Anthropic signed a massive deal to rent roughly $45 billion in computing capacity from Nscale, a British infrastructure firm. The move marks another huge step in Anthropic’s effort to secure raw processing power and keep pace with primary industry rivals like OpenAI.

Founded in 2024, Nscale previously signed infrastructure agreements with major tech companies like Microsoft. Under this new deal, Nscale will supply Anthropic with processing power using Nvidia Vera Rubin systems. The Vera Rubin system combines six different silicon chips to work as a unified unit, offering top-tier processing performance for training complex models. Anthropic expects to start running services on this hardware in late 2027.

Bloomberg first reported the deal, which spans six years. Most of the dedicated processing capacity will originate from Nscale’s flagship data center in West Virginia.

Over the past eight months, Anthropic went on a spending spree to expand its infrastructure. Earlier this month, the company inked a $10 billion agreement with cloud startup Volta to secure six years of cloud computing resources out of Norway. Back in July, Anthropic signed a $5 billion processing deal with AMD to diversify its chip suppliers.

A few months before the AMD deal, Anthropic locked down a massive agreement with Elon Musk’s SpaceX. That deal routes compute power from two distinct SpaceX data centers, providing Anthropic with roughly $1.25 billion worth of processing capacity every month.

In April, Anthropic expanded its existing partnership with Amazon to secure access to an additional 5 gigawatts of power. That same month, the company broadened working relationships with Google and Broadcom to guarantee steady chip deliveries and custom hardware integration.

Anthropic is not alone in buying up massive processing capacity. Major tech labs including OpenAI, Google, and Meta continue spending billions to secure chips, land, and power access. As models grow larger, access to raw computing capacity dictates who builds the most capable systems.

Securing long-term compute contracts gives research labs the foundation needed to train next-generation models without running into capacity bottlenecks. By locking in hardware years in advance, Anthropic guarantees that its engineering teams have the horsepower required to compete at the absolute top of the tech industry.