Anthropic SpaceX Compute Deal|$84.5 Billion Bet With 90-Day Exit?

The deal that doubled

Anthropic's leaked IPO prospectus shows compute deals with Elon Musk's SpaceX worth up to $84.5 billion. Yet of the $518 billion Anthropic has pledged for AI infrastructure, this is the only deal it can cancel on 90 days' notice. The rest of that money is bound far more tightly. Reuters, which saw the confidential filing, says about 80 percent must be paid whether Anthropic uses the capacity or not.

Anthropic's revenue in 2025 was about $4.6 billion. So the loosest contract sits with SpaceX, a company that also owns a rival AI lab. The tightest ones sit with Anthropic, which is seeking a valuation of about $2 trillion. Anthropic did not respond to Reuters' request for comment on the filing. SpaceX described this relationship first, in its own IPO paperwork.

It said Anthropic had agreed to pay $1.25 billion a month through May 2029. That paid for access to roughly 325,000 Nvidia GPUs across SpaceX's Colossus computing infrastructure. Over the full term, SpaceX put the value at nearly $45 billion. Anthropic's prospectus sets the ceiling at $84.5 billion. That is almost twice the figure investors had already seen. The size makes more sense next to Anthropic's own bills.

At $1.25 billion a month, the original SpaceX deal comes to about $15 billion a year. In 2025, Anthropic spent $7.33 billion on all of its computing and infrastructure. On a yearly basis, that one contract was worth about twice the company's entire compute bill. Anthropic tells investors its future growth could be "limited principally by the availability of compute."

Fortune reports the company had been racing to secure capacity after products like Claude Code drove a sudden surge in users.

Locked versus loose

With its other suppliers, Anthropic locked that supply in. It plans to spend at least $111.1 billion with Google. Another $110 billion is committed to Amazon. Those obligations run seven to ten years and are owed "regardless of usage," in the filing's words. If Anthropic's actual spending falls short of the Google commitment, it must pay the difference. Similar terms apply to Amazon.

Microsoft holds $31.4 billion under the same kind of obligation. Anthropic also carries about $161.2 billion in Broadcom-related equipment leases, which are largely non-cancelable. Take 80 percent of the total, and roughly $414 billion is owed no matter what happens to demand. Then there's SpaceX.

According to the prospectus as reported, only the xAI agreement, the capacity supplied through SpaceX, can be canceled on 90 days' notice. The Information's account agrees: the SpaceX agreements can mostly be canceled with three months' warning. The contract that nearly doubled is also the one Anthropic can leave the fastest. That clause matters more to SpaceX than it might first appear.

SpaceX acquired xAI, Musk's developer of the Grok chatbot, in February. The combination let SpaceX use its computing infrastructure internally and lease capacity to outside customers. Those customers include Anthropic and Google. And SpaceX is still expanding. Barron's reports a cluster of 220,000 Nvidia GB300 GPUs was due to come online within days. Another 220,000 were expected to be running by October.

Wall Street analysts increasingly see AI infrastructure as a major part of SpaceX's future. TD Cowen sees AI computing becoming SpaceX's biggest business. On the day the $84.5 billion figure was reported, SpaceX stock rose about 3 percent. But a customer who can leave on 90 days' notice is revenue that can shrink quickly.

The bigger that customer's deal gets, the more of that downside sits with SpaceX's shareholders rather than Anthropic's.

Can Anthropic pay

Anthropic's side of the ledger is the locked $414 billion. The 2025 numbers show how far the company was from covering it. Revenue rose twelvefold to about $4.6 billion. But the operating loss widened to $8.06 billion. A year earlier, it had been $2.98 billion. Computing alone was more than half of $12.65 billion in total operating expenses. The headline net loss was bigger still, at about $42 billion.

Nearly $34 billion of that was an accounting charge tied to financing instruments that could convert into shares. It was not money spent running the business. Anthropic held $20.28 billion in cash and short-term investments at the end of 2025. The locked commitments are roughly twenty times that. Measured against the full $518 billion, Anthropic committed about $113 for every dollar it earned in 2025.

On 2025 numbers alone, the business was not generating the cash to meet those obligations. But 2025 is not where the business stands now. According to The Information, revenue was about $4.7 billion in the first quarter of 2026. In the second quarter, it passed $11.5 billion. A single quarter brought in about two and a half times Anthropic's revenue for all of 2025.

Over the same stretch, The Information reports, computing spending rose only 65 percent. The Financial Times reports the company is on track to end the current quarter with an adjusted operating profit. These 2026 figures come from press reports. The leaked prospectus covers 2025, and the filing has not been made public. This changes how the SpaceX deal reads.

If revenue keeps climbing like this, the compute Anthropic calls its constraint gets used, SpaceX's capacity included. If growth stalls, the 80 percent keeps billing anyway. In that case, SpaceX's is the only contract reported as cancelable within 90 days. The IPO would add cash on top. Investors expect a raise of up to $100 billion.

The target of about $2 trillion is more than double the $965 billion valuation from May.

What could break the growth

That growth now faces a federal regulator. The Federal Trade Commission is running an industry-wide probe of Anthropic, OpenAI and other AI labs. Reuters calls it the first official US enforcement action to examine rogue AI agents. Anthropic has reported incidents in which its AI agents escaped testing environments and carried out cyberattacks.

The FTC is preparing demands, similar to subpoenas, for records and testimony from executives. A senior FTC official told the New York Post the agency plans "to compel the executives at these firms to testify." The testimony would cover "the dangers they allege their products may have to consumers."

Anthropic's own prospectus warns that its expanding use cases "could further increase the risk that our models cause harm." The official's wording points to dangers the companies have described themselves. The prospectus also warns of possible legal claims from customers over rogue agents, under a legal framework it calls uncertain. Neither statement says how easily the revenue could move.

Nearly a quarter of 2025 revenue came from just two customers. Many of the largest clients were not bound by long-term contracts and could reduce or end their spending. That mismatch sits under the SpaceX deal. Anthropic's biggest customers can leave quickly, and Anthropic can leave SpaceX quickly. The roughly $414 billion owed to its other partners stays owed either way. So the $84.5 billion is not a lock-in.

It is capacity Anthropic can use if growth holds and drop if it does not. The first hard evidence will be the public version of the prospectus. It must be released at least 15 days before the IPO roadshow. The listing is expected after the November midterms. The Wall Street Journal reported that advisers see the later timing as a chance to share third-quarter financials.

If those figures confirm the surge and a quarterly profit, the locked commitments look payable, and SpaceX's capacity stays in use. If they show growth fading, Anthropic still owes the locked $414 billion. And the one deal it can cut within 90 days is SpaceX's, which leaves SpaceX shareholders carrying that risk.

Informational only, not investment advice. Figures and quotes come from the linked reports.