That’s the kind of deal that AMD and Anthropic just made public. Please read it again. The company that makes chips is putting up to $5 billion into Anthropic, the company that makes Claude. In exchange, Anthropic has agreed to buy up to two gigawatts of AMD’s newest Instinct MI450 chips, which AMD says could cost in the tens of billions of dollars just for the hardware. In the first half of 2027, the deployment starts. It’s hard to fully understand the size of it.
This is a pattern that the business world has quietly accepted as normal. Companies that make chips invest in AI companies that buy their chips, and those AI companies use the money to buy more hardware. It is said that Nvidia is in talks to invest $30 billion in OpenAI. AMD and OpenAI already had a deal like this in October of last year. There are real concerns about how long this model can last because of these circular deals, but for now, everyone seems happy with the arrangement.
There is an engineering layer under the money that makes the AMD-Anthropic partnership feel a little different. Both companies say they will work together directly to make workloads run more smoothly on AMD’s Instinct GPUs and speed up development on AMD’s ROCm software stack. AMD will also use Claude on a large scale across its own product and engineering teams. That is more than just a purchase order. It’s more like a working relationship, the kind that grows over time in ways that simple dollar amounts can’t fully show.
Anthropic has been building infrastructure quickly this year, and they haven’t been quiet about it. In April, the company said that the high demand for Claude had really put a strain on its systems, making them less reliable during busy times. It has since made deals with Amazon, Google, Broadcom, and SpaceX to share computing power.
SpaceX, run by Elon Musk, pays $1.25 billion a month to use the Colossus 1 data center in Memphis. Regardless of the size, that monthly bill is huge, and it shows how quickly Anthropic is growing. Adding AMD’s Helios rack-scale systems to that list shows that the company is trying to make its hardware more diverse. Tom Brown, co-founder and chief compute officer of Anthropic, made this clear by saying that the company can match different workloads to the right infrastructure by running on a variety of hardware.
AMD can see what’s at stake more clearly. The company has been steadily building a reputation as a credible alternative to Nvidia in the market for AI accelerators. They aren’t trying to beat Nvidia, but they do want to be a reliable second choice for serious customers. Anthropic landing at gigawatt scale sends a message to the market as a whole. It boosts AMD’s credibility at a time when buyers are wondering if there are real alternatives to Nvidia that can work in production settings. Analysts have said that each win like this helps AMD’s long-term story, but doesn’t always change the way competitors work right away.
When the news came out, AMD shares went up about 2.4%. This year, the stock has more than doubled, which is an amazing run. Investors’ faith in AMD’s AI strategy seems to have quietly changed from being cautious to being genuinely optimistic. Whether that optimism lasts depends on how well things are carried out—on whether the chips ship on time, the software works, and the two engineering teams work together to make something real.
Anthropic, on the other hand, is moving into a completely different phase. In May, the company’s run-rate revenue was said to have topped $47 billion, a number that would have seemed crazy just a year earlier. It raised money at a valuation of $965 billion and is reportedly getting ready for a possible IPO. There is a bigger picture behind the AMD deal. The company is trying to secure the computing power it needs before it becomes even harder to access AI infrastructure. They are paying high prices for now, but when demand picks up, the prices will look reasonable.
As these numbers get bigger, it’s still not clear if the circular investment model will still work or if there’s a point where the math stops making sense. But for now, it looks like AMD and Anthropic have found something that works for both of them. In this race, that might be enough.
