AMD commits up to $5B as Anthropic plans 2GW of Instinct GPUs
AMD will invest up to $5 billion in Anthropic, which plans to deploy 2 gigawatts of Instinct MI450-series GPUs for Claude starting in early 2027.
AMD and Anthropic announced a sweeping compute partnership on July 22: AMD will invest up to $5 billion in the Claude maker, which in turn commits to deploying up to 2 gigawatts of AMD Instinct MI450-series GPUs for training and serving its models.
Deal structure
The hardware will arrive as Helios rack-scale systems pairing MI455X GPUs with AMD's EPYC "Venice" processors and AMD networking. The first gigawatt phase is slated to come online in the first half of 2027, with a multi-year buildout to follow. Anthropic is not a new AMD customer — it already runs MI355X GPUs in production — but this is by far its largest commitment to the chipmaker.
Beyond silicon, the two companies will run a multi-year engineering collaboration in which Claude is used to improve ROCm, AMD's software stack and the perennial weak point in its challenge to Nvidia's CUDA. AMD will also roll out Claude internally across its development teams. Anthropic co-founder Tom Brown framed the deal as part of a deliberate multi-vendor strategy, saying working with different hardware providers lets the company match each workload to the best-suited chips.
The circularity question
AMD has struck similar arrangements with Meta and OpenAI, and the deal follows a now-familiar template: the chip vendor invests in an AI lab that then spends the money on the vendor's hardware. Critics argue these circular financing structures inflate reported demand and blur the line between revenue and investment.
Why it matters
For Anthropic — which already splits its compute across Nvidia, Google TPUs, and Amazon's Trainium — the deal adds a fourth major silicon supplier and real leverage over pricing. For AMD, landing a frontier lab as a ROCm co-development partner may matter more than the revenue: software maturity, not chip specs, is what keeps most AI workloads on Nvidia. And at 2 gigawatts, the agreement is another data point in the industry's stampede toward power-denominated compute deals.