Waymo has started running a custom-built chip inside the newest generation of its robotaxis, replacing hardware it previously sourced from Nvidia and AMD. Bloomberg first reported the shift, and Alphabet’s self-driving subsidiary confirmed the chip is already deployed, not merely announced.
The stakes are straightforward: chip supply and cost structure. Nvidia has spent 2026 watching its largest customers walk the same path Waymo just took, and that pattern deserves naming plainly rather than treated as a one-off. Meta, Microsoft, and several other hyperscale Nvidia buyers have all pushed custom silicon programs this year, each chasing the same goal of trimming what they pay a single supplier for compute.
Inside the vehicle, the perception models that make sense of camera, lidar, and radar feeds now run on that silicon, letting a robotaxi work out what is happening around it and decide how to respond. Waymo says this setup moves sensor readings through the system at a quicker clip than the hardware it replaced. That handoff, from raw sensor input to a driving decision, is the core workload, and it is where latency carries direct safety consequences.
On raw specifications, Waymo puts the new chip’s throughput above 1,000 TOPS, a unit that counts trillions of computing operations executed each second. TSMC fabricates the part on a 5-nanometer process, the company says, and Waymo further claims that output puts the chip on par with Nvidia’s current autonomous-driving systems. Both the TOPS figure and the parity claim come from Waymo itself: neither Bloomberg nor The Decoder cites independent benchmarking, so read them as a vendor’s self-assessment rather than a verified result.
Cost reduction is the stated motive, and the timing matters. Waymo is installing the new chip in a fresh robotaxi design it is building with Zeekr, the Chinese automaker. Pairing a self-designed processor with a Chinese manufacturing partner is a second dimension of this story worth naming honestly rather than folding into the chip headline: it puts a US autonomous-vehicle leader building hardware alongside a Chinese carmaker at a moment when AI supply chains between the two countries are under heavier policy scrutiny than at any point in Waymo’s history.
Nvidia’s autonomous-driving business has been smaller than its data-center franchise, but losing a marquee customer like Waymo to in-house silicon still signals where margin pressure is heading across the industry. The company has not disclosed unit costs for the new chip or how much it expects to save per vehicle, so the financial case remains a stated intent rather than a published figure.
Operators evaluating robotics or autonomy hardware contracts should treat Waymo’s TOPS and parity numbers as a claim to verify, not a benchmark to cite, and should watch whether the Zeekr partnership becomes a template other AV programs copy for cost control.
Reported by Maximilian Schreiner for The Decoder on August 21, 2026, citing Bloomberg as the original source.
