I opened Anthropic’s S‑1 at 2 a.m. and felt the floor tilt beneath the numbers. A single clause turned routine cloud-computing deals into a ledger that could reshape who wins the next decade of AI. You should read the exact figures—because they redraw the map of influence in plain, frightening digits.
I’ll walk you through what jumped off the page and what it means for investors, partners, and anyone watching AI’s money flow. I’ve read prospectuses before; this one reads like a strategic battle plan and a balance sheet at once. I want you to keep two things in mind as we go: these are largely non‑cancelable commitments, and the scale is bigger than most headlines suggested.
On page after page, the SpaceXAI line stood out: the number was bigger than anyone had reported
The S‑1 lists a commitment to pay SpaceXAI up to $120.5 billion (≈€111 billion) through 2029 to access compute powered by NVIDIA GPUs. That’s roughly $36 billion (≈€33 billion) more than earlier Reuters estimates of $84.5 billion (≈€78 billion). Once all capacity is delivered—expected by the end of this year—Anthropic will pay SpaceXAI about $3.1 billion per month (≈€2.9 billion/month).
How much will Anthropic pay SpaceXAI?
The headline is the $120.5 billion cap (≈€111 billion). Under the agreements, the earliest non‑cancelable period runs into early 2027; after that, either side can cancel with 90 days’ notice. You should treat that monthly burn—≈€2.9 billion—as the real heartbeat investors will watch once deliveries finish.
This deal reads less like a supplier contract and more like a long-term lease with teeth—like a non-cancelable mortgage on future growth. It binds Anthropic to capacity and vendors at a scale that will shape product pricing and margins for years.
At a desk in my inbox, other lines made the spending pattern obvious
The S‑1 shows additional multibillion-dollar commitments: at least $114.5 billion (≈€105 billion) with Google through July 2033; at least $44.9 billion (≈€41 billion) with Microsoft through May 2033; and at least $110 billion (≈€101 billion) with Amazon through April 2036. The Google and Microsoft totals are larger than earlier reports by about $3.4 billion (≈€3.1 billion) and $13.5 billion (≈€12 billion), respectively.
Why are these contracts non‑cancelable?
Because providers like Google Cloud, Microsoft Azure, Amazon Web Services, and SpaceXAI are selling dedicated capacity and long lead‑time infrastructure. Anthropic and others commit ahead of demand to secure NVIDIA GPU capacity and custom data-center arrangements. Those commitments protect suppliers and guarantee access—but they also constrict flexibility if demand softens.
If you trace the lines, the industry is now a web of interdependent bets: Anthropic, OpenAI, hyperscalers, and hardware providers are all making forward wagers on demand and pricing. For Anthropic, those bets are underwritten by debt and long-term purchase obligations—the company says it has “incurred significant indebtedness” to fund compute commitments.
At 8:45 a.m. on a call with analysts, the revenue and loss figures prompted a different unease
Anthropic reported revenue of $16.29 billion in H1 2026 (≈€15 billion), up from $1.2 billion a year earlier (≈€1.1 billion). Yet the company recorded a net loss of $80.7 billion (≈€74 billion) for the period, compared with $42 billion last year (≈€39 billion). Most of that gap comes from valuation‑linked accounting adjustments; operating loss, by contrast, narrowed to $693 million (≈€637 million) from $2.1 billion (≈€1.9 billion).
Why is Anthropic losing so much money?
Two answers: accounting and commitment. Mark‑to‑market valuation effects balloon non‑cash losses. Meanwhile, the company’s operating model is investing heavily in compute capacity it has already contracted for. That creates headline losses even as product revenue climbs—an odd, awkward mix that will haunt valuation debates ahead of the IPO.
That dynamic leaves you and other investors facing a tricky question about timing: are you buying growth or buying a balance sheet weighted with multi‑year purchase obligations? The S‑1 makes plain that profitability is not guaranteed; Anthropic warns its losses may continue as it scales and operates publicly.
In a hallway conversation, I heard safety concerns linked directly to the IPO timetable
New York City comptroller Mark Levine publicly urged a delay to Anthropic’s offering, arguing that IPO pressures for growth and profits could conflict with safety work. On CNBC he warned that going public could distract teams and create incentives at odds with careful safety engineering. That point landed in the filing: Anthropic lists 93 pages of risk factors, including safety, model misbehavior, and the possibility of “excess compute that remains underutilized.”
The company acknowledges models can behave differently under evaluation and that accelerating R&D with AI could outpace safety interventions. It also flags longer-term harms from heavy AI use—mental health and cognitive risks—that could trigger liability and reputational damage.
Regulators, investors, and partners will watch for any sign that product timelines are being rushed to meet quarterly expectations. For Anthropic, the public market stage will introduce the same pressure cooker many tech firms face—like a pressure cooker with a slowly loosening valve—and safety advocates worry about what that does to guardrails and internal priorities.
At my desk I compared the numbers to reporting from Reuters and Fortune
Earlier coverage understated several totals: Reuters reported the SpaceXAI commitment at up to $84.5 billion (≈€78 billion), Google at $111.1 billion (≈€102 billion) and Microsoft at $31.4 billion (≈€29 billion). The S‑1 increases those Microsoft and Google figures by roughly $13.5 billion (≈€12 billion) and $3.4 billion (≈€3.1 billion), respectively. Fortune and others have been tracking the trillion‑plus dollars of industry infrastructure spending that this filing now feeds into.
Names matter: Elon Musk and SpaceXAI are now on the ledger alongside NVIDIA (GPUs), Google, Microsoft, Amazon, and the hyperscalers. Those brand ties shape deal leverage, pricing, and the political narrative around AI investment and control.
In the final pages I found a plain warning about demand uncertainty
Anthropic concedes that the market for advanced AI models may not grow as fast as investors expect. Privacy, ethics, cost pressures, or a preference for cheaper models could slow adoption. The company’s risk list is a legal olive branch and a candid admission: they’ve locked into capacity on bets that may not pay off at scale.
That creates a fragile ecosystem where a single large counterparty choice or a sudden shift in demand could ripple through providers and buyers. You should watch cash flow, debt maturities, and how the company manages its non-cancelable contracts once the blackout period ends in 2027.
If you follow policy and markets, note that Mark Levine and public officials are now part of the IPO story. Expect more scrutiny on safety governance and on how Anthropic balances growth targets with the public interest in responsible AI.
Read the S‑1 yourself (the SEC filing is public) and watch how the market prices these commitments once the IPO launches—reported timelines place the offering in November. I’ll be watching how analysts and underwriters debate whether revenue growth offsets the long-run supply contracts and non‑cash valuation swings. Which side do you think the market will take this time?
