Advanced Micro Devices and Anthropic have struck a deal worth tens of billions of dollars for artificial intelligence servers, a landmark agreement that strengthens AMD’s standing against market leader Nvidia while providing Anthropic with a massive infusion of computing power.

Under the agreement, Anthropic will purchase up to 2 gigawatts of AMD‘s next-generation Instinct MI450 chips, with rollouts beginning in the first half of 2027. AMD will also make its first direct investment in the AI firm, committing up to $5 billion as key deployment milestones are achieved.

“We have very much wanted to be a major part of their infrastructure,” AMD Chief Executive Lisa Su said, adding that engineering teams from both companies have already been working closely together for some time.

Anthropic currently spreads its computing workloads across multiple hardware platforms, including Google tensor-processing units, Amazon Trainium chips, and Nvidia graphics processing units. As part of this new agreement, Anthropic will purchase AMD chips for its own data centers and lease further capacity through third-party cloud providers and neoclouds. AMD and Anthropic are actively collaborating to identify suitable data centers to house the chips.

“You can’t just wake up one morning and say, ‘Oh, I want a gigawatt of compute tomorrow,’” Su said. “You actually have to plan, you know, 12, 18, 24 months in advance for what you want.”

AMD is also in discussions to provide a financial backstop for Anthropic’s future data-center leases. Established tech giants with strong credit ratings have increasingly stepped in to back lease commitments or debt for AI startups, giving those younger firms access to capital on far more favorable terms. Google has previously agreed to backstop certain data-center arrangements to help Anthropic access its tensor-processing units.

The agreement highlights AMD’s ongoing push to build a larger presence in the GPU market, benefiting from AI developers seeking alternatives to Nvidia. AMD recently locked in major agreements with OpenAI and Meta Platforms as it competes for business among top AI companies. To win over large clients, AMD has occasionally offered sweeteners such as warrants for future shares in the company.

As part of an engineering collaboration built into the agreement, AMD will use Anthropic’s Claude models to help improve the performance of its own chip technology. For Anthropic, securing this infrastructure is vital as it manages rapid user growth that recently led to usage limits and outages. The AMD deal complements previous infrastructure agreements Anthropic signed earlier this year with Amazon, Google, and SpaceX.