Musk says SpaceX-Anthropic Colossus lease only runs six months
TL;DR:
- Elon Musk says the SpaceX-Anthropic deal that was previously framed as a $1.25 billion-a-month commitment through May 2029 is in fact a 180-day lease with mutual 90-day cancellation rights, with no guarantee of extension.
- SpaceX’s IPO filing last week disclosed the cancellation terms but did not mention the six-month duration; Musk made the disclosure on X, saying “the short term was our request, not Anthropic’s”.
- For UK AI policy watchers, the episode is a reminder that frontier-AI compute commitments are politically contestable, not durable infrastructure — and that the UK’s reliance on overseas compute for the same workloads carries similar fragility.
SpaceX CEO Elon Musk has contradicted the framing of one of the largest publicly disclosed AI compute arrangements ever announced, saying the company’s deal with Anthropic to provide Colossus and Colossus II data centre capacity in Memphis is only a 180-day lease — not a multi-year commitment. SpaceX filed for IPO last week disclosing that Anthropic was paying $1.25 billion a month for capacity that ran through May 2029, with both companies able to terminate with 90 days’ notice.
What Musk actually said
In a post on X, Musk wrote: “SpaceX has not committed to leasing Colossus for years, although it’s possible that may be what happens.” He went on to clarify that the agreement is a 180-day lease with mutual 90-day cancellation thereafter, adding: “The short term was our request, not Anthropic’s. 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.”
The IPO filing did not mention the six-month duration. SpaceX’s AI segment lost about $2.5 billion from operations in the March quarter on segment revenue of $818 million.
Why this matters
The Musk-Anthropic compute relationship has been one of the most-watched data points for understanding frontier AI’s true infrastructure costs and timelines. The earlier framing — multi-year commitment, $1.25 billion a month, through 2029 — supported a particular narrative about durable AI compute pricing. The revised picture is materially less stable: Anthropic has only six months of guaranteed capacity, with extension subject to SpaceX’s own internal needs.
That uncertainty is structural for Anthropic, but it also reveals something broader about the politics of frontier AI. The largest compute deals are now mediated by individuals — Musk publicly negotiating duration on X, after the SEC filing — rather than by stable commercial frameworks. Last week Musk separately posted that SpaceX was “in discussions with other companies about offering AI compute as a service at significant scale”.
UK implications
UK AI firms increasingly rely on overseas compute — including Microsoft’s Loughton supercomputer (covered separately this week) — for the same kinds of workloads Anthropic is running on Colossus. The fragility of compute commitments matters for the UK’s AI Opportunities Action Plan, which assumes UK AI scale-ups can secure stable compute access. If even a $15 billion-a-year customer like Anthropic can have its compute timeline shortened by an X post, smaller UK firms should plan accordingly.
Looking forward
How Anthropic responds — by diversifying compute providers, building closer relationships with hyperscalers, or pursuing its own infrastructure — will shape the next year of frontier AI economics. For UK observers, the question is whether the Strategic Compute Programme and forthcoming UK AI infrastructure investments can offer the kind of stability that the Colossus episode shows is no longer available from individual operators.