1 hour ago
1
xAI, Musk’s artificial intelligence company, has leased the entire compute capacity of its Colossus 1 supercomputer in Memphis to Anthropic, the maker of Claude. The deal is worth approximately $1.25 billion per month, with a potential total value exceeding $40 billion if extended through May 2029.
A supercomputer looking for a purpose
Colossus 1 is no small operation. The facility packs more than 220,000 NVIDIA GPUs and draws over 300 megawatts of power. It was built in a remarkable 122 days and became operational in mid-2024, originally designed to train xAI’s Grok models.
The supercomputer’s mixed GPU architecture turned out to be a poor fit for effectively training Grok. Rather than retrofit or mothball the facility, xAI pivoted to monetization. The leasing agreement, signed on May 6, 2026, makes Anthropic the sole tenant of Colossus 1. Anthropic plans to use the compute primarily for inference tasks, which is the process of running trained AI models to generate responses, rather than training new ones from scratch.
Training requires all GPUs working in tight coordination on the same task. Inference is more forgiving of mixed hardware configurations, which makes Colossus 1’s architecture far better suited for Anthropic’s needs than it was for Grok’s.
Anthropic will channel that compute power toward serving its Claude Pro and Claude Max subscribers.
The money and the fine print
Musk confirmed on May 28, 2026, that the initial lease commitment is only six months in duration. Six months at $1.25 billion per month comes to roughly $7.5 billion. The possibility for extensions remains open, but nothing beyond the initial period is locked in.
What this means for the AI compute market
This deal confirms that raw GPU supply is now so scarce that rivals will do business with each other rather than go without. Anthropic and xAI compete directly in the large language model space, with Claude and Grok targeting many of the same users, yet here they are, landlord and tenant.
At roughly $1.25 billion per month for 220,000-plus GPUs, the market now has a reference point for what top-tier inference capacity costs at scale.
There is already speculation about a separate deal for Google to rent capacity at xAI’s planned Colossus 2 facility.
The risk for Anthropic is dependency. Building your inference stack on a competitor’s hardware creates obvious strategic vulnerabilities. If xAI decides it needs Colossus 1 back after six months, or if lease renewal negotiations get contentious, Anthropic could face a capacity crunch at the worst possible time.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) · 