Anthropic Just Hired the Man Who Built Google’s TPU From Scratch

Anthropic hired Amir Salek, the executive who founded and led Google’s Tensor Processing Unit program from 2013 to 2022, to join its compute team, the company confirmed Friday.

Salek led Google’s custom silicon effort through its first seven chip generations before leaving for Cerberus Capital Management. His move to Anthropic comes just weeks after the Claude chatbot maker confirmed it was building an in-house chip design team.

Hiring the person behind Google’s TPU program gives Anthropic’s chip ambitions far more weight than a job listing or a stated plan alone.

The Executive Who Built the Blueprint Everyone Else Is Copying

As Seeking Alpha notes, Salek founded and led Google’s TPU program for nearly a decade, helping deliver its first seven chip generations that now power much of Google Cloud’s AI infrastructure. 

Notably, it’s the same Cloud business that recently helped ease investor concerns over Google’s high AI spending. 

That track record matters because TPUs became one of the industry’s clearest examples of a hyperscaler successfully building custom silicon that meaningfully undercut Nvidia’s pricing for its own workloads. 

According to Bloomberg, Salek will report to James Bradbury within Anthropic’s compute team, joining the same group behind its long-running multi-chip strategy and partnerships with Amazon, Google, Nvidia and AMD. 

A Résumé That Reads Like an AI Chip All-Star Roster

Before Google, Salek spent roughly eight years at Nvidia, where he founded and led its System-on-a-Chip design organization and worked on its GPU and Tegra product lines.

This gives Anthropic firsthand design experience from both companies whose hardware powers much of the AI industry. 

The combination of Nvidia’s GPU architecture and Google’s TPU experience is rare and valuable for early custom silicon programs where core architecture decisions are made.

Salek most recently served as senior managing director at Cerberus Capital Management, focusing on semiconductor and deep tech investing, before returning to hardware after years on the investor side.

The move also follows Nobel laureate John Jumper, co-creator of AlphaFold, who left Google DeepMind to join Anthropic in June.

Why This Hire Signals More Than the Job Listings Did

What separates Salek’s arrival from Anthropic’s earlier chip team announcements is the shift from stated intent to demonstrated commitment. 

Job listings and a spokesperson’s confirmation are weaker signals; any company can post a silicon engineer role paying up to $485,000 and call it a strategy. 

Hiring Google’s custom chip creator proves Anthropic plans a multi-year silicon effort, not just using it as an edge in pricing negotiations with Nvidia, which also develops competing, open-weight AI models. 

It also puts Anthropic in a more direct race with rival OpenAI, whose Broadcom-built Jalapeño inference chip debuted in June, the same month former OpenAI chip lead Clive Chan joined Anthropic to strengthen its own silicon effort.

With Salek now anchoring the team’s compute leadership, Anthropic has assembled significant hardware talent. 

The question is no longer whether Anthropic is serious about building its own chips, but whether its team can move fast enough before the AI chip shortage ends.

Source: Anthropic Taps Google Chip Veteran as Part of Push Into Hardware