Anthropic is in the middle of a compute crisis that is genuinely unusual for a company growing this fast. Its CEO, Dario Amodei, said this month that the company has had "difficulties with compute," which is a careful way of saying demand for Claude has outrun the company's ability to serve it. Claude Code, the tool that helps software developers write code with AI, became popular faster than Anthropic's infrastructure could absorb.
The response has been a remarkable string of deals. Anthropic now has a 10-year arrangement with Amazon worth over $100 billion for access to Amazon's custom chips. It has a $200 billion, five-year deal with Google for cloud capacity and Google's own chips. It signed a deal with SpaceX just weeks ago, paying $1.25 billion per month through May 2029 for access to over 220,000 Nvidia GPUs sitting in a Memphis data center that used to belong to Elon Musk's other AI company. And in November, Microsoft invested $5 billion in Anthropic while Anthropic committed to spending $30 billion on Microsoft's Azure cloud.
Now Microsoft wants to go further. The Maia 200 is a chip Microsoft built specifically to run AI models, not for general computing. Microsoft says it costs 30% less per unit of output than the previous generation of hardware it was using. The chip is already running in Microsoft's own data centers, but has not been offered to outside customers yet. Giving Anthropic access would make it the first external company to use it at scale.
This matters because Microsoft is playing catch-up. Google has offered its custom chips to outside cloud customers for years. Amazon has done the same with its Trainium chips, and has already deployed over 1.4 million of them. Microsoft built Maia later, hit production delays, and has not yet put it in the hands of any paying customer. An Anthropic deployment would change that in a single move.
For Anthropic, the appeal is straightforward: cheaper computing capacity from a company that has already invested in it and that it already spends heavily with. The Microsoft chip, if it performs as claimed, could reduce the cost of answering every question Claude gets. Given the scale Anthropic is operating at, even a modest reduction per answer adds up to serious money.
The broader lesson here is about how AI companies are now thinking about infrastructure. Anthropic is not picking one supplier and committing to it. It is spreading its compute across Amazon, Google, SpaceX, and potentially Microsoft, treating each as a different pool of capacity to draw from when it needs it. That multi-supplier approach gives Anthropic negotiating leverage and reduces the risk that a single provider's shortage or outage becomes its problem.
For businesses using Claude today, or thinking about it, this spending race is actually good news in the medium term. The more compute Anthropic locks in, the more stable and consistent Claude's availability should become. The rate limits and slowdowns that have frustrated users this year are a direct result of demand outrunning supply. The supply is now being built, fast.
The Microsoft deal is not signed yet. It could still fall through. But the direction is clear: Anthropic wants every large source of computing power it can get, and Microsoft badly wants to prove its own chips can compete in the real world.