The Anthropic-SpaceX compute deal that AI Insiders covered on May 22 as a $45 billion, three-year commitment has a complication: Elon Musk publicly described it this week as a 180-day lease with a 90-day mutual cancellation clause, a characterization that directly contradicts SpaceX’s own S-1 filing.

TechCrunch, reporting May 28, surfaced Musk’s post on X in which he wrote that SpaceX “has not committed to leasing Colossus for years.” He described the structure as a 180-day initial lease followed by 90-day mutual termination rights, adding that the short term was SpaceX’s preference and that the company would provide Anthropic a “reasonable off-ramp” if needed.

SpaceX’s S-1 filing says something different. Page F-62 states that Anthropic “has agreed to pay a monthly fee through May 2029” under a cloud services agreement signed May 3, 2026. The same language appears on pages F-96, 13, and 146, with one instance specifying “$1.25 billion per month through May 2029.” That is not one typo; it is a consistent disclosure across four pages of a document filed with the SEC during a public offering process.

The gap between those two framings matters for more than one reason.

IPO pricing and revenue quality. SpaceX’s S-1 presents Anthropic as a major recurring-revenue customer through May 2029. Public market investors reading that filing would price SpaceX’s revenue visibility accordingly. A deal that either side can exit in 90 days is structurally different from a three-year contract, regardless of the monthly dollar figure. The headline $45 billion is the product of $1.25 billion per month times 36 months; it only holds if all 36 months close.

Anthropic’s compute flexibility. A 90-day termination right is actually worth something to Anthropic. The company is building on Colossus while AWS invests heavily in Trainium chips, while Microsoft scales Maia capacity, and while DeepSeek’s efficiency gains continue to compress the cost per useful compute unit. An exit right lets Anthropic respond to any of those shifts on a single quarter’s notice rather than absorbing the cost of a locked multi-year term.

The source problem. TechCrunch’s confirmation of the 180-day structure comes from “people familiar” with the agreement, not from the contract itself. Musk’s public statement on X is the kind of performative framing he routinely uses to shape narrative around his companies. SpaceX’s S-1 is the legally accountable disclosure, drafted by counsel and filed during a quiet period when material misrepresentations carry formal risk. Each of those three sources has an incentive to frame the deal favorably to their own position: the people familiar may want the deal characterized as flexible to protect Anthropic’s optionality, Musk may want it characterized as short-term to reduce the impression of dependence on a competitor-adjacent partner, and SpaceX’s lawyers want it characterized as long-term to support the revenue multiple. TechCrunch reporter Russell Brandom noted plainly that a misrepresentation made while marketing a security “is bad karma at the very least,” and that the SEC “probably will not do anything.”

The legally filed text says Anthropic has agreed to pay through May 2029. The cancellation clause is also there, on page F-62. Both things are simultaneously true, and both matter.

Enterprise procurement teams reading AI infrastructure contracts should now treat headline duration and cancellation terms as separate line items. A $45 billion agreement with a 90-day exit window is closer in risk profile to a rolling month-to-month arrangement than to a capital commitment. If your procurement model is calibrated to a multi-year duration, the cancellation clause is the number to read first.

Reported by TechCrunch on 2026-05-28.