Anthropic Signs $45 Billion Nscale Compute Deal as AI Infrastructure Race Accelerates

Anthropic has agreed to rent about $45 billion in AI computing capacity from Nscale, securing infrastructure for the company’s expanding services. The six-year arrangement centers on Nscale’s flagship data center development in West Virginia, Bloomberg reported, and ranks among the largest disclosed commitments in the company’s rapid push for more computing power.

The deal gives Anthropic a long-term supply of capacity rather than a short-term burst of rented servers. TechCrunch reported that the systems are expected to begin powering Anthropic services in late 2027. The timing places the commitment alongside the company’s effort to scale its business and prepare for a possible public offering, while also tightening its race with OpenAI for model training and customer-serving capacity.

Nscale will provide the computing power through Nvidia’s Vera Rubin chip systems, according to TechCrunch. Vera Rubin combines 6 different chips in a coordinated system, making the hardware a central part of the arrangement rather than a generic cloud supply contract.

The choice of Nvidia hardware links Anthropic’s capacity plans to the chipmaker’s newest platform and to the physical buildout of specialized AI data centers. Nscale, founded in 2024, has also reached infrastructure agreements with Microsoft, showing how quickly newer providers are entering a market shaped by demand from the largest AI companies. Bloomberg’s account places the capacity at Nscale’s West Virginia project, giving the deal a defined site as well as a multiyear financial scale.

The Nscale agreement is the latest in a series of large infrastructure commitments by Anthropic. The company signed a $10 billion, 6-year arrangement with AI cloud startup Volta for capacity from a data center in Norway, followed by a $5 billion compute-related deal with AMD in July.

Anthropic also expanded access through SpaceX data centers in May. That arrangement was reported to provide about $1.25 billion in capacity each month, while an April agreement with Amazon added 5 gigawatts of compute. Anthropic expanded its relationship with Google and Broadcom during the same month, adding further capacity across cloud and custom-chip infrastructure. Together, the deals show a strategy built around several suppliers and locations rather than dependence on a single cloud platform.

Anthropic’s spending reflects a contest in which computing access has become a core competitive resource. The company has sharply increased capacity over the past 8 months as it competes most directly with OpenAI, while other model developers pursue their own chip, cloud and data center arrangements.

Long-term reservations help AI companies plan model training, inference and enterprise service expansion, but they also commit them to substantial fixed spending. The Nscale contract starts service in late 2027, so its value depends on Anthropic sustaining demand for Claude products and future systems over several years. Nvidia gains another major customer for its newest platform, while cloud and infrastructure providers gain a large anchor tenant for facilities designed around AI workloads.

Nscale’s West Virginia development places a major Anthropic computing commitment in a state better known for energy and industrial infrastructure than for frontier AI. The project illustrates how model growth is pulling investment toward large facilities that need substantial power, cooling, networking and chip supply.

The physical location matters because AI capacity requires more than access to software interfaces. Data centers must be built, connected to power and equipped with tightly integrated accelerator systems before they can serve customers. Anthropic’s agreement gives Nscale a large customer commitment, while the late-2027 start date leaves time for construction, equipment deployment and systems integration.

Anthropic’s immediate compute needs will continue to be served through its existing and recently expanded relationships with Amazon, Google, Broadcom, AMD, SpaceX and Volta. The Nscale capacity is scheduled to come online in late 2027, extending the company’s infrastructure planning well beyond its current wave of expansion.

The agreement also gives investors and competitors a clearer view of the capital required to operate frontier AI services at scale. Anthropic has secured capacity across multiple providers, geographies and hardware platforms, but each commitment raises the financial and operational stakes of growth. The company’s next test is converting that reserved computing power into reliable products and sustained revenue.