Anthropic is paying Elon Musk’s SpaceX $1.25 billion a month for the entire compute capacity of the Colossus 1 data center in Memphis — 220,000 NVIDIA GPUs across 300 megawatts. The Anthropic-SpaceX compute deal, which runs through May 2029, has already translated into real benefits for Claude users: rate limits are up, and the frustrating peak-hours throttle is gone.
Anthropic announced the partnership on May 6, rolling out three immediate changes for paid subscribers. Claude Code’s five-hour rate limits are now doubled for Pro, Max, Team, and seat-based Enterprise plans. The peak-hours limit reduction for Claude Code — the cap that would quietly halve your throughput during busy periods — has been removed entirely for Pro and Max users. And Anthropic has significantly raised its API rate limits for Claude Opus models, giving developers more headroom for production workloads.
The $15 Billion-a-Year Compute Contract Nobody Saw Coming
The financial terms became public on May 20, when SpaceX filed its S-1 IPO prospectus with the SEC. Anthropic is paying approximately $1.25 billion per month through May 2029, with a discounted rate for the first two months as xAI completes its ramp-up. The deal could hand xAI over $40 billion in total revenue, according to reporting by TechCrunch.
Anthropic is effectively renting xAI’s entire Colossus 1 data center — the same cluster Musk’s company built to train its Grok models. The Memphis facility runs more than 220,000 NVIDIA GPUs including H100, H200, and next-generation GB200 accelerators, connected across over 300 megawatts of power capacity.
The move has given xAI a hybrid position in the AI market. SpaceX calls it a “dual monetization strategy” in its IPO filing — building compute for itself while selling capacity to others when utilization runs low. TechCrunch calls this model a “neocloud.” The timing helps explain why Grok usage has dropped significantly throughout 2026: those idle servers needed a buyer.
Musk’s Public Exit Clause
Elon Musk has been careful to frame the arrangement as short-term. “SpaceX has not committed to leasing Colossus for years, although it’s possible that may be what happens,” he wrote on X after the IPO filing. He clarified the structure: a 180-day initial lease followed by a mutual 90-day cancellation provision. Either party can walk away on three months’ notice.
“The short term was our request, not Anthropic’s,” Musk added in a follow-up post on June 1. Then came the line that should matter to anyone depending on Claude: “We won’t leave them hanging and will provide a reasonable off-ramp, but if compute gets super tight I said we might need it back at some point.”
SpaceX’s AI division posted an operating loss of roughly $2.5 billion in Q1 2026 while generating $818 million in segment revenue, per its IPO filing. If xAI’s own compute needs spike — say, if Grok usage reverses — Colossus could get recalled.
Why Anthropic Needed 220,000 GPUs, and What Comes Next
Rate limits on Claude weren’t a quirky policy choice — they were a real constraint. Users on paid plans were hitting five-hour ceilings, and high-demand coding sessions got throttled during peak hours. As Claude Code became a serious competitor to GitHub Copilot and Cursor, those caps were a competitive liability.
The SpaceX deal bridges the gap while Anthropic’s larger builds complete. The company also has a deal for up to 5 gigawatts with Amazon (nearly 1GW online by end of 2026), a 5GW partnership with Google and Broadcom (2027), a $30 billion Azure arrangement with Microsoft and NVIDIA, and a $50 billion U.S. infrastructure commitment with Fluidstack. This is part of a broader AI infrastructure arms race — Anthropic’s $65 billion funding round from earlier this year was specifically cited as fuel for this kind of infrastructure investment.
Beyond raw compute, Anthropic has expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity. That’s still at the interest stage, but it signals where Anthropic sees the infrastructure ceiling eventually landing.
Check out our Claude Opus 4.8 review for a look at the model powering this. For anyone on Claude Pro or Max, the headroom is real and available today. Whether it lasts through 2027 depends partly on how well Grok competes — which makes the Anthropic SpaceX compute deal one of the stranger business arrangements in recent tech history.
Frequently Asked Questions
What is the Anthropic-SpaceX compute deal?
Anthropic signed an agreement to use all of the compute capacity at SpaceX’s Colossus 1 data center in Memphis, Tennessee, giving it access to over 220,000 NVIDIA GPUs across more than 300 megawatts of power. The deal, announced May 6, 2026, immediately enabled Anthropic to raise Claude’s usage limits for paid subscribers.
How much is Anthropic paying SpaceX?
Anthropic is paying approximately $1.25 billion per month through May 2029, according to SpaceX’s S-1 IPO filing. The deal includes a discounted rate for the first two months and could bring xAI over $40 billion in total revenue over the contract period.
How do Claude’s new rate limits affect paid users?
Claude Code’s five-hour rate limit has been doubled for Pro, Max, Team, and Enterprise plans. The peak-hours reduction — which throttled Pro and Max users during busy periods — has been removed entirely. API users running Claude Opus models also received significantly higher rate limits as of May 6, 2026.
Can Elon Musk cancel the Anthropic compute agreement?
Yes. The initial term is 180 days, followed by a mutual 90-day cancellation provision. Musk wrote on X that “if compute gets super tight I said we might need it back at some point.” Both parties can exit the contract with 90 days’ notice at any time after the initial period.
Does Anthropic have other compute deals besides SpaceX?
Yes. Anthropic also has a deal for up to 5 gigawatts with Amazon (including nearly 1GW new capacity by end of 2026), a 5GW partnership with Google and Broadcom (coming online 2027), a $30 billion Azure arrangement with Microsoft and NVIDIA, and a $50 billion U.S. infrastructure commitment with Fluidstack. The SpaceX deal is the only one already delivering compute today.
Last Updated: June 2026
