Anthropic accelerates in-house chip push, recruits Google TPU architect

Anthropic, the company behind the Claude AI model, is stepping up its push to develop its own AI chips and has brought on a semiconductor expert who led Google's TPU program.

According to Bloomberg on Friday, Anthropic has hired Amir Salek, who headed Google's TPU project.

Salek worked at Google until 2022, leading development of the TPU from its first through seventh generations. Before joining Google, he spent eight years at Nvidia.

Anthropic's drive to secure semiconductor talent stems from its ambition to build its own AI chips in earnest. Demand for AI models is growing rapidly, but the computing infrastructure needed to support that demand remains insufficient. Anthropic plans to address the shortfall by developing proprietary chips.

The talent push is ongoing. Anthropic recruited Clive Chan, an early member of OpenAI's custom chip team, in June. The company is also reported to be in discussions with Samsung Electronics' foundry division.

Alongside its in-house chip development, Anthropic intends to continue using third-party semiconductors. The company has said Nvidia's GPUs, Google's TPUs and Amazon's Trainium chips will continue to play a central role in its computing strategy.

Competition to develop proprietary AI chips is intensifying across the industry. Rival OpenAI unveiled its own AI chip, "Jalapeño," in June.

Anthropic and OpenAI each filed confidential S-1 registration statements with the US Securities and Exchange Commission (SEC) in June, initiating the process toward a public listing.