Anthropic to Spend $45 Billion on SpaceX Compute to Scale Claude

Anthropic has agreed to pay SpaceX $45 billion over three years for computing resources for its Claude AI assistant. The deal is confirmed in an official securities filing and reflects the scale of investment required to compete in the artificial intelligence market.

The scale of the deal

This is one of the largest deals in the history of relations between AI companies and infrastructure providers. By comparison, even the most ambitious cloud computing contracts rarely exceed a few billion dollars a year. But training and running modern LLMs demands capacity of a completely different order: hundreds of thousands of chips working in sync.

Anthropic is growing exponentially: Claude competes with OpenAI's GPT-5, its user base is expanding, and API integrations appear in new services every month. Every jump in traffic requires a proportional jump in infrastructure. The $45 billion is an investment not only in current capacity but also in a reserve for the coming years, insurance against demand spikes.

The contract with SpaceX is especially interesting because the company had not previously positioned itself as a cloud provider. Elon Musk, meanwhile, owns xAI, a direct competitor of Anthropic in the AI market. Still, the deal shows that in a hypercompetitive industry, business interests overlap even between financial rivals. This is not an alliance but pure commercial calculation: SpaceX has spare computing capacity, Anthropic is willing to pay, and the contract is signed.

Why computing is so expensive

Modern large language models are not just programs but enormously resource-hungry systems. Training a new generation of LLMs takes hundreds of thousands of hours of continuous GPU and TPU work, on specialized processors that cost tens of thousands of dollars apiece.

But training is not the only cost. Once in production, Claude has to serve requests from millions of users in real time, 24/7. Each of your queries is processed across thousands of compute cores in parallel. This creates constant, growing demand for capacity.

Anthropic has several options: build its own data centers (capital-intensive and takes years), rent capacity from giants such as AWS or Google Cloud (expensive and competitive, since big IT companies consume most of their own capacity first), or sign a long-term contract with an independent provider. SpaceX offered a third path: guaranteed computing capacity for three years at a fixed price, which gives Anthropic long-term predictability and guarantees infrastructure scale at the same time.

What this means

The deal is a signal of the real cost of the AI war in 2026. Not blockchain, not NFTs, not the metaverse: the real money in the industry goes to hardware, meaning servers, GPUs and energy. Companies are ready to pay tens of billions a year just for access to compute.

For SpaceX, it means stable long-term revenue and a new business segment beyond space. For Anthropic, it is a critical advantage: controlling compute means controlling the pace of development and the scale of operations.

For the industry, it is a sign that in the coming years AI leadership will be won not by the companies that code best, but by those that can afford to pay billions for hardware and electricity.