Anthropic has agreed to commit roughly $45 billion over six years to lease cloud computing capacity from London-based infrastructure startup Nscale, according to a Bloomberg report published Wednesday. The deal, which represents one of the largest AI infrastructure agreements to date, underscores the escalating competition for chips, electrical power, and data center space among AI developers.
The agreement provides Anthropic with approximately 460 megawatts of power capacity from the initial structure at Nscale's Monarch data center campus in West Virginia. The facility will operate on Nvidia's upcoming Vera Rubin chips, an architecture that combines six separate chips, and is projected to launch by late 2027. This commitment follows several recent infrastructure investments by the Claude developer, including a $10 billion Norwegian data center partnership with Volta, a $5 billion server agreement with Advanced Micro Devices, a $1.25 billion monthly computing arrangement with SpaceX, and a $9.1 billion, 20-year computing contract with Riot Platforms, along with strengthened relationships with Amazon, Google, and Broadcom. For Nscale, which started operations in 2024, the deal establishes a foundational anchor customer as the company readies its own initial public offering planned for as soon as next month. Anthropic took over the West Virginia project after Microsoft withdrew from an earlier letter of intent for the location during the summer. Constructing the complete three-building, 1.35-gigawatt Monarch campus will require an estimated $71 billion, with $47 billion earmarked for AI processors. The site will employ 4,375 construction workers, while the finished development will generate 645 direct on-site positions and roughly 3,600 indirect jobs across the state.
The company acknowledged earlier this year that surging demand for its Claude models created "inevitable strain" on its network, degrading "reliability and performance" for users during periods of peak traffic, according to CNBC. Obtaining dedicated capacity also positions the startup for an anticipated public market launch. Anthropic, valued at $965 billion in May, filed a confidential initial public offering prospectus with the Securities and Exchange Commission in June.
The Nscale agreement illustrates how the AI competition is extending beyond software models into the physical infrastructure needed to operate them. Long-term contracts with outside data center operators allow AI developers to access massive computing capacity without needing to construct and manage every facility on their own. For Anthropic, reserving capacity years ahead could help lower the risk that shortages of chips, power, or data center space limit Claude as enterprise demand expands. This matters to channel partners and corporate customers because AI infrastructure availability increasingly decides how fast providers can grow services, handle larger workloads, and sustain performance as usage climbs. Organizations that can lock in the most computing capacity may secure a considerable edge as customers choose which AI platforms can dependably support production deployments.
There's a trade-off involved. Anthropic is locking in multibillion-dollar obligations for infrastructure and processors that won't be completely operational for years. Gains in model efficiency, changes in chip costs, or slower-than-projected revenue expansion could render those commitments more costly than expected. The broader signal is that AI leadership is evolving into as much an infrastructure battle as a software competition. Model quality remains important, but so do access to electricity, cutting-edge chips, data center capacity, and the capital required to obtain all three before rivals can. Companies that can afford to reserve infrastructure at this scale may find themselves with operational flexibility that competitors simply can't match, while those that overcommit risk financial strain if technology or market dynamics shift faster than anticipated.