Inside Anthropic’s $40B+ Deal to Rent xAI’s Colossus AI Supercomputer

That computing power is intended to support Anthropic’s rapidly growing AI services, particularly the Claude family of models, which require enormous GPU capacity for both training and inference workloads.

Large AI labs increasingly compete to secure compute infrastructure, because limited GPU supply and data‑center capacity can constrain how quickly models can scale.

Why the Deal Could Generate Over $40 Billion

The enormous headline value comes from a straightforward calculation.

• $1.25 billion per month × roughly three years (mid‑2026 through May 2029) produces a theoretical revenue run‑rate of around $45 billion.

However, many reports describe the deal as worth “over $40 billion” rather than the full $45 billion for several reasons:

• Some months may include initial discounts or ramp‑up pricing.
• The contract reportedly includes a 90‑day mutual termination clause, meaning either party could exit before the full term.

Because of those conditions, the multi‑year figure represents potential contract value rather than guaranteed revenue.

The Rise of the “Neocloud” Model

One of the most interesting aspects of the deal is what it says about xAI’s business model.

Rather than using all of its massive GPU clusters exclusively to train its own models, xAI is leasing infrastructure to other AI companies. This approach resembles a specialized AI cloud provider—sometimes called a “neocloud.”

In this model:

• Companies build enormous GPU‑dense data centers.
• They use part of the capacity internally to train their own models.
• Any surplus capacity is sold to other AI developers.

TechCrunch described the arrangement as turning xAI from a pure consumer of compute into a provider of it, effectively monetizing infrastructure it already built.

This strategy can help offset the enormous cost of building AI supercomputers, which require billions of dollars in GPUs, power infrastructure, and cooling systems.

Why xAI Had Capacity to Lease

Reports suggest that xAI shifted much of its own model training to a newer facility called Colossus II, leaving Colossus 1 available for external customers.

That made it possible for the company to lease the entire cluster without halting development of its own AI systems.

For Anthropic, the deal solves a different problem: securing enough compute to keep scaling Claude as demand rises.

The Grok Usage Context

The timing of the agreement also coincides with reports that xAI’s consumer chatbot, Grok, has struggled to maintain growth compared with rivals.

According to third‑party analytics cited in industry coverage:

• Grok’s app downloads reportedly fell nearly 60% between January and April 2026, dropping from over 20 million to about 8.3 million installs.
• Surveys also indicated very low paid adoption among U.S. users, suggesting limited monetization from the consumer product so far.

While these figures come from external analytics rather than official company data, they highlight why selling compute infrastructure can be strategically valuable.

If a company has already built huge GPU clusters, leasing them out ensures the hardware generates revenue regardless of how its own AI products perform.

What the Deal Signals About the AI Industry

The Anthropic–xAI agreement highlights several broader trends shaping the AI race:

1. Compute is becoming the key bottleneck.
AI labs increasingly compete to secure power, GPUs, and data‑center capacity as model sizes grow.

2. Infrastructure itself is becoming a product.
Companies with massive clusters can sell access just like traditional cloud providers.

3. Competitors may cooperate at the infrastructure layer.
Even companies building rival models may rely on each other’s compute capacity when supply is tight.

In that sense, the deal is not just a financial milestone. It also signals a structural shift: AI infrastructure may become as strategically important—and tradable—as the models themselves.