Anthropic Signs $45 Billion Computing Deal with British Startup Nscale
The AI company behind Claude is locking up computing power at a pace that suggests it expects AI demand to keep climbing hard through the rest of the decade.
Key points
- Anthropic agreed to a $45 billion, six-year deal with Nscale, a British computing infrastructure company founded in 2024, for AI computing capacity.
- Nscale will supply computing power through Nvidia Vera Rubin chips, Nvidia's most advanced chip system, from a data center in West Virginia.
- Computing capacity from the deal is expected to start running Anthropic's services in late 2027.
- Since April 2025, Anthropic has signed at least five separate large computing deals worth tens of billions of dollars combined.
- Rivals including Google, OpenAI, and Meta are pursuing similar computing expansions.
Anthropological has agreed to spend up to $45 billion renting computing power from Nscale, a British AI infrastructure company founded just last year, according to a source familiar with the agreement, first reported by Bloomberg. The deal spans six years and draws on Nscale's main data center in West Virginia.
The computing power will come from Nvidia's Vera Rubin system, a next-generation chip setup (chips are the specialised pieces of hardware that do the heavy number-crunching AI needs to run) that combines six different chips working together. It is currently Nvidia's most advanced chip design. Capacity is not expected to flow into Anthropic's systems until late 2027.
Why is Anthropic spending so much on computing?
More computing power means faster, more capable AI. Anthropic, the company behind the Claude family of AI assistants, is trying to close the gap with OpenAI, its main competitor, and that requires a lot of hardware.
The Nscale deal is the largest single computing agreement Anthropic has announced, but it is far from the only one. The pace over the past few months has been striking.
| Date | Partner | Value | Notes |
|---|---|---|---|
| April 2025 | Amazon | Undisclosed | 5 extra gigawatts of computing capacity |
| April 2025 | Google / Broadcom | Undisclosed | Additional power capacity |
| May 2025 | SpaceX | ~$1.25 bn/month | Two SpaceX data centers |
| July 2025 | AMD | $5 billion | Computing-related deal |
| July 2025 | Volta | $10 billion | Six-year deal, Norway data center |
| 2025 | Nscale | $45 billion | Six-year deal, West Virginia |
Volta, it is worth noting, was itself founded only in January 2025, making two of Anthropic's biggest recent partners very young companies.
Anthropics is not alone in this spending race. Google, OpenAI, and Meta are all signing large computing deals of their own, betting that demand for AI services will justify the cost.
What does this mean for ordinary users?
Nothing changes today for people who use Claude. The West Virginia capacity does not switch on until late 2027, so any effect on the speed or capability of Anthropic's products is a couple of years away.
Longer term, more computing capacity generally allows AI companies to train more powerful models and serve more users without slowdowns. Whether those improvements reach free-tier users or sit behind paid plans is a business decision Anthropic has not yet made public for this capacity.
