Anthropic is making sure it’s not dependent on any one of AI chip provider.
The company has signed a new agreement with AMD covering “tens of billions of dollars” worth of AI servers, according to a Wall Street Journal report, adding a fourth major chip supplier to a stack that already includes NVIDIA, Google and Amazon.
Under the terms of the deal, Anthropic will purchase up to 2 gigawatts of AMD’s latest-generation AI chip, the Instinct MI450, with deployment starting in the first half of 2027. In return, AMD will invest up to $5 billion in Anthropic as certain deployment milestones are met, marking the chipmaker’s first ever investment in the AI company. The two sides are also reportedly in talks for AMD to provide a financial backstop for Anthropic’s future data-center leases, a structure that would tie AMD even more closely to Anthropic’s compute build-out over the coming years.
For AMD, the agreement is another shot at closing the gap with NVIDIA, which has dominated AI training and inference hardware for years. AMD has been chasing similar mega-deals across the industry, including a $100 billion agreement with Meta for 6 gigawatts of compute and a deal with OpenAI that includes warrants for up to 10% of AMD’s stock. Landing Anthropic gives AMD a customer that is not just buying chips, but one whose revenue has been compounding at a pace that makes its future compute needs genuinely enormous.
Anthropic’s Widening Circle Of Chip Partners #
What makes the AMD deal notable is less the dollar figure and more where it fits into Anthropic’s broader hardware strategy. The company has spent the last year methodically adding chip suppliers rather than betting everything on one horse.
Its relationship with NVIDIA remains foundational — Claude has trained and run on NVIDIA GPUs since the earliest days of the company, and that partnership was formalized further last November when Anthropic, Microsoft and NVIDIA struck a three-way deal worth $30 billion in Azure capacity, with NVIDIA committing to invest in Anthropic as part of the arrangement.
Google has been an equally significant partner. Anthropic signed a deal last October to access up to one million of Google’s TPUs, a deal also worth “tens of billions of dollars.” That relationship deepened further this April when the two companies expanded their partnership to include Broadcom, adding multiple gigawatts of next-generation TPU capacity. Google has also invested billions into Anthropic directly, building what has been reported as a roughly 14% stake in the company.
Amazon, meanwhile, remains Anthropic’s primary cloud and training partner through Project Rainier, the Trainium2-based supercomputer cluster in Indiana that AWS has called the world’s most powerful AI training system. Amazon has invested billions into Anthropic since 2025 and has continued adding capacity commitments, including agreements for additional gigawatts of new compute disclosed alongside Anthropic’s most recent funding round.
Layered on top of all this, Anthropic has also tapped SpaceX for GPU access through its Colossus supercomputers, and signed a memory and storage partnership with Micron. The result is a company running Claude across at least four distinct chip architectures at once, with each supplier also holding some financial stake in Anthropic’s success.
Why The Diversification Makes Sense #
Anthropic’s revenue has gone from roughly $1 billion in annualized terms at the start of 2025 to over $40 billion within about a year and a half, a growth rate that CEO Dario Amodei has said outpaced even the company’s own internal projections by a wide margin. That kind of demand curve makes single-supplier dependence a real business risk — any delay in chip availability from one vendor becomes a direct constraint on how many customers Anthropic can serve.
Spreading purchases across NVIDIA, Google, Amazon and now AMD gives Anthropic negotiating leverage on pricing and delivery timelines, while also reducing exposure to supply shocks at any single chipmaker. It is a strategy that mirrors what OpenAI and Meta have also been doing this year, locking in chip commitments years ahead of actual deployment because, as AMD CEO Lisa Su put it, you cannot simply decide you want a gigawatt of compute tomorrow — that kind of planning takes 12 to 24 months.
For AMD specifically, the Anthropic deal also comes with a software angle. The two companies are launching a multi-year engineering collaboration where Claude will be used to help optimize AMD’s ROCm software stack and accelerate chip development work internally, while AMD adopts Claude more broadly across its own product teams. It’s a dynamic increasingly common in these chip partnerships, where the AI company doesn’t just buy hardware, it also becomes a development partner shaping how well that hardware performs against NVIDIA’s more mature software ecosystem. AMD shares dipped slightly in premarket trading following the report, even as the deal was framed as a meaningful win in the company’s long push to chip away at NVIDIA’s dominance of the AI accelerator market.