Anthropic’s $45B Compute Deal Signals AI Arms Race

Anthropic Signs $45 Billion Compute Deal as AI Arms Race Reaches New Heights

The Claude maker has now committed over $60 billion to compute in 2026 alone — more than many countries spend on technology.

Anthropic has signed a deal to rent approximately $45 billion in AI compute from Nscale, a British AI infrastructure company founded just two years ago. The compute will come from Nscale’s flagship data center in West Virginia, powered by Nvidia’s Vera Rubin chips, with capacity expected to come online in late 2027.

The deal spans six years and represents the single largest compute commitment Anthropic has made — but it’s far from the only one.

Anthropic’s 2026 Spending Spree

To understand the scale of what’s happening, here’s Anthropic’s compute spending timeline this year:

• April: Expanded Amazon partnership for an additional 5 gigawatts of compute
• April: Expanded Google and Broadcom TPU deal
• May: Large deal with SpaceX data centers, reportedly worth $1.25 billion per month in capacity
• July: $5 billion compute deal with AMD
• August: $10 billion deal with Volta, an AI cloud startup founded in January 2026
• August: $45 billion deal with Nscale (this story)

That’s over $60 billion in compute commitments in a single year. For context, Anthropic’s own valuation is roughly $60 billion. The company is essentially betting its entire worth on having enough computing power to stay competitive.

Who Is Nscale?

Nscale was founded in 2024 and has quickly become a major player in AI infrastructure. The company also has deals with Microsoft, and its rise shows how the AI boom is creating entirely new categories of billion-dollar companies.

Two years ago, Nscale didn’t exist. Today, it’s signing $45 billion deals with one of the most well-funded AI labs in the world.

Why So Much Compute?

The simple answer: training and running frontier AI models requires enormous amounts of computing power, and that demand is growing faster than supply.

Anthropic’s Claude models compete directly with OpenAI’s GPT series, Google’s Gemini, and Meta’s Llama. Each generation of models requires significantly more compute than the last. The companies that secure the most compute capacity today will have the strongest models tomorrow — or at least that’s the bet.

The Risks

This level of spending carries real risk. If AI model improvements hit diminishing returns — if throwing more compute at the problem stops producing better results — these deals could become expensive liabilities. Some researchers have already warned about a “compute bubble” where companies overinvest based on projections that don’t materialize.

There’s also the question of revenue. Anthropic generates significant income from Claude subscriptions and API usage, but $60 billion in annual compute spending requires massive revenue growth to justify.

What This Tells Us

Regardless of whether these bets pay off, one thing is clear: the AI industry has entered an infrastructure arms race that dwarfs anything we’ve seen in technology before. The companies building AI aren’t just competing on algorithms and data — they’re competing on who can secure the most physical computing infrastructure.

For the rest of us, this means two things: AI capabilities will continue advancing rapidly, and the cost of that advancement is being measured in tens of billions of dollars.