Anthropic inks $45B cloud deal with Nscale

Anthropic commits $45 billion to Nscale cloud deal

Anthropic has agreed to spend roughly $45 billion to rent artificial intelligence cloud computing power from Nscale. The agreement centers on a large, dedicated capacity reservation rather than building facilities directly. It reflects Anthropic’s need for sustained access to high-performance infrastructure to support its AI systems. The deal is described as a single, large-scale commitment rather than a series of smaller arrangements.

Under the arrangement, Anthropic will rely on Nscale as a core provider of compute resources. The contract is structured around a specific data center development, giving Anthropic clarity about where its workloads will run. The large financial value underscores the cost of securing state-of-the-art AI compute at scale. It also highlights the importance of long-term planning for access to specialized chips and power.

Scale of compute at West Virginia data center

The agreement covers about 460 megawatts of compute capacity at an Nscale data center development in West Virginia. This single site will serve as a major hub for Anthropic’s AI computing needs. Concentrating such capacity at one development underscores the scale of infrastructure required for advanced AI training and inference. The power allocation is tied specifically to this facility rather than spread across multiple regions.

Locating the contracted capacity at a dedicated development allows Nscale to design the facility around Anthropic’s workloads. The site is being built to accommodate intensive AI operations that demand significant power and cooling. For Anthropic, anchoring the deal in a named development provides greater certainty over the physical footprint behind its cloud resources. It also aligns data center planning with the anticipated growth of AI compute requirements.

Use of Nvidia’s Vera Rubin chips

Nscale will deploy Nvidia’s (NVDA) Vera Rubin chips at the West Virginia facility to serve Anthropic’s needs. These chips are designed for high-end AI work, positioning the site for demanding training and inference tasks. The hardware choice indicates that the contracted capacity is intended for cutting-edge model development rather than general-purpose computing. Anthropic’s workloads at the site are expected to run on this specialized Nvidia platform.

The Vera Rubin chips at the facility are expected to start coming online in late 2027. This timing sets a clear horizon for when significant new compute capacity will be available to Anthropic. The phased arrival of hardware means Anthropic can plan future model development around the expansion of this infrastructure. It also marks late 2027 as a key milestone for the practical ramp-up of the $45 billion commitment.

Strategic implications of the infrastructure deal

By securing such a large block of capacity, Anthropic is tying its future AI development closely to Nscale’s West Virginia buildout. The agreement gives Anthropic a long-term path to scale its compute usage without constructing and operating its own data centers. For Nscale, the contract anchors demand for its flagship development and supports the deployment of Nvidia’s Vera Rubin chips at substantial scale. The deal illustrates how major AI companies are locking in specialized infrastructure to meet growing computational needs.