Anthropic's SpaceX compute deal: what's confirmed about the $1.25 billion a month, and what's only reported

Anthropic's deal to rent computing power from SpaceX is back in the headlines. Around October 9–10, 2026, the Wall Street Journal published a profile of Anthropic co-founder Tom Brown, describing him as the person who brokered a compute deal worth $1.25 billion a month with Elon Musk's company after a tense period with the White House in June. The monthly figure isn't new, and it isn't only reported: SpaceX disclosed it in its own IPO filing in May. Other parts of the story, including how the deal came together, rest on anonymous sourcing.

This story separates the two. It is based on the companies' announcements, SpaceX's S-1 and reporting by TechCrunch, Axios and the Wall Street Journal (via summaries, because the profile is paywalled), checked October 12, 2026.

The short version

• Announced (May 6, 2026): Anthropic will use "all of the compute capacity" at SpaceX's Colossus 1 data center in Memphis: more than 300 megawatts and over 220,000 Nvidia GPUs.
• Filed with the SEC (May 20, 2026): SpaceX's S-1 says Anthropic agreed to pay $1.25 billion per month through May 2029, and that either party can end the agreements on 90 days' notice.
• Reported (WSJ, October 2026): Tom Brown, Anthropic's compute chief, met Musk in March and later helped resolve a June standoff over two Anthropic models.
• Reported, not confirmed (Reuters via TheStreet, September 29): Anthropic's leaked IPO prospectus shows it could pay up to $84.5 billion through 2029 for xAI-supplied Nvidia capacity.

What did Anthropic and SpaceX announce?

On May 6, 2026, Anthropic said it had "agreed to a partnership with SpaceX that will substantially increase our compute capacity," and would use all of Colossus 1's capacity. It said that gave it "more than 300 megawatts of new capacity (over 220,000 NVIDIA GPUs) within the month," aimed at Claude Pro and Max subscribers.

The same day, Anthropic doubled Claude Code's five-hour rate limits on paid plans and raised Opus API rate limits; our Claude app explainer covers those plan limits.

SpaceXAI's own post confirmed the Colossus 1 access and the GPU count, including H100, H200 and GB200 accelerators, and said Anthropic had "expressed interest" in partnering on "multiple gigawatts of orbital AI compute capacity." Neither announcement gave a price or a term.

What does SpaceX's IPO filing say?

SpaceX made its S-1 public on May 20, 2026. In a section titled "Compute Services Agreements with Third Parties," it says:

• "we entered into Cloud Services Agreements with Anthropic PBC"
• "the customer has agreed to pay us $1.25 billion per month through May 2029"
• "The agreements may be terminated by either party upon 90 days' notice."

TechCrunch, which cited the filing's financial-statement notes, reported that the agreement was signed on May 3, 2026, with capacity "ramping in May 2026 at a reduced fee," and that Anthropic keeps ownership of its models and data.

Simple arithmetic shows the scale: $1.25 billion a month is $15 billion a year. Run for the full three years, it would come to roughly $45 billion. That is an upper bound, though, since the early months were billed at a reduced fee and either side can walk away with three months' notice.

How long is the deal, really?

The filing and Musk don't quite agree. On May 28, Musk wrote on X: "This is a 180 day lease with 90 day notice mutual cancellation thereafter," adding that "the short term was our request, not Anthropic's." TechCrunch pointed out the gap between that and the S-1's "through May 2029" language. Neither company has clarified, and Anthropic's announcement doesn't give a duration.

Both accounts agree on the part that matters most: the arrangement can be ended on 90 days' notice. That makes it far less locked in than Anthropic's multi-year commitments to Amazon, Google and Microsoft, which our Anthropic explainer describes.

Musk has since said he wouldn't use that exit against Anthropic. In a July 9 post on X, as reported by Benzinga, he wrote, "I was clearly wrong about Anthropic," and "I would never cut them off in a way that hurt them badly, even as a competitor. That's not my style."

Is the deal bigger now?

Possibly, but only reported figures say so. In May, Brown wrote on X that Anthropic would be "scaling up on (Nvidia) GB200 capacity in Colossus 2 throughout June," according to Axios, which suggests the relationship extends beyond Colossus 1.

On September 29, Reuters reported on Anthropic's confidential IPO prospectus, which it had reviewed. According to TheStreet's summary of that report, the prospectus shows Anthropic could spend up to $84.5 billion through 2029 on Nvidia-based capacity supplied by xAI, most of it cancellable on 90 days' notice, as part of at least $518 billion in cloud and compute commitments across six partners. Anthropic hasn't published the prospectus or confirmed those numbers, and the S-1 language doesn't say whether the $1.25 billion monthly fee covers Colossus 2.

What did the Wall Street Journal report?

The Journal's profile of Brown is paywalled. According to summaries published October 9–10 by Shopifreaks and AI Weekly, the WSJ reported that:

• Brown, a Republican co-founder who runs Anthropic's compute deals, has become its main go-between with the Trump administration and Musk.
• He met Musk at SpaceX in March and argued that Anthropic was less "woke" than Musk believed, which cleared the way for the $1.25-billion-a-month deal.
• After the White House made Anthropic shut off two of its models in June, Brown ended a weeks-long standoff by convincing officials at the Commerce Department, the Pentagon and the White House that Anthropic had tightened its safeguards.

The June episode lines up with the public record. On June 12, 2026, the US government applied export controls to Claude Fable 5 and Mythos 5. Anthropic suspended access, the controls were lifted on June 30, and access returned on July 1. Brown's personal role is the Journal's reporting; Anthropic hasn't commented on it publicly.

Why it matters

The deal is a measure of how hard frontier labs are scrambling for compute. Anthropic is paying a direct competitor, the maker of the Grok models, about $15 billion a year, a sum Axios set against SpaceX's annual revenue of about $18 billion. It also makes SpaceX a cloud provider, one of the shifts our gigawatt datacenter tracker follows.

The 90-day exit cuts both ways: it limits Anthropic's liability, but capacity behind Claude's higher usage limits could, on paper, vanish within a quarter. Those compute costs sit behind the revenue figures discussed in our analysis of how Anthropic and OpenAI count revenue.

What's confirmed, reported and uncertain

• Confirmed (Anthropic and SpaceXAI, May 6): Colossus 1 access, more than 300 MW and 220,000+ GPUs, the rate-limit increases, and interest in orbital compute.
• Confirmed (SpaceX S-1, May 20): $1.25 billion a month through May 2029 and termination by either party on 90 days' notice.
• Stated by Musk (X): a 180-day lease, then 90-day mutual cancellation.
• Reported: the May 3 signing date and reduced ramp fee (TechCrunch, citing the S-1), the Colossus 2 expansion (Axios, citing Brown), the $84.5 billion and $518 billion prospectus figures (Reuters via TheStreet), and Brown's role (WSJ via summaries).
• Uncertain: the actual term, whether Colossus 2 is covered by the same fee, and how much Anthropic has paid so far.