Anthropic Leases the Full Capacity of SpaceX's Colossus 1 Data Center, Adding 300 MW and Over 220,000 Nvidia GPUs to the Claude Backend
The May 6 deal hands Anthropic all of xAI's Memphis supercomputer and unlocks immediate rate-limit increases for Claude Code, Claude Pro, Max, and the Opus API.
Editor's Note ·
- Clarification:
- Two direct quotes attributed to Yahoo Finance — 'Everyone I met was highly competent and cared a great deal about doing the right thing' and 'SpaceXAI had already moved training to Colossus 2' — cannot be independently verified from the local source snapshot because Yahoo Finance redirected to consent.yahoo.com. Tom's Hardware paraphrases similar substance ('senior members of the company [...] understand what they do to ensure Claude is good for humanity'; 'as the company focuses on building Colossus 2') but the exact verbatim wording is not in any of the three openly-fetched snapshots. The substance of the claims is corroborated; the specific verbatim attribution is unverifiable.
Overview
Anthropic has signed a deal with SpaceX to take over the full compute capacity of Colossus 1, the Memphis, Tennessee supercomputer that xAI built and operated, in an arrangement announced on May 6 that gives Claude immediate access to more than 300 megawatts of new capacity and over 220,000 Nvidia GPUs, according to Anthropic. The capacity will come online “within the month” of the announcement, the company said. As part of the same agreement, Anthropic also expressed interest in working with SpaceX to develop multiple gigawatts of orbital AI compute capacity over the longer term.
What We Know
The lease covers the entirety of Colossus 1, which is located in Memphis, Tennessee and houses more than 220,000 Nvidia processors, as reported by Al Jazeera. The 300-megawatt figure and 220,000 GPU count were also confirmed in Tom’s Hardware’s coverage of the deal.
Elon Musk, who owns SpaceX and merged xAI into the company earlier this year, addressed the seemingly counterintuitive decision to hand a flagship facility to a direct rival. He explained that “SpaceXAI had already moved training to Colossus 2,” leaving Colossus 1 available, and that he had personally vetted Anthropic’s leadership. “Everyone I met was highly competent and cared a great deal about doing the right thing,” he said, per Yahoo Finance. He added “No one set off my evil detector,” a line that surfaced both in Tom’s Hardware and Al Jazeera.
Anthropic, led by Dario Amodei, is using the new capacity to immediately raise customer-facing limits, as reported by Al Jazeera. The company said it is doubling Claude Code’s five-hour rate limits across Pro, Max, Team, and seat-based Enterprise plans, removing the peak-hours limit reduction on Claude Code for Pro and Max accounts, and raising API rate limits considerably for Claude Opus models, according to Anthropic.
The orbital ambition is the more speculative half of the announcement. Anthropic said it had expressed interest in partnering with SpaceX to develop “multiple gigawatts of orbital AI compute capacity,” according to Anthropic, with no timeline attached. Al Jazeera framed the orbital component as aligned with one of Musk’s longer-running goals.
What We Don’t Know
The announcement did not disclose financial terms — neither Anthropic nor SpaceX has stated what Anthropic is paying for the Colossus 1 lease, what portion is upfront versus consumption-based, or how long the agreement runs. The exact mix of GPU generations inside Colossus 1 was also not specified in either company’s announcement, beyond Nvidia branding.
The orbital compute strand carries even less detail: no design, no launch cadence, no candidate satellite bus, and no first-customer milestone were named. Anthropic’s wording — “expressed interest” — is a memorandum-grade commitment, not a binding plan.
The broader competitive picture is also unsettled. The Al Jazeera account noted that Musk is currently suing OpenAI and its chief executive, Sam Altman, alleging that the company breached its nonprofit mission — context that helps explain Musk’s willingness to supply Anthropic but not directly Anthropic’s strongest rival. How that lawsuit and its outcome ripple back into compute-allocation decisions across the AI sector remains to be seen.