Anthropic signs a six-year, $45 billion compute deal with Nscale
Anthropic has agreed to rent roughly $45 billion of AI compute from Nscale, a British AI infrastructure company. Bloomberg reported the deal first, and a source familiar with it told TechCrunch that the contract runs six years and pulls capacity from Nscale's flagship data center in West Virginia. The capacity is powered by Nvidia's Vera Rubin systems and is expected to start running Anthropic's workloads in late 2027.
The chip behind the contract
Vera Rubin combines six chips working together in one system and is Nvidia's newest flagship, which the company announced in January. It is the hardware Nscale will deploy to serve Anthropic under this contract, and TechCrunch describes the six-chip design as the current top of chip engineering. The late-2027 start date means the deal is buying capacity that does not exist yet: Anthropic is paying for a data center expansion that will only start producing a few years from now. Nscale is a young company. It was founded in 2024 and already supplies compute to Microsoft, among other customers, with the West Virginia site as its flagship facility.
Eight months of back-to-back deals
The Nscale contract is the latest in a string of compute agreements Anthropic has signed over the past eight months while scaling capacity to compete with OpenAI and other rivals:
- Earlier this month, a $10 billion deal with AI cloud startup Volta, securing six years of cloud compute from a data center in Norway. Volta was founded in January.
- In July, a $5 billion compute-related deal with AMD.
- In May, a large compute arrangement with SpaceX covering two of the company's data centers, reportedly worth about $1.25 billion of capacity per month. That is an unusual pairing: the centers run under Elon Musk, who has a long-standing rivalry with OpenAI CEO Sam Altman, making him an unlikely ally of Anthropic.
- In April, an expansion of its Amazon partnership adding 5 gigawatts of compute, plus a deepened relationship with Google and Broadcom.
The disclosed dollar figures on that list add up to more than $60 billion, before counting the SpaceX contract, which is reported on a monthly rather than total basis. The pattern is consistent across the deals: multi-year terms, capacity at data centers that are being built or expanded, and a growing share of the supply coming from companies that did not exist a couple of years ago.
The wider race
Anthropic is hardly alone in this push. The scramble for compute capacity is still running, with Google, OpenAI, and Meta all signing their own long-term deals. The Nscale contract is notable less for the chips involved than for what it shows: frontier labs are locking in multi-year, multi-billion dollar capacity well ahead of demand, and the suppliers they are turning to are increasingly young neoclouds rather than the hyperscalers.
What it means if you buy compute
If you are planning your own compute, two signals matter here. First, the big buyers are prepaying for capacity years before they need it, which keeps the newest hardware reserved and pushes the rest of the market toward older generations or token-based pricing. Second, the vendor list is diversifying fast: in eight months Anthropic added Volta, Nscale, AMD, SpaceX, Amazon, Google, and Broadcom as capacity suppliers. For anyone renting compute, that is more bargaining room than there has been in a while, at least until the late-2027 wave of new capacity actually comes online.
Source: techcrunch.com