Anthropic is writing another eye-popping check.
The Claude developer agreed to spend roughly $45 billion over six years to lease cloud computing power from London-based infrastructure startup Nscale, Bloomberg first reported on Wednesday.
The deal secures Anthropic around 460 megawatts of power capacity from the first building at Nscale’s flagship Monarch data center campus in West Virginia. The infrastructure will run on Nvidia Corp.’s next-generation Vera Rubin chips, an architecture integrating six distinct chips, and is expected to go live by late 2027.
The massive commitment follows a string of infrastructure investments by Anthropic. In recent months, the startup has locked down a $10 billion Norwegian data center contract with Volta, a $5 billion server deal with Advanced Micro Devices, a $1.25 billion monthly compute agreement with Elon Musk’s SpaceX, and a $9.1 billion, 20-year compute deal with Riot Platforms, alongside expanded ties with Amazon, Google, and Broadcom.
The aggressive expansion directly addresses Anthropic’s recent service challenges. The company acknowledged earlier this year that surge demand for its Claude models caused “inevitable strain” on its network, degrading “reliability and performance” for users during peak traffic periods, per CNBC.
Securing dedicated capacity also clears the runway for an expected public market debut. Valued at $965 billion in May, Anthropic submitted a confidential initial public offering prospectus to the Securities and Exchange Commission in June.
Nscale secures an anchor tenant
For Nscale, which launched in 2024, the agreement locks in a critical cornerstone customer as the company prepares its own initial public offering as soon as next month.
Anthropic stepped into the West Virginia project after Microsoft Corp. walked away from an earlier letter of intent for the facility during the summer, according to Bloomberg. Developing the entire three-building, 1.35-gigawatt Monarch campus will cost an estimated $71 billion, with $47 billion allocated toward AI processors. Construction at the site will employ 4,375 workers, while the completed development will support 645 direct on-site roles and approximately 3,600 indirect jobs statewide.
Why this matters: AI’s infrastructure race gets more expensive
Anthropic’s agreement with Nscale shows how quickly the AI race is expanding beyond models and into the physical infrastructure required to run them.
Long-term agreements with external data center operators can give AI developers access to enormous computing capacity without requiring them to build and operate every facility themselves. For Anthropic, securing capacity years in advance could also help reduce the risk that shortages of chips, power or data center space constrain Claude as enterprise demand grows.
That matters to channel partners and corporate customers because the availability of AI infrastructure increasingly determines how quickly providers can expand services, support larger workloads and maintain performance as adoption rises. Companies capable of reserving the most computing capacity could gain a significant advantage as customers decide which AI platforms can reliably support production deployments.
There is a trade-off. Anthropic is making multibillion-dollar commitments around infrastructure and processors that will not be fully deployed for years. Improvements in model efficiency, shifts in chip economics or slower-than-expected revenue growth could make those commitments more expensive than anticipated.
The bigger signal is that AI leadership is becoming as much an infrastructure contest as a software contest. Model quality still matters, but so do access to power, advanced chips, data center capacity and the capital needed to secure all three before competitors do.
Also read: Nvidia’s $500 billion AI infrastructure push shows how Wall Street financing is becoming another critical piece of the race to secure data centers, chips, and compute capacity.