Anthropic talks to Samsung about its own chip because depending on Nvidia costs too much

From April musings to real negotiations

Back in April, Reuters reported that Anthropic was toying with making its own AI chips in response to chip shortages. On Thursday, The Information added that the company is in contact with Samsung about a collaboration. According to the report, no decision has been made yet on what the chip will be used for, how it will fit into the server architecture, or how powerful it should be.

Diversification, not a breakup with suppliers

Anthropic told TechCrunch that a diversified hardware stack with chips from Google, Amazon and Nvidia will remain pivotal to its compute strategy. On a potential Samsung partnership, the company had nothing further to add. Reading this as a split from Nvidia would be premature; it is an insurance policy and a negotiating position.

Jalapeño raised the bar and Microsoft poured in billions

A week earlier, OpenAI and Broadcom announced their custom inference processor, Jalapeño, claiming better performance per watt than competing chips. Amazon and Google have offered custom accelerators in their clouds for years. And Microsoft joined the same wave by launching its own AI deployment company with a $2.5 billion commitment. Vertical integration has stopped being an eccentricity; it is now table stakes.

Samsung plays every side

Samsung manufactures chips for Nvidia, is building an AI chip factory with Nvidia in South Korea, and has discussed chipmaking with Google as well. For Anthropic, it is a partner that knows frontier hardware production first hand and has no reason to protect anyone’s monopoly.

Lilith's verdict

Custom silicon is above all a bargaining lever. Even a chip that does not exist yet lowers the price Anthropic will pay for its next generation of compute.