Anthropic has signed a roughly $45 billion cloud agreement with UK-based AI infrastructure company Nscale, according to people familiar with the deal. The agreement will give Anthropic access to around 460MW of compute capacity at an Nscale data center development in West Virginia. The facility is expected to come online toward the end of 2027 and will use Nvidia’s advanced Vera Rubin chips. Bloomberg first reported the agreement, with CNBC subsequently confirming the deal.

The scale of the commitment underscores how quickly compute requirements are expanding for frontier AI companies. Anthropic has been expanding its infrastructure footprint as demand for its Claude models and products puts increasing pressure on available capacity. The company said earlier this year that growing demand had caused “inevitable strain” on its infrastructure, affecting “reliability and performance” for users, particularly during peak hours. The Nscale agreement gives Anthropic a major future capacity block as it prepares for another phase of Claude growth.

Anthropic Locks In 460MW For Future Claude Demand

The Nscale agreement centers on roughly 460MW of compute capacity, making it one of the largest infrastructure commitments disclosed by Anthropic. The capacity will come from an Nscale development in West Virginia, giving Anthropic access to a dedicated future footprint rather than relying solely on existing cloud availability. The planned deployment will use Nvidia’s Vera Rubin generation of AI accelerators, connecting Anthropic’s future model workloads to Nvidia’s next-generation computing platform. Nscale expects the facility to become operational at the end of 2027.

The deal further expands Nscale’s role in the rapidly developing AI infrastructure market. Rather than simply providing conventional cloud resources, Nscale is positioning its infrastructure around the power, facilities and accelerator requirements of large AI workloads. Anthropic’s commitment gives that strategy a major customer with one of the industry’s most aggressive compute requirements. It also places West Virginia within Anthropic’s expanding network of infrastructure locations.

Claude’s Infrastructure Pressure Is Driving New Deals

Anthropic has spent much of 2026 expanding its infrastructure relationships as Claude adoption increases. The company has announced or pursued agreements involving Advanced Micro Devices, SpaceX, Google and Broadcom, adding different sources of compute, networking and infrastructure capacity to its broader strategy. Those partnerships give Anthropic more options as accelerator technologies change and demand becomes harder to forecast. The Nscale agreement adds another large-scale capacity source to that network.

Meanwhile, the company’s expanding infrastructure footprint shows how AI capacity planning is moving beyond conventional cloud procurement. Frontier AI developers increasingly need to secure accelerator supply, power, cooling, networking and physical data center space as a single strategic problem. Anthropic’s growing collection of infrastructure agreements reflects that shift. The company is effectively building a distributed compute supply chain around Claude.

Nvidia Vera Rubin Becomes Part Of Anthropic’s Roadmap

Anthropic’s planned use of Nvidia’s Vera Rubin chips adds another layer to the Nscale agreement. The Rubin platform represents Nvidia’s next major generation of AI computing technology, giving Anthropic access to hardware designed for increasingly demanding model workloads. The decision suggests that Anthropic is planning its infrastructure several years ahead rather than simply expanding around the hardware available today. That approach becomes important as model complexity and inference requirements continue to rise.

The 2027 deployment window gives Anthropic time to align the new capacity with future model and product requirements. Large AI facilities require coordination across accelerator procurement, electricity supply, cooling systems, networking and construction, so capacity cannot appear immediately when demand rises. Locking in infrastructure years ahead can reduce the risk of reaching a point where model demand exceeds available compute. It can give Anthropic a more predictable foundation for its next generation of Claude services.

Anthropic’s IPO Puts Compute Spending Under Scrutiny

Anthropic’s infrastructure expansion comes as the company prepares for increased scrutiny from public-market investors. Anthropic confidentially filed its IPO prospectus with the Securities and Exchange Commission in June and has held preliminary meetings with prospective investors. The company is now approaching a period in which investors will examine not only Claude’s growth but the infrastructure costs required to sustain that growth. Compute commitments will form an important part of that financial picture.

The company faces pressure to justify a reported $965 billion valuation as competition from OpenAI and Google intensifies. Anthropic must demonstrate that its model demand can support the enormous infrastructure requirements associated with frontier AI. Large capacity agreements can strengthen the company’s ability to serve customers, but they can also create substantial long-term financial obligations. The balance between securing enough compute and maintaining attractive economics will become increasingly important as Anthropic approaches the public markets.

Anthropic’s Compute Strategy Is Becoming More Diversified

Anthropic’s infrastructure strategy now spans multiple major technology and infrastructure companies. Agreements involving AMD, SpaceX, Google, Broadcom and Nscale indicate that the company is spreading its requirements across different suppliers and infrastructure models. That diversification can give Anthropic greater flexibility as chip generations, cloud economics and deployment requirements change. It can reduce the risk of depending too heavily on a single source of compute.

Nscale brings a particularly large capacity commitment into that strategy. Around 460MW gives Anthropic a substantial future footprint that can support the computational requirements of increasingly capable Claude models. The exact division between training and inference workloads remains undisclosed. The size of the agreement nevertheless points to expectations of sustained and significant compute demand.

The $45 Billion Deal Signals A New Compute Era

Anthropic’s Nscale agreement illustrates how frontier AI companies are moving toward infrastructure commitments measured in hundreds of megawatts and tens of billions of dollars. Claude’s growth is creating requirements that traditional cloud capacity planning cannot easily satisfy. Anthropic is responding by securing infrastructure years before the West Virginia facility is expected to become operational. The strategy gives the company a larger future compute base as competition for advanced accelerators and power capacity intensifies.

For Anthropic, the immediate objective is straightforward: secure enough capacity to keep Claude growing without repeating the infrastructure strain the company has already acknowledged. The longer-term question is whether that enormous infrastructure commitment can translate into sustainable economics as the company competes with OpenAI and Google. As Anthropic moves toward a potential IPO, its infrastructure strategy will become a central part of the investment story. The $45 billion Nscale agreement puts that strategy into unusually sharp focus.