Elon Musk's xAI to Become Anthropic's Compute Provider at $1.25 Billion per Month

Anthropic has agreed to pay xAI — Elon Musk's company — $1.25 billion per month for access to computing resources. The unexpected industry deal emphasizes how fiercely AI companies compete for compute — the scarcest resource in the modern artificial intelligence market.

Colossus: A Cluster Powerful as an Entire City

xAI owns one of the world's most powerful computing clusters — Colossus, consisting of 100,000 H100 GPUs. These graphics processors are the foundation for training large language models. A single H100 costs approximately $40,000, so equipment procurement for Colossus alone cost several billion dollars. These are the most expensive and powerful chips on the market.

But equipment costs are only part of the expense. The cluster consumes as much electricity as a small American city. xAI invests billions in building its own power plants and purchasing long-term electricity contracts from suppliers. Compute is not just servers in a data center — it's an entire ecosystem: electricity generation and delivery, facility cooling, network infrastructure, 24/7 technical support. Colossus data is rarely publicly disclosed, but it's believed to be one of the three largest clusters in the world alongside Microsoft and Google resources. Control over such power gives Musk and xAI enormous influence on the development of the AI industry.

Why Anthropic Pays $15 Billion Per Year

Anthropric faces fierce competition from OpenAI (which has access to Microsoft resources) and Google (with its own TPUs — tensor processing units). For a long time, Anthropic was severely limited in compute access and couldn't train models at the same scale as its competitors. This was a serious strategic disadvantage.

The xAI deal is an attempt to close this gap. Direct access to H100 GPUs without intermediaries is critical for developing the next versions of Claude. $1.

25 billion per month ($15 billion per year) is a massive sum, but for an AI leader, it's the price of entry into the race for dominance. Without control over compute, Anthropic simply cannot develop competitive next-generation models on par with OpenAI and Google. For comparison: Microsoft pays billions annually for cloud resources.

Now Anthropic is also joining this club of companies willing to spend tens of billions to dominate AI.

Compute Shortage as a Strategic Problem

Currently, H100 GPUs cannot be purchased on the open market in the required volumes. They are controlled by cloud providers (AWS, Google Cloud, Azure) and major AI companies. This creates a bottleneck for the entire industry. The Anthropic deal shows that cloud resources alone are insufficient. Companies need long-term contracts and direct partnerships with GPU owners to guarantee access for years ahead. This transforms computing power into a strategic asset, similar to oil fields in the energy sector.

• Compute has become the scarcest resource in AI
• Cloud providers cannot meet the demand of all companies
• Long-term contracts are the new norm for AI companies

What This Means

The deal between Anthropic and xAI signals a transition of the AI race into a new phase. Now, ownership of computing power is more critical than algorithm development. Companies that secure long-term access to GPUs and electricity will win. Others will either seek partners, optimize models for less compute, or exit the market.