AMD and Anthropic announced a deal on July 22 where AMD will invest up to $5 billion in Anthropic. In return, Anthropic commits to deploying two gigawatts of AMD's latest AI chips, the Instinct MI450 series, in large server systems called Helios racks. The first half of that capacity goes live in the first half of 2027.
To understand the size of this: one gigawatt of computing capacity costs roughly $50 billion to build out in full infrastructure. The hardware alone in two gigawatts is worth tens of billions of dollars. AMD's $5 billion investment is essentially the down payment that makes the purchase commitment happen.
This deal follows a clear pattern. Nvidia invested up to $10 billion in Anthropic in November 2025. Microsoft invested up to $5 billion in the same deal. Anthropic committed to spending $30 billion on Microsoft's cloud services in return. Amazon has put $8 billion into Anthropic. Anthropic uses Amazon's computing infrastructure. Now AMD is doing the same thing. The chip or cloud company writes a check, the AI lab commits to spending far more back on that company's products.
The criticism is fair: these deals look circular. A company invests in an AI lab, which then spends that money and more back on the investor's own products. But there is a real reason this structure exists. AI labs need enormous amounts of computing power, and the suppliers of that power want guaranteed, long-term customers. The investment is how you lock in both sides of that relationship. It is less financial wizardry and more a pre-commitment contract.
For AMD, the stakes are clear. Nvidia currently holds around 80 to 85 percent of the market for AI computing chips. AMD holds somewhere between 7 and 15 percent depending on the segment, and it is growing slowly. Deals like this one, alongside a similar arrangement with Meta for six gigawatts of AMD chips and 50,000 AMD chips being deployed by Oracle, are AMD's way of planting flags with the biggest AI spenders before Nvidia cements every relationship permanently.
There is also a software angle. Anthropic and AMD are starting a multi-year engineering project where Claude, Anthropic's AI, will be used to improve AMD's own software for running AI workloads. AMD will also use Claude internally across its engineering teams. This is becoming standard: AI labs and chip companies trade access to help each other improve.
For anyone whose business runs on Claude, the more important takeaway is that Anthropic is clearly preparing for significant growth in demand. The company flagged infrastructure strain in April 2025, noting that reliability and performance took hits during peak hours. The answer is more hardware from more suppliers, starting in 2027.
The broader question is whether AI labs will ever generate enough revenue on their own to fund this scale of spending without continuous investment from suppliers and cloud companies. That question remains open. For now, the pattern holds: each new chip partnership adds another layer to a structure where almost everyone in AI is, in some way, investing in everyone else.