A leaked SpaceX IPO filing reveals Anthropic's $45 billion compute contract, shedding light on the massive capital requirements and Nvidia infrastructure dependency driving the current AI arms race.
SpaceX's IPO filing, made public this week, has pulled back the curtain on one of the most consequential — and until now, largely opaque — financial arrangements in the AI industry: Anthropic will pay nearly $45 billion over three years for computing resources, working out to approximately $1.25 billion per month. The disclosure, first reported by Bloomberg and TechCrunch, marks the first time the true scale of AI infrastructure economics has been laid bare in a public regulatory document — and the numbers are staggering.
The same filing also revealed that Elon Musk's xAI burned through $6.4 billion in 2025 while aggressively expanding its Grok model infrastructure, signaling that the compute arms race is far from cooling down.
A $45 Billion Compute Contract: What It Means
The Anthropic deal is not a vague partnership announcement or a press-release figure. It is a hard contractual obligation disclosed in a securities filing — the kind of number that lawyers and auditors have signed off on. At $1.25 billion per month, Anthropic's compute spend alone exceeds the annual revenue of most enterprise software companies.
To put it in context: Anthropic's total funding raised to date across all rounds is estimated at roughly $12–14 billion. The company is committing to spend more than three times that amount on compute over the next three years. This is not a company buying servers — it is a company making a generational bet that frontier model capability is worth almost any price.
The arrangement also underscores a structural shift in how AI labs source infrastructure. Rather than relying solely on hyperscaler cloud agreements with AWS, Google Cloud, or Azure, Anthropic is locking in capacity through a compute provider tied to SpaceX's broader infrastructure ecosystem. The strategic logic: securing guaranteed access to scarce GPU and accelerator capacity before competitors can.
The Nvidia Infrastructure Dependency at the Core
The nvidia ai infrastructure investment impact running through this story is impossible to ignore. Whether the underlying hardware is sourced through Nvidia's H100s, B200s, or next-generation Blackwell Ultra chips, contracts of this scale are ultimately bets on Nvidia's supply chain and roadmap. Every dollar of that $45 billion flows, in some form, through data centers stacked with Nvidia silicon.
This creates a compounding dynamic: AI labs sign multi-year compute contracts → compute providers must pre-purchase or lease massive GPU clusters → Nvidia's order books fill years in advance → pricing power for Nvidia remains elevated. The SpaceX-Anthropic deal is, in effect, a downstream confirmation of why Nvidia's data center revenue has grown from roughly $15 billion in fiscal 2023 to over $115 billion in fiscal 2025.
"Anthropic will pay nearly $45 billion over three years for computing resources — approximately $1.25 billion per month." — Bloomberg, May 20, 2026
For infrastructure investors and enterprise buyers, this is a critical signal. The companies that control compute — whether through ownership, long-term contracts, or preferential allocation — hold structural leverage over the AI value chain that no amount of model fine-tuning can replicate.
xAI's $6.4 Billion Burn: Expansion, Not Inefficiency
The xAI figure deserves equal scrutiny. According to TechCrunch's reporting on the SpaceX filing, xAI burned $6.4 billion in 2025 — and the filing makes clear the spending is far from over, with Grok expansion plans requiring continued massive capital deployment.
xAI's burn rate, annualized, is running at a pace that rivals or exceeds OpenAI's reported infrastructure spend during its own rapid scaling phase. The difference is that xAI is building its own data center infrastructure — most visibly the Memphis "Colossus" cluster — rather than relying primarily on third-party cloud providers. That capital-intensive model means higher upfront burn but potentially lower marginal costs at scale.
The strategic contrast between Anthropic and xAI is instructive:
- Anthropic is paying a premium for guaranteed external compute capacity, preserving capital flexibility but creating a long-term contractual obligation.
- xAI is burning capital to own infrastructure, accepting short-term losses in exchange for long-term control.
Both approaches reflect the same underlying conviction: that compute scarcity is the binding constraint on AI progress, and that securing it — by any means — is the primary strategic imperative.
Why Public Disclosure Changes the Game
Prior to this filing, AI infrastructure economics were almost entirely opaque. Labs disclosed funding rounds but not compute spend. Hyperscalers reported aggregate cloud revenue but not per-customer breakdowns. The result was an industry where the true cost of building frontier AI was a matter of informed speculation.
The SpaceX IPO filing changes that. As a public offering document, it requires material contracts and financial obligations to be disclosed with specificity. The Anthropic compute deal qualifies. So does xAI's burn rate, which appears as a related-party disclosure given Musk's ownership stake in both SpaceX and xAI.
This transparency will have ripple effects. Investors evaluating AI companies will now have a benchmark for what frontier compute actually costs. Enterprise buyers negotiating cloud contracts will have a reference point. And competitors — including OpenAI, Google DeepMind, and Meta AI — will face pressure to either disclose similar arrangements or explain why their economics differ.
What to Watch Next
Several developments are worth tracking closely in the weeks ahead:
SpaceX IPO timeline: The filing itself is the trigger for all of this disclosure. If SpaceX moves toward a public listing, additional financial details about its compute business — and its relationships with AI labs — will become part of the public record.
Anthropic's revenue trajectory: Committing $1.25 billion per month to compute is only sustainable if Anthropic's API and enterprise revenue scales to match. The company's Series E and F valuations (reportedly in the $60–75 billion range) imply investors believe it will — but the math requires aggressive commercial execution.
Nvidia's next earnings: With multi-year, multi-billion-dollar contracts now confirmed in public filings, Nvidia's forward visibility on data center demand is stronger than any analyst model previously assumed. Watch for guidance language around long-term contracted revenue.
Competitive infrastructure moves: If Anthropic is spending $45 billion on compute over three years, what are OpenAI, Google, and Meta spending? The pressure to disclose — or to match — will intensify.
The SpaceX filing has done something rare in the AI industry: it has replaced speculation with specifics. The numbers confirm what many suspected — that the race for AI supremacy is, at its foundation, a race for compute, and the capital required is measured not in millions but in tens of billions.
Last reviewed: May 21, 2026
