Anthropic’s IPO filing commits the company to at least $518 billion in cloud and compute contracts over the next decade, mostly payable whether the compute gets used or not, as the AI lab angles for a valuation above $2 trillion. The leaked prospectus, first reported by Fortune, also shows a $42 billion net loss for 2025 on $4.6 billion of revenue.
The filing lays out how concentrated that spending is. Google accounts for at least $111.1 billion of the commitments, Amazon $110 billion, Microsoft $31.4 billion, and roughly $161.2 billion sits in equipment leases tied to Broadcom. About 80 percent of the total is non-cancelable or owed regardless of usage. Spread evenly over ten years that averages $52 billion a year, though the contracts do not require even payments. The remaining slice spreads across smaller infrastructure partners and capacity reservations.
Revenue against the bill
The financials read like a startup scaling far ahead of its income. Revenue grew twelvefold to nearly $4.6 billion in 2025, against roughly $8 billion in operating losses and $7.3 billion spent on compute and infrastructure, up threefold from 2024. Once accounting charges land, the Globe and Mail put the net loss at $42 billion, a figure that drew pushback from bulls who argue non-cash items distort the picture.
Two customers supplied nearly a quarter of 2025 revenue, and many of the largest contracts are not locked into long-term commitments, a mix the filing itself flags as a risk. Concentration that tight means a single enterprise cutting spend would dent the model the valuation rests on, and the document says so plainly.
A very large number
Anthropic is targeting a valuation above $2 trillion, more than double the $965 billion its May 2026 funding round implied. Nasdaq Private Market put the private valuation at $1.32 trillion as of late September. Outlets including TheStreet suggest an offering could raise more than $100 billion, with a debut expected after the November midterm elections. For comparison, SpaceX went public in June at a record $1.77 trillion, the largest IPO in history. Anthropic would top it.
The company filed confidentially in June and must make the prospectus public at least 15 days before its investor roadshow, so the leak moved disclosure forward by weeks without changing the substance.
Risk disclosure with teeth
What drew as much attention as the money was the risk section. Roughly 80 of the 261 pages deal with AI risks, and the filing warns plainly that Anthropic’s own models could pose catastrophic or existential risk to humanity, as Fortune reported. That language reads oddly beside a pitch asking investors to value the company above $2 trillion, and analysts have started quoting the passages side by side.
The disclosure is consistent with the company’s public posture. Anthropic publishes a policy on what it calls the AI exponential and regularly argues frontier development carries systemic risk. Investors now get the same warning inside the securities document they would rely on to price the shares, which is unusual in either direction. A disclosure that scares investors is rare; one that does so while claiming a two trillion dollar premium is rarer still.
The compute bet
The commitments reflect a bet that access to chips and data centers, not model design, is the binding constraint on building better systems. Anthropic has already locked one unusual deal along those lines: TheStreet reports a SpaceX arrangement paying $1.25 billion a month for compute capacity running through May 2029, part of a push to extend training infrastructure beyond conventional data centers.
Locking in capacity makes sense while scarcity holds. It turns cloud costs into a fixed bill, which is tolerable at twelvefold revenue growth and painful if growth flattens. Investors were already questioning pay-as-you-go assumptions across the sector; a filing where four fifths of a half-trillion-dollar commitment survives a demand shock sharpens that debate. OpenAI, which spends on a similar scale through its Microsoft and Oracle relationships, is weighing its own public offering, so these numbers will be read as a benchmark for the whole category.
The filing also describes reliance on large tech firms that are simultaneously suppliers, investors, customers, distributors and potential competitors. Google’s stake and cloud deal sit alongside Gemini, its own frontier model line, and Microsoft both sells compute to Anthropic and competes through OpenAI. That entanglement runs across the industry, but the sums here make it the clearest case study available. Nobody has previously put this much contracted future spending on paper in a public securities filing.
What the market will argue about
Reaction split along familiar lines. Bullish coverage treats the $518 billion as proof management sees demand far ahead of supply, pointing to capital expenditure at the large cloud providers growing every quarter since 2023 with no peak in sight. Skeptics call it the clearest evidence yet of an infrastructure bubble, noting 2025 revenue covers under one percent of the decade’s committed spend and that depreciation schedules on this hardware run five to six years, faster than the contract terms they finance.
Both readings share the same document, which usually means the argument runs through the IPO window and past it. Between now and the expected roadshow, the numbers to watch are quarterly revenue against the monthly compute bill, whether enterprise contracts convert into multi-year commitments, and how the SEC treats a risk section that warns investors the product itself might be dangerous. None of that gets settled by the filing alone, which is the point of a roadshow. The debut timing after the midterms leaves roughly two months for those questions to move from commentary into pricing.