> ## Content Index
> Fetch the complete content index at: https://www.metatalks.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Waymo's first in-house chip, built with TSMC, is already running in its newest robotaxi
- URL: https://www.metatalks.ai/waymos-first-in-house-chip-built-with-tsmc-is-already-running-in-its-newest-robotaxi/
- Published: 2026-08-22T13:21:00.000Z
- Updated: 2026-08-22T13:21:00.000Z
- Author: Al
- Tags: News, Semiconductors, Physical AI, #newswire

**The custom part does only the front end of the job — turning raw LiDAR, radar and camera feeds into usable data — and Nvidia, AMD and Waymo's other silicon suppliers stay in the stack.**

Waymo has designed [a chip of its own](https://www.just-auto.com/news/waymo-develops-custom-chip-robotaxi-generation?ref=metatalks.ai) for the first time, and the part is already being manufactured. In a blog post dated 20 August the company set out that silicon alongside the architecture of its sensors and software: [an application-specific integrated circuit](https://www.automotiveworld.com/news/waymo-reveals-a-custom-chip-powering-its-robotaxi-fleet?ref=metatalks.ai) made on TSMC's 5nm process technology, now running inside Ojai, its newest generation of robotaxi.

The chip sits at the earliest stage of the driving system. As unprocessed LiDAR, radar and camera feeds arrive, it pulls usable information out of them on the fly — temporal denoising, which Waymo says improves perception in dim light, is one of the operations — and passes the result to the onboard machine-learning systems behind the car's driving decisions. Waymo puts the part's neural-network capability at more than 1,000 TOPS, or trillion operations per second, a capacity that covers sensor processing rather than the vehicle's whole computing workload. The latest system, the company says, can work through high-fidelity streams from 13 high-resolution cameras at the same time.

Until now the computers Waymo designed for that work were built around chips bought from Nvidia and AMD. The company presents the new component as an addition to those relationships rather than a break with them: AMD, Micron, Nvidia, Samsung, SanDisk, Socionext and TSMC all remain suppliers to parts of the computing stack, and the ASIC is one of several bespoke pieces of hardware it has under way, layered on top of bought-in equipment instead of displacing it.

Nvidia's published figure for its DRIVE AGX Thor platform, roughly 1,000 INT8 TOPS, lands in much the same range as the number Waymo quotes, but the two do not stand side by side. Waymo has never said what precision its own number refers to, and Thor is a single consolidated system-on-chip handling driving, cabin and entertainment tasks at once, while Waymo's part does nothing beyond preparing sensor data for the company's principal machine-learning system.

Waymo has put a figure on what its chip does and none on what it costs. The per-vehicle numbers in circulation come from outside the company, and they price a robotaxi's whole self-driving hardware rather than the silicon: $20,000 to $25,000 for the sixth-generation hardware, against $100,000 to $125,000 for the generation before it. Those figures may have to fall further still before a fleet at the size Waymo has in mind can be run at a profit.