The First Frontier S-1: Anthropic's Prospectus Shows $4.6B Revenue, a $42B Net Loss, and 47% of Sales Flowing Through Its Own Investors

For two years, the financial guts of the frontier AI labs have been a matter of leaks, analyst models, and stage-managed run-rate announcements. That ended this week. Reuters reviewed Anthropic’s confidential IPO prospectus, and the document — the first detailed look inside a frontier lab’s books — sketches a company growing faster than almost any in software history, losing money at a scale to match, and bound by contractual obligations larger than the GDP of most G20 members.

Anthropic confidentially filed its S-1 back on June 1, becoming the first major AI lab to formally begin the listing process. The prospectus, as reported by Reuters, Fortune, and The New York Times’ DealBook, is expected to precede a Nasdaq debut in November — an offering that could raise more than $100 billion at a valuation above $2 trillion, which would make it the largest IPO in history. Here is what the numbers actually say.

The headline financials

The shape of the business is now public. Revenue grew roughly twelvefold in 2025 to nearly $4.6 billion, up from around $380 million in 2024. Few software companies have ever scaled that fast at that base. But the cost side scaled even harder: Anthropic reported an operating loss of $8.06 billion and a net loss of $42 billion for 2025 — the net figure inflated by non-cash charges, including the accounting treatment of preferred-stock instruments that will flip on IPO. Compute and infrastructure spending hit $7.33 billion, roughly triple its 2024 level, and it is still climbing.

In other words: for every dollar of revenue Anthropic booked last year, it spent about $1.60 on compute alone. That is the unit economics of a company betting that revenue compounds faster than the infrastructure bill — and the prospectus commits it to that bet in writing.

The concentration problem

The detail that will occupy underwriters’ risk committees: 47% of Anthropic’s 2025 sales — about $2.16 billion — were routed through the cloud marketplaces of Amazon and Google. Enterprises buy Claude credits inside AWS and Google Cloud, Anthropic recognizes the revenue, and it paid roughly $351 million back to its cloud partners in marketplace and channel fees, per the filing. Two unnamed customers each accounted for about 12% of revenue on their own.

The structural issue is not the marketplace mechanic — channel sales through hyperscalers are standard enterprise software practice. It is who the channel is. Amazon and Google are simultaneously Anthropic’s distribution channel, its chip and data-center suppliers, and two of its largest investors. The same filing reveals a related structure on the supply side: Broadcom has agreed to lend Anthropic up to $42 billion in convertible notes to finance a $125.2 billion, five-year TPU lease — a supplier-as-financier arrangement we covered earlier this week. Every layer of Anthropic’s stack — capital, compute, and customers — runs through a handful of the same names. If those relationships soured, there is no version of this business that is unaffected.

The $518 billion wall

The single largest number in the document is $518 billion in planned spending on cloud, computing, and infrastructure obligations over the coming years — and, as Investing.com highlighted, much of it sits in contracts that cannot be canceled. This is the frontier-lab business model in one figure: growth is purchased years in advance, in cash commitments, before the revenue exists. Anthropic is not alone in this — OpenAI’s circular compute deals and Microsoft’s Azure commitments follow the same template — but Anthropic is the first to have to disclose it to public-market investors, who will now price that rigidity every single quarter.

The scale deserves emphasis. A company with $4.6 billion of annual revenue has signed onto more than a hundred times that amount in infrastructure commitments. The bull case in the filing is that annualized revenue keeps compounding toward the $100-billion-plus range investors project for the end of 2026; the bear case is written into the same risk factors.

The risk factor nobody has ever filed

Then there is the passage that made the prospectus famous before most people had read a single financial line. Among its risk factors, Anthropic warns that its own models could pose a “catastrophic or existential risk to humanity.” Coverage of the filing describes roughly 80 pages of AI-risk disclosure — misalignment, misuse, loss of control — sitting alongside the standard securities-law boilerplate.

There is genuine novelty here. A company whose founding charter is safety research has now told prospective shareholders, in a legally binding disclosure document, that its product line carries tail risks of civilizational scale. securities lawyers will be parsing the liability implications for years: if such risks are material enough to disclose, what does it mean to sell accelerator products anyway? And if they are not material, why are they in the filing? Either answer is uncomfortable. It is also, arguably, the most honest paragraph in any AI lab’s marketing — an honesty purchased, fittingly, only under the threat of shareholder lawsuits.

What November means

The listing window matters beyond Anthropic. OpenAI filed confidentially a week after Anthropic and is queued behind it; a pipeline of AI companies worth an estimated $3.6 trillion in total valuations waits on the other side. A successful November debut at $2 trillion-plus would validate the reference prices for all of them. A stumble — or even a cold reception to those risk factors — would chill the entire queue.

Public investors are about to own a piece of the frontier for the first time. They will get quarterly earnings instead of leaks, audited margins instead of analyst models, and risk disclosures that read like safety research. The prospectus’s message is consistent from its first page to its last: the growth is real, the losses are real, the obligations are locked in, and the downside scenarios include some that no discount rate can price. Anthropic has always asked the world to take both halves of that sentence seriously. In November, the market gets to place the trade.