Anthropic Will Pay xAI $1.25 Billion Per Month for Access to Colossus Data Centers Through 2029

Anthropic has agreed to pay xAI $1.25 billion per month through May 2029 for access to the Colossus and Colossus II data centers, according to SpaceX’s S-1 filing with the SEC published on May 20. The deal could generate over $40 billion in revenue for xAI across its full term, according to TechCrunch.

Anthropic confirmed the figures to WIRED. An Anthropic spokesperson validated the deal terms. Anthropic co-founder and chief compute officer Tom Brown said on X that the company is “scaling up on (Nvidia) GB200 capacity in Colossus 2 throughout June,” according to Axios.

The Numbers

The deal gives Anthropic access to 300 megawatts of compute from Colossus 1 (the data center’s entire output) plus capacity in Colossus 2. Together, the two facilities straddle Tennessee and Mississippi with more than one gigawatt of computing power, according to WIRED.

Anthropic pays a reduced rate for May and June while xAI completes its ramp-up, with the full $1.25 billion monthly rate taking effect after that. Either side can terminate the contract with 90 days’ notice.

At $15 billion annually, the deal dwarfs previous AI compute agreements. For context, Anthropic’s second-quarter 2026 revenue is expected to exceed $10 billion, according to The Wall Street Journal (cited by WIRED). The company is spending more on compute from a single provider than most AI labs generate in total revenue.

Why xAI Is Selling

xAI built the Colossus facilities for Grok, its flagship AI assistant. But Grok’s usage has dropped significantly in recent months, according to Forbes, freeing up servers that xAI now sells to a direct competitor.

The SpaceX S-1 frames it as strategy rather than surplus: “We believe our dual monetization strategy provides multiple pathways to generate returns on invested capital.” The filing also states SpaceX expects to “enter into additional similar services contracts,” according to TechCrunch.

TechCrunch described this as a “neocloud” model, where AI companies that overbuilt infrastructure offset costs by acting as cloud providers. Most players either build data centers for themselves or for others. xAI is doing both simultaneously, selling excess GPU capacity to Anthropic while continuing to train its own models.

The SpaceX IPO Context

The deal surfaced through SpaceX’s IPO paperwork. SpaceX is pursuing the largest IPO in history, targeting a roughly $75 billion raise at a $1.75 trillion valuation, according to WIRED. The company plans to list on Nasdaq under the ticker SPCX, with reports suggesting a debut as early as June 12.

SpaceX (including X and xAI) generated nearly $4.7 billion in revenue and lost $4.3 billion in Q1 2026. For 2025, the combined entity posted $18.7 billion in revenue against $4.9 billion in losses, driven by heavy spending on AI infrastructure and rocket development.

Elon Musk said on X that SpaceX was “in discussions with other companies to do the same,” suggesting the Anthropic deal may be the first of several compute-for-revenue arrangements ahead of the public offering.

What This Means for Compute Economics

The $1.25 billion monthly figure crystallizes a reality that has been abstract until now: running frontier AI models at production scale requires infrastructure spending that rivals the GDP of small nations. Anthropic’s Claude powers coding tools, enterprise deployments, and third-party agent integrations through platforms like OpenClaw. That usage is growing fast enough that Anthropic needs an entire data center’s output from a competitor just to keep up with demand.

The deal also reveals the fragility of the current compute market. xAI overbuilt because it expected Grok adoption to scale. It didn’t. Now that excess capacity flows to a competitor. If Anthropic’s demand plateaus or the 90-day termination clause gets exercised, xAI loses its largest revenue line heading into a public market debut.