Waymo Built Its Own Chip. Nvidia Just Lost Its Biggest Robotaxi Customer.

Waymo just declared hardware independence.

The Alphabet subsidiary that runs the largest commercial robotaxi fleet in the United States revealed this week that it has designed, manufactured, and begun deploying its own custom chip for self-driving. The silicon delivers over 1,000 trillion operations per second on a TSMC 5-nanometer process -- performance on par with Nvidia's current automotive platforms, built specifically for Waymo's sensor stack and AI models.

Until now, those chips came from Nvidia and AMD.

The announcement, published in a blog post by Waymo's Vice President of Engineering Satish Jeyachandran and Compute Lead Daniel Rosenband, marks the culmination of a decade-long hardware co-design effort. The company didn't just buy a chip design off the shelf. It rebuilt its compute architecture from the ground up, matching its custom silicon to the specific sensor suite and driving models that power its robotaxis.

This is not theoretical. The chips are already in mass production, built by TSMC, and being installed in Waymo's fleet including the newly unveiled Ojai robotaxi.

Why Waymo walked away from the GPU standard

For years, the autonomous vehicle industry followed a simple formula: buy the most powerful Nvidia or AMD GPU you could fit in a car, strap on enough cooling, and write software that makes the most of it. It worked well enough that Waymo itself relied on those chips for its earlier generations.

But autonomous driving has a peculiar hardware requirement that consumer GPUs were never designed for: deterministic, low-latency performance under a strict power and thermal budget. A GPU that drives 4K gaming at 120 frames per second can stutter on frame 121 if a background process jumps in. In a robotaxi, that stutter is the difference between a smooth stop and a pedestrian injury.

Waymo's custom ASIC (Application-Specific Integrated Circuit) eliminates that uncertainty. Because Waymo designed the chip alongside the driving software, every transistor is allocated to a known task at a known time. Sensor data arrives, gets processed, feeds the AI model, and produces a driving command -- all on silicon purpose-built for that exact pipeline.

More than a chip announcement

The custom silicon launch coincided with Waymo unveiling its next-generation Ojai robotaxi platform. The Ojai, described by Waymo as a reimagined vehicle cabin that serves as a flexible, premium rider space, represents the physical manifestation of this vertical integration push. The chip is the brain; the Ojai is the body.

The timing is strategic. Waymo's fleet has grown to over 100,000 weekly driverless rides across multiple US cities, including its recent Houston public launch and a partnership with the Las Vegas Raiders at Allegiant Stadium. Scaling to those volumes made the economics of custom silicon increasingly attractive. Off-the-shelf GPU pricing, especially for automotive-grade Nvidia hardware carrying premium margins, becomes hard to justify when you are buying them by the thousands.

What this means for Nvidia

Nvidia's automotive business, while a small fraction of its total revenue, has been one of its fastest-growing segments. Losing Waymo as a chip customer -- even partially -- is a signal the GPU giant would rather not send to the market. If the leading robotaxi operator decided custom silicon was worth the enormous upfront investment, other autonomous vehicle developers may start asking the same question.

Tesla has been designing its own AI chips for years. Amazon's Zoox (acquired for $1.2B, same price as Waymo) uses custom compute. Now Waymo. The trend toward in-house silicon in autonomous driving is unmistakable.

Nvidia still has deep relationships with automakers from Mercedes-Benz to Hyundai, and its DRIVE platform remains the default choice for most automotive AI development. But the Waymo departure suggests a ceiling on how long the GPU giant can remain the default supplier for every self-driving project. When your robotaxi fleet reaches the scale where chip design makes financial sense, you build your own.

The deeper picture

This move is not just about chips. It is about what happens when autonomous driving matures from a research project into an industrial-scale operation. The early years of self-driving were about proving the software works. The current phase is about optimizing the hardware that runs it so efficiently that the economics of robotaxi fleets become viable at scale.

Waymo's custom chip is a bet that vertical integration -- designing the chip, the software, and the vehicle as one system -- delivers lower latency, higher reliability, and ultimately lower cost per mile than any combination of off-the-shelf components. Apple proved the model works in consumer devices. Waymo is testing whether it works on the road.

The company is quick to note it still works with Nvidia, AMD, and other chipmakers for supplemental compute. This is not a divorce. But it is a pointed statement about where Waymo sees the future of its hardware stack: in-house, purpose-built, and not for sale.

Sources

• Bloomberg: Alphabet's Waymo Has Built a Custom Chip for Its Robotaxis
• The Decoder: Waymo builds its own chip for its robotaxis, cutting its reliance on Nvidia
• The Verge: Waymo lifts the lid on the 'brain' powering its robotaxis
• Waymo Blog: A look under our trunk: what's in our compute
• Ars Technica: Waymo develops custom chip for latest robotaxi generation