Anthropic could pay SpaceX up to $84.5B for computing capacity through 2029

Anthropic is preparing to spend more on rented computing power than most countries collect in taxes. A confidential IPO prospectus dated around September 29, 2026 reveals the company could pay up to $84.5 billion to Elon Musk’s xAI for NVIDIA-based compute capacity through 2029, secured through SpaceX’s Colossus 1 data center.

That number is large enough to make most readers blink twice, so here is the context: $84.5 billion over roughly three years works out to an estimated $15 billion annually by May 2029, with monthly payments ramping up to approximately $1.25 billion once fully operational.

What Anthropic is actually buying

The partnership traces back to May 2026, when Anthropic secured exclusive access to Colossus 1, a SpaceX facility that houses more than 220,000 NVIDIA GPUs running on over 300 megawatts of power.

The arrangement is notable for one particular clause: the bulk of this commitment is cancelable on just 90 days’ notice. That is an unusually short leash for a contract measured in the tens of billions, and it stands in contrast to Anthropic’s other infrastructure agreements with partners including Google, Amazon, and Microsoft, which carry more binding terms.

The bigger picture: $518B in total commitments

Zoom out further, and the xAI deal is only one line in a much longer ledger. Anthropic’s prospectus discloses total AI infrastructure commitments exceeding $518 billion over the next decade, spanning its full roster of compute and cloud partners.

Roughly 80% of those commitments are classified as non-cancelable or usage-independent, meaning Anthropic owes the money whether or not it actually burns through the compute. The xAI deal, sitting in the cancelable 20%, is actually the flexible portion of an otherwise very rigid balance sheet.

The prospectus also reveals that Anthropic and SpaceX are in exploratory discussions around multi-gigawatt orbital AI compute capacity. Its inclusion in an IPO document suggests the company wants investors to see it as a serious long-term direction rather than a science project.

What this means heading into an IPO

An IPO prospectus is, at its core, a sales document. Every data point Anthropic chose to include is there because it tells a story the company wants potential investors to believe.

The story here is clear: Anthropic has locked in relationships with xAI, Google, Amazon, and Microsoft simultaneously, spreading its compute across multiple providers while maintaining flexibility through cancelable terms where possible.

The downside to that story is also visible in the same document. Five hundred and eighteen billion dollars in infrastructure commitments is a lot of future cash out the door, and 80% of it cannot easily be unwound.