Anthropic in Talks With Samsung for Custom AI Chip Manufacturing

Anthropic is pushing to build a custom AI chip to escape the cost and supply constraints of relying entirely on Nvidia GPUs. The company has held preliminary discussions with Samsung Electronics about manufacturing the chip, according to The Information, which cited people familiar with the plan. The move follows Reuters reporting from April that Anthropic was weighing its own chip production to respond to shortages.

The startup's computing costs have ballooned as it trains and runs ever-larger Claude models. Controlling its own silicon would let Anthropic optimize chip architecture specifically for its AI workloads rather than using general-purpose GPUs. It is also part of a broader pattern, as OpenAI announced its own custom AI chip partnership with Broadcom just a week earlier.

Samsung is offering its 2-nanometer manufacturing technology for the project, according to Jawlah's report citing The Information. That process node would place the chip at the leading edge of semiconductor fabrication, competing with TSMC's most advanced offerings. Samsung has struggled to win high-profile AI chip contracts against TSMC, which manufactures chips for Nvidia and most other major AI companies.

For Samsung Foundry, landing Anthropic would be a significant validation of its advanced process technology. The division has been chasing major logic-chip customers aggressively, and Anthropic represents a fast-growing, well-funded prospect in the booming AI infrastructure market.

Samsung's reported work on a custom SoC for OpenAI has slowed, Digitimes noted, making Anthropic an even more attractive alternative customer. The parallel tracks, OpenAI with Broadcom and now Anthropic with Samsung, show how the major AI labs are diverging in their chip strategies. OpenAI has leaned on its deep Microsoft partnership for infrastructure, while Anthropic is pursuing a more independent path.

This fragmentation in chip sourcing could reshape supplier relationships across the industry. Each major AI lab is now building or negotiating its own silicon pipeline, reducing collective dependence on Nvidia while creating new competitive dynamics among foundries.

In a separate but related development, The Information also reported that Anthropic is in talks to use Microsoft's AI chips. This suggests Anthropic is pursuing a multi-pronged silicon strategy, not committing exclusively to Samsung. Microsoft has been developing its own AI accelerators, notably the Maia chips, to reduce reliance on Nvidia in its Azure data centers.

If Anthropic were to use both Samsung-manufactured custom chips and Microsoft AI chips, it would mirror the hybrid approach that other cloud providers have taken. The Microsoft talks add complexity to any Samsung deal, as Anthropic may split workloads across different chip architectures depending on cost and availability.

Samsung and SK hynix both took strategic stakes in Anthropic's $65 billion Series H funding round in May, which valued the startup at $965 billion, according to The Korea Herald and The Star. That investment now looks like a down payment on closer business ties, with chip manufacturing orders the obvious next step. Samsung's foundry division and SK hynix's HBM memory business both stand to benefit if Anthropic builds out custom infrastructure.

The funding round made Anthropic the world's most valuable AI company by valuation, ahead of OpenAI. That financial heft gives it the capital to design and tape out custom silicon, an endeavor that typically costs hundreds of millions of dollars before manufacturing even begins.

Nvidia currently holds an estimated 74% global market share in AI chips, but custom silicon efforts from Anthropic, OpenAI, Google, and Amazon are gradually chipping away at that position. None of these alternatives eliminate Nvidia GPUs entirely, they typically handle specific inference or training workloads where optimization pays off. Still, each custom chip that ships represents revenue Nvidia won't book.

The foundry battle matters too. If Samsung can land Anthropic, it proves its 2nm node is competitive for demanding AI workloads, potentially attracting other customers away from TSMC. For now, the talks remain preliminary and no manufacturing agreement has been finalized.