Anthropic Builds an In-House Silicon Team to Design Its Own Chips for Claude, Chasing a Lower Cost Per Query at Massive Scale

Anthropic has taken its clearest step yet toward owning more of the physical machinery behind its artificial intelligence, confirming on August 5, 2026, that it is assembling an in-house team to design custom chips for its Claude models.

The company framed the effort around a practical goal, namely running its technology faster and more cost efficiently at the enormous scale its customers now require, rather than a symbolic bid for independence from existing chip suppliers.

A team that spans hardware and software

Anthropic said it is looking to bring on engineers whose backgrounds span both hardware and software, so that the same people can develop the chips and the AI models side by side rather than in separate silos.

The published Silicon Engineer listing seeks talent across front-end design, pre-silicon verification, physical design, design for test, analog and mixed-signal work, and packaging, a sweep that covers nearly the entire lifecycle of building a modern processor.

The salaries attached to those roles, ranging from three hundred twenty thousand to four hundred eighty five thousand dollars, signal how fiercely companies are now competing for the relatively small pool of engineers who can design cutting edge chips.

The economics behind the move

The push is driven by arithmetic more than ambition, because serving billions of tokens a day at a revenue run rate of roughly thirty billion dollars turns even tiny savings per query into very large sums over the course of a year.

By co-designing its hardware and its models in tandem, rather than buying off the shelf accelerators, the company believes it can cut the cost of each query in ways that simply purchasing general purpose chips cannot match.

That logic mirrors the path taken by other large technology firms, which learned that once a workload becomes big enough and predictable enough, designing a chip tuned precisely to it can pay for the considerable effort involved.

Not an escape from Nvidia

Anthropic was explicit that this is not an attempt to walk away from its current suppliers, saying it plans to keep using Nvidia, AMD, Google's tensor processing units, and Amazon's Trainium chips alongside whatever it eventually builds itself.

On manufacturing, the company has reportedly been scouting Samsung as a potential partner, though it offered no timeline for when its own chips might arrive, nor did it say whether it would try to handle production in any way on its own.

For now the announcement is a statement of direction rather than a finished product, but it places Anthropic firmly inside a race where the companies building the models increasingly want a hand in building the silicon that runs them.