At an Anthropic developer conference in London on May 19, a laptop with Claude stickers was shown. Photographer: Chris Ratcliffe/Bloomberg
Dina Bass | August 21, 2026, 3:59 PM EDT
The race to build better AI is starting to look a lot like a race to build better chips.
Anthropic, the company behind the Claude chatbot, just hired the person who helped build Google's custom chip business from the ground up. It is the clearest sign yet that the AI company wants to stop renting its computing power and start owning its own.
Amir Salek is joining Anthropic's compute team and will report to James Bradbury. Salek ran Google's tensor processing unit division until 2022, and he oversaw the creation of seven successive versions of those processors. Tensor processing units are Google's custom-built processors designed specifically for AI work.
Before Google, Salek worked at Nvidia, the company that makes most of the AI chips on the market today. After leaving Google, he took a senior managing director role at Cerberus Capital Management, the private equity firm co-founded by Stephen Feinberg, who is now the Deputy Secretary of Defense.
That mix of experience matters. Salek has seen chip development from the inside at two of the biggest names in the industry, and now he is bringing that knowledge to a company that has relied on others for computing power.
As a firm races to build their own chips, grab the free Always Be Buying E-Book to build wealth steadily
Anthropic currently buys chips from Nvidia, Google, and Amazon. But the company has been clear that it wants to create its own semiconductor business, and it has started recruiting for the effort.
The logic is simple. When you rent chips from someone else, you take what they offer. When you design your own, you can build exactly what your AI models need. Custom chips could also help Anthropic get around supply shortages that have slowed AI development across the industry.
Anthropic is also placing bets beyond its own designs. The company has agreed to buy about $250 million worth of chips from Fractile, a UK startup, with the option to expand the deal. It also recently signed capacity agreements with Riot Platforms Inc. and Volta Infra Holdings Ltd. to lock in more computing power.
Anthropic is not the only one moving in this direction. OpenAI, its main rival, is developing a chip with Broadcom and plans to deploy it later this year.
Read More: OpenAI, Broadcom Unveil Chip to Run Models Faster, Cheaper
The picture is becoming clear: Anthropic is building optionality. Buying from startups, signing infrastructure deals, and bringing in senior silicon talent are all part of a hedge across the whole hardware stack.
The chip market is becoming the battleground for AI dominance. Whoever controls the silicon controls how quickly AI can improve, and that is why the largest AI developers are all trying to bring chip design in-house.
For investors, two early signals matter: the $250 million order from a startup means the supply picture is still tight, and any shift in custom-chip planning will change the relationship between AI companies and suppliers like Nvidia.
This transition will not take effect immediately. Designing a new chip from zero takes years, and even a veteran like Salek cannot bypass the physical engineering. But the decision to bring in a leader who has watched Google and Nvidia scale their own chips says something important: hardware is no longer a side project - it is a competitive weapon.
The AI boom is not purely about software. It is about the data centers, the processors, the memory, and the power that make large models possible. As technology firms open their own chip path, their choices ripple into startups, established manufacturers, and utilities.
The path may be long, but the direction is set. Anthropic, once content to be a customer, is now aiming for a partner and, likely, a competitor.
Chip success takes patience, and the Always Be Buying E-Book teaches that same habit for your money