Anthropic is reportedly assembling a US$45 billion capital pool dedicated to securing long-term AI computing rentals and infrastructure access.
Anthropic is reportedly preparing a US$45 billion commitment to lease AI computing capacity from infrastructure firm Nscale over a six-year period. The agreement, designed to support the development and operational requirements of Anthropic’s AI models, will be fulfilled through Nscale’s data center complex in West Virginia, United States. On Wednesday, August 26, 2026, Reuters—citing a Bloomberg report—outlined the deal, though Anthropic has not yet provided official confirmation regarding its specific terms.
The US$45 billion figure represents the total value of the lease commitment rather than a lump-sum payment. Spread evenly across the six-year term, it equates to approximately US$7.5 billion annually, though this is merely a simple average as the actual payment schedule remains unpublished. This distinction is critical: capital expenditure for constructing data centers differs fundamentally from computing service leases. Under this arrangement, Anthropic secures access to server capacity and supporting infrastructure without assuming outright ownership of the facilities.
The reported agreement encompasses approximately 460 megawatts of capacity at Nscale’s West Virginia site. This substantial power requirement underscores the immense computational demands required to train AI models, execute inference processes, and serve daily user traffic for Anthropic’s products. Nscale is reportedly deploying systems based on Nvidia’s Vera Rubin-generation chips. These AI accelerators are essential to modern data centers, engineered to handle massive volumes of parallel processing workloads. However, specifics regarding capacity activation timelines, server delivery phases, and usage targets have not been disclosed; the 460-megawatt figure should therefore be understood as contracted capacity rather than currently operational throughput.
Anthropic develops the Claude family of models, which power conversational interfaces, document analysis, code generation, and various enterprise applications. The expanding adoption of services like Claude and Claude Code directly drives computing demand, both for iterative model development and real-time user request processing. Across the broader AI sector, developers are engaged in intense competition to secure advanced chips, reliable power supplies, efficient cooling systems, and viable data center land. While long-term contracts provide capacity certainty for model builders, they also lock in significant financial obligations should demand growth fall short of projections.
The reported US$45 billion Nscale lease is distinct from Anthropic’s previously announced US$50 billion investment in domestic AI infrastructure, revealed in November 2025. At the time, Anthropic projected that the infrastructure initiative would generate approximately 800 permanent positions and 2,400 construction jobs, with multiple facilities slated to enter phased operation throughout 2026. These two figures should not be conflated as realized expenditures: the US$45 billion pertains to the six-year lease arrangement with Nscale, while the US$50 billion represents a separate, earlier-announced capital deployment strategy.
As of publication, Anthropic reportedly declined to comment. With no official disclosure from either Anthropic or Nscale detailing the full contract terms, information regarding valuation, duration, and technical specifications continues to rely on media reporting. Should the deal proceed as reported, it would rank among the largest computing lease commitments in the AI industry. The agreement further illustrates that competition among AI developers now extends beyond model architecture and performance, increasingly hinging on the ability to secure long-term energy and data center infrastructure.