Anthropic signs multi-year CoreWeave deal to rent Nvidia cloud capacity
Anthropic Is Renting Cloud Computing by the Billion. Its Competitors Are Buying the Data Centers.
When a company explores designing its own chips while simultaneously renting someone else's GPUs, the message is the same either way: there is not enough compute.
Anthropic signed a multi-year agreement with CoreWeave on Friday to rent the cloud provider's Nvidia-powered computing capacity, bringing additional capacity online later this year to power its Claude family of models. The financial terms were not disclosed. The deal is CoreWeave's latest in a string of large infrastructure agreements: an $11.9 billion contract with OpenAI last year, a $6.3 billion initial order with Nvidia in September, and a $21 billion expanded agreement with Meta on Thursday. CoreWeave shares surged more than 13 percent on Friday following the announcement.
The timing is not incidental. Reuters exclusively reported on Thursday that Anthropic is exploring designing its own artificial intelligence chips — plans that are in early stages and may not result in a finished product, according to three people with knowledge of the matter. The company has not assembled a dedicated chip design team or committed to a specific architecture. A spokesperson declined to comment. The reported cost of designing an advanced AI chip is roughly $500 million, factoring in engineering talent and the precision required in semiconductor manufacturing.
Anthropic is also deepening its relationship with Google and Broadcom. Earlier this week it signed a long-term deal committing to invest $50 billion in U.S. computing infrastructure through those partners, building on the custom tensor processing unit architecture Google has offered since the TPU's inception. That commitment, combined with the CoreWeave rental and the custom silicon exploration, represents three parallel tracks in Anthropic's answer to the same problem: its revenue is growing faster than its ability to build or buy the hardware to serve it.
Anthropic reported run-rate revenue surpassing $30 billion, up from approximately $9 billion at the end of 2025. That growth is placing acute pressure on compute availability. Claude has been one of the most demanded AI products in the market, and the gap between what Anthropic needs and what it has committed — through Google TPUs, through the Broadcom partnership, through rented cloud capacity — is apparently wide enough to justify exploring a half-billion-dollar custom chip project that may never ship.
The CoreWeave deal itself is structurally unremarkable: a cloud rental agreement with no disclosed price. What makes it significant is the context. With the addition of Anthropic, nine of the leading ten AI model providers now leverage CoreWeave's platform. The exception is Google, which builds its own chips. Every other major AI developer is, in some form, renting its compute from the same company or its underlying suppliers. That concentration is not a business story. It is an infrastructure story — one in which the physical layer of the AI race is being built by a single specialist supplier while the labs above it compete for access.