Alphabet’s autonomous driving unit Waymo is ending its reliance on merchant silicon from Nvidia and AMD, deploying a proprietary processor to power its sixth-generation robotaxi fleet, Bloomberg reports. As commercial ride-hailing operations scale across major US metros, off-the-shelf automotive platforms from merchant chipmakers have become a margin drag that Alphabet can no longer justify.
Manufactured on TSMC's 5-nanometer node, the custom silicon delivers over 1,000 TOPS of compute—matching or exceeding Nvidia’s Drive Thor baseline. Waymo engineered the architecture specifically for direct sensor stream ingestion, slashing latency for on-board multimodal vision models and localization stacks without paying the vendor tax on generic GPU compute.
The custom chip is debuting in Waymo’s purpose-built vehicles developed alongside Geely’s Zeekr brand. While building proprietary silicon demands heavy upfront non-recurring engineering (NRE) capital, vertical integration is the only viable path to reducing hardware bill-of-materials (BOM) and total cost of ownership (TCO) at fleet scale. For Nvidia, Waymo's departure signals a familiar threat: when AI compute shifts from experimental R&D to high-volume commercial inference, hyperscalers invariably design their own silicon.