CourionAI
EN
Newsletter
← All news
chips 2 min read

Waymo Quietly Built Its Own Chip, and It Is Already Driving Around in the Cars

More than 1,000 TOPS, built by TSMC on a 5-nanometre process, and running in the latest generation of vehicles. Waymo joins Google, Amazon, OpenAI and Anthropic in designing silicon instead of buying it.

A car silhouette with a glowing chip under the hood above simple road lines

Waymo, Alphabet’s self-driving taxi arm, has designed its own chip and is already running it in its latest vehicles. Bloomberg reported the move on 20 August. Until now the company used chips from Nvidia and AMD.

The specifications Waymo gives are more than 1,000 TOPS, meaning trillions of operations per second, manufactured by TSMC on a 5-nanometre process. That puts it roughly level with Nvidia’s current autonomous driving hardware rather than ahead of it, which tells you what this is actually about. The chip handles two jobs: processing the flood of data coming off the cameras, radar and lidar, and running the AI models that decide what the car does next. Waymo is fitting it to the new robotaxi it is building with the Chinese manufacturer Zeekr.

Why everyone is suddenly making chips

Matching Nvidia’s performance is not the goal. Cost is. A robotaxi only makes money if the hardware in it gets cheap enough to put in thousands of cars, and a general-purpose chip carries a lot of capability a self-driving car never uses, plus a supplier’s margin on top. Designing your own means you build exactly the parts you need and pay a foundry like TSMC to make it. The upfront cost is enormous and the payoff only arrives at volume, which is why this is a decision about scale, not about engineering pride.

You may notice this is the same story as several others this week. Anthropic hired the founder of Google’s TPU programme and has a chip order with a UK startup. OpenAI is working on a Broadcom-designed chip. Amazon and Google have been building their own accelerators for years. The pattern is consistent: once a company’s AI workload becomes predictable and large, buying general-purpose hardware starts to look like paying rent. What is different about Waymo is that the chip runs inside a car rather than in a data centre, so power and heat matter far more than raw speed.

A fair caveat: these are Waymo’s own numbers, and TOPS is a headline figure that says little about how efficiently a chip actually runs a given model. It is a useful comparison point, not a verdict.

What this means for you: Directly, nothing, unless you live somewhere Waymo operates. Indirectly, it is a decent signal about where self-driving is in its lifecycle. Companies do not spend years and hundreds of millions designing custom silicon for a research project. They do it when they are confident enough about the design to optimise for cost. That is a more meaningful indicator than any demo video.

Sources

Source: https://www.bloomberg.com/news/articles/2026-08-20/google-s-waymo-has-built-a-custom-chip-for-its-robotaxis

Next story

A Robot That Only Tracks Objects Predicts the Wrong Move. Somebody Finally Measured By How Much

Adding beliefs, intentions and social norms to a world model lifted action prediction from 63.3 to 87.9 F1. GPT-4.1 with the framework beat GPT-5.6 Sol without it.

A robot with a thought bubble containing small human figures and prediction arrows