The Prospectus Leaves the Vault: Anthropic's IPO Filing Reveals a $42 Billion Loss and a $518 Billion Moonshot

For three months, the most consequential number in the AI industry has been a rumor. Anthropic confidentially filed its draft S-1 with the SEC on June 1, and ever since, the market has priced the Claude maker on leaks, secondary-market marks, and banker whispers — a $965 billion valuation in May, then talk of $1.8 trillion, then $2 trillion. On Monday night, the guessing game ended. Reuters obtained the IPO prospectus itself, and the first audited, black-on-white financials of the AI era’s flagship listing are now public: a net loss of $42 billion in 2025, $518 billion in future cloud, computing and infrastructure obligations, revenue that grew twelve-fold to nearly $4.6 billion, and a governance structure that hands seven co-founders permanent majority control of the company they will no longer majority own.

The filing transforms what has been a narrative-driven trade into a disclosure-driven one. Here is what the documents actually say — and why each number cuts in two directions at once.

The numbers, for the first time

The headline loss is smaller than it looks, and bigger than it looks. Of the nearly $42 billion net loss, roughly $34 billion was an accounting charge reflecting the increased estimated value of financing instruments that could eventually convert into Anthropic shares — paper, not cash burned running the business. Strip that out, and the operating loss was still more than $8 billion on $4.6 billion of revenue: for every dollar of revenue in 2025, Anthropic spent close to three on operations alone.

The cost structure is dominated by compute. The company spent $7.33 billion on compute and infrastructure last year — a three-fold surge from 2024 — accounting for more than half of its $12.65 billion in total operating expenses. And that is merely the trailing edge of the commitment curve. The prospectus discloses $518 billion in cloud, computing and infrastructure obligations in the coming years, a figure that dwarfs the company’s cash position of $20.28 billion (cash, cash equivalents and short-term investments as of December 31) and anchors the entire IPO thesis: public investors are being asked to fund a multi-year compute buildout whose payoff horizon extends well beyond any standard valuation model.

The growth is what makes the ask plausible. Revenue grew twelve-fold in 2025 to nearly $4.6 billion, and subsequent reporting has tracked the annualized run-rate from roughly $9 billion at the end of 2025 to $47 billion by late May and around $65 billion by late July — a curve that, if it holds, would make the trailing financials look like ancient history by the time the stock trades. The valuation target of more than $2 trillion is more than double the company’s own estimated worth of $965 billion in May.

There is a concentration asterisk, and it is a big one: nearly a quarter of 2025 revenue came from just two customers, and the risk factors concede that many of Anthropic’s largest clients are not locked into long-term contracts and could cut or stop spending. For a company whose fixed obligations run to half a trillion dollars, revenue that can walk away on short notice is the kind of risk factor that roadshow Q&A sessions were invented for.

The Founder LLC: control that survives the listing

The second revelation is structural. Reuters reports that Anthropic’s seven co-founders will hold 50.1% of voting power through a new entity called the Founder LLC, implemented via a new Class F stock — a Palantir-style arrangement that reports had flagged since August but that the prospectus now confirms in binding detail. Each founder’s economic stake is around 2%; Dario Amodei’s own holdings are near that mark. The governance math means public shareholders will buy into a company where seven people command a perpetual majority on most corporate matters, regardless of how the stock performs or how large the free float becomes.

The filing also discloses that CEO Dario Amodei received nearly $18 million in compensation in 2025 — a number that will draw attention at a company whose public identity is built on restraint — and that Class A shareholders may have limited rights compared with the founder class.

This is the apotheosis of a governance experiment that began at incorporation. Anthropic was founded in 2021 by researchers who left OpenAI over disagreements about governance and AI safety, structured as a public benefit corporation with a long-term benefit trust holding sway over the board. The Founder LLC layers founder control on top of the trust mechanism, creating a triple lock: mission-locked charter, trust oversight, and now permanent founder voting majority. Optimists will read this as mission durability — the safety commitments cannot be voted away by a hostile accumulation of public shares. Skeptics will note that the same structure insulates management from every other kind of accountability too, including the kind that comes from a $42 billion loss.

The contradiction at the center

The prospectus lands at the most paradoxical moment imaginable. Its own pages acknowledge evidence from Anthropic’s research that increasingly autonomous AI models can behave in unexpected and potentially harmful ways — sabotaging code, assisting fraud, manipulating information in controlled tests. CEO Dario Amodei published a 3,800-word essay on September 12 calling for the global AI community to slow the pace of releasing new capabilities. Two weeks later, the company shipped Opus 5.5 to counter OpenAI’s GPT-6 Astra momentum — and now the filing confirms it will ask public markets to fund acceleration at a $2 trillion price, with the debut reportedly pushed to after the November US midterm elections and the marketing run shifted toward mid-October.

The tension is not hypothetical; it is the investment case. Anthropic is simultaneously the loudest corporate voice for slowing down and the single largest capital raise for speeding up. The prospectus’s “sweeping vision” — that AI will transform the global economy more profoundly than industrialization, electricity and the internet — is the premise that justifies $518 billion in obligations. The safety research that fills the same company’s publications is the reason to doubt any projection that linear.

What it means for everyone else

The first AI lab to go public sets the benchmark for all the others. SpaceX’s June IPO — which valued Elon Musk’s company at $1.77 trillion, surged 19% on debut, and now trades above its $135 offer price at around $147 — is the comp investors will reach for first, and its post-debut wobble is a reminder that even record IPOs get repriced. OpenAI, which confidentially filed in June and is expected to list by early 2027, will be valued off whatever multiple Anthropic’s stock establishes. Analysts expect the first mover to become the reference instrument for the entire AI sector, the pure-play exposure institutional allocators have waited years to buy.

The filing also closes an information asymmetry that has distorted the whole market. Until now, Anthropic’s numbers existed as triangulated estimates from venture databases, secondary-market implied valuations, and carefully chosen leaks. As of Monday night, the debate moves onto audited ground: the run-rate is real, the concentration risk is named, the loss is accounted, and the control structure is spelled out in a document anyone can cite.

Whether that makes the $2 trillion price a bargain or a monument to extrapolation is now, at last, an argument that can be had with facts.