than two lower-resolution sensors combined.

Custom silicon attacks the other half of the cost equation. A fixed-purpose ASIC at a mature 5nm process node delivers far more performance per watt and per dollar than general-purpose compute doing the same work. When you intend to scale a factory in Metro Phoenix toward “tens of thousands of units per year,” as Waymo has said it is doing, every dollar and every watt saved per vehicle multiplies across the fleet.

There is also a strategic dimension. Waymo disclosed a partner list that reads like a map of the global semiconductor industry: AMD, Micron, Nvidia, Samsung, Sandisk, Socionext, and TSMC. The message is nuanced — designing your own chip does not mean going it alone. Nvidia remains a partner even as Waymo’s ASIC claims performance parity with Thor, and TSMC manufactures the silicon. Waymo is walking the now-familiar Big Tech line: own the architecture that differentiates you, rent the fabrication that commoditizes you.

The industry read: autonomy is becoming a silicon business

The announcement lands at a moment when the entire autonomous vehicle industry is converging on the same conclusion. Tesla has long staked its full self-driving stack on in-house inference chips. China’s robotaxi leaders similarly pair custom compute with domestic fabs. And Nvidia’s DRIVE Thor platform exists precisely because automakers that lack Waymo’s engineering depth still need 1,000-class TOPS in a automotive-grade package.

What makes Waymo’s move distinctive is that it sits upstream of the main brain. Rather than competing head-on with the flagship autonomous computing platform, Waymo built a specialized pre-processor that makes the entire pipeline more efficient — an admission that in a system ingesting gigabits per second of sensor data, the bottleneck is often the plumbing, not the model.

The timing is also pointed. Just days earlier, Nevada regulators approved permits for Tesla, Uber’s affiliate Aviari, and Waymo that could put up to 8,000 robotaxis on Las Vegas-area streets within a year. The industry is moving from proving autonomy works to proving it scales — and scale wars are won on unit economics. The company that can cheapest manufacture, fuel, and maintain a fleet that drives millions of miles a week will out-expand rivals regardless of whose benchmark numbers look better.

Waymo’s own framing is careful — “a critical piece of a system that can react fast and safely in complex, high-density environments” — but the subtext is unambiguous. After seven years of paid autonomous service and nearly 200 million fully autonomous miles across more than ten major cities, Waymo has learned that the path from impressive pilot to durable business runs through supply chains and semiconductor roadmaps as much as through neural network architecture.

What to watch

Three things will determine whether this chip matters as much as its TOPS figure suggests. First, whether the Ojai’s cost-per-mile actually drops enough to show up in Waymo’s expansion cadence — the company is now opening cities at a pace that demands vehicle production, not just software updates. Second, how the compute relationship with Nvidia evolves: partner and substitute at the same time is a delicate equilibrium. Third, whether rivals without in-house silicon can keep their bill of materials competitive as sensor resolution — and therefore pre-processing load — keeps climbing across the industry.

One thing is already clear: the race to make robotaxis profitable has moved below the surface of the vehicle. The next generation of autonomous driving will be decided as much by 5-nanometer lithography and imager physics as by model weights — and Waymo just showed its hand.