Anthropic's IPO Filing Shows $500 Billion in Commitments

By Olivia Hart

Anthropic's draft IPO prospectus, filed Monday, discloses more than $500 billion in committed payments for hardware, cloud and data-center capacity over the coming years — against first-half operating cash flow of $3.9 billion.

The company behind the Claude chatbot released the first draft of its initial public offering document this week. The filing omits the two numbers most investors want: the stock price and the valuation. What it does contain, spread across hundreds of pages, is a detailed picture of how expensive it is to build an AI business at scale.

Two details stand out immediately. One is the sheer size of the spending commitments. The other is that the founders' headline philanthropic "pledge" is, by the company's own admission on page 118, not a pledge at all.

What has Anthropic actually committed to spend?

The obligations run to a small group of suppliers, and most of them are fixed regardless of how much capacity Anthropic actually uses.

• $114.5 billion to Google over seven years for AI and cloud infrastructure
• $110 billion to Amazon for the same purpose
• $44.9 billion to Microsoft for cloud infrastructure
• $161.2 billion to Broadcom for data-center and equipment leases
• $44.6 billion to NScale Global and $41.3 billion to Lambda, two private AI infrastructure companies

That list alone totals $516.5 billion. On top of it sit two open-ended commitments: an unspecified amount expected to exceed $20 billion with chipmaker Advanced Micro Devices, and another unspecified amount of "up to $120.5 billion" with SpaceX.

Every payment commitment in the cloud contracts is "regardless of actual usage," according to the prospectus. In plain terms, Anthropic pays whether or not it needs the capacity — which makes these obligations behave like debt, even though they never appear on a balance sheet as borrowings.

How big is that relative to Anthropic's cash generation?

The prospectus shows operating cash flow of $3.9 billion for the first half of this year, on page 168. Annualized, that is roughly $7.8 billion — against more than half a trillion dollars in committed outlays stretching over years.

Brett Arends, writing his ROI column on the filing, put the gap in perspective: "A hundred billion dollars here and a hundred billion dollars there, and pretty soon you're starting to talk about serious money."

There is nothing improper about these commitments; they are the cost of securing compute in a capacity-constrained market. They work out fine so long as Anthropic's revenue booms. But any valuation of the company that ignores these fixed obligations is valuing something other than the actual business.

Is the founders' 80% giving pledge legally real?

No — and the prospectus says so itself. The document states repeatedly near the top that "our co-founders have each pledged to dedicate 80% of their personal equity in Anthropic to philanthropic causes." Page 118 then concedes that these commitments are "voluntary" and "revocable," express nothing more than intent, and "do not constitute legally binding obligations."

A statement of intent that can be withdrawn at will is not a pledge. Arends is blunt about the framing: "it's not a pledge if it's not a pledge." His column notes that cynical readers may wonder whether the prominent placement of the giving commitment near the front of the document, repeated several times, amounts to pre-IPO spin — with the get-out clause buried deep in the back, in the hope that few readers make it to page 118.

To be fair, it is the founders' money. They are under no obligation to give any of it away, and they deserve credit if they follow through. The issue is disclosure design, not generosity: a legally binding promise and a revocable intention should not carry identical language in an offering document.

What should investors take from the filing?

The filing leaves the market's central questions unanswered. No share price, no valuation range, no revenue history beyond the cash-flow figure. What it does establish is the structure of the risk: fixed, usage-independent obligations to a concentrated set of counterparties — Amazon, Google, Broadcom, SpaceX, AMD, Microsoft, NScale and Lambda — that dwarf the company's current cash generation.

Investors with long memories, as Arends points out, will recall that similar commitments and financing arrangements turned out badly when the technology bubble burst at the end of the 1990s. Whether Anthropic's bets compound into boom-era Microsoft or bubble-era Global Crossing is precisely what the final prospectus, and the market's appetite for it, will have to price.