Anthropic's Confidential IPO Filing: $42 Billion Lost, $518 Billion Owed — and Investors Won't Be in Charge

The most consequential document in the AI economy is one the public cannot read. On Monday evening, Reuters reported that it had reviewed Anthropic's confidential IPO prospectus, and the figures are staggering: a $42 billion net loss in 2025, $518 billion in future cloud, computing and infrastructure obligations, and a public sale that could value the five-year-old lab at more than $2 trillion. Anthropic declined to comment, and the filing itself is not public — every number below comes from reporting on the document, not from the document itself.

The numbers: $4.6 billion in, $42 billion out

Start with the top line. Anthropic's revenue grew roughly twelvefold in 2025 to nearly $4.6 billion, up from about $400 million a year earlier, according to the prospectus reviewed by Reuters. That is explosive commercial growth by any standard.

Now the other side of the ledger. The company reported a net loss of about $42 billion for 2025. Most of that figure — roughly $34 billion — is a non-cash accounting charge reflecting an increase in the estimated value of financing that could eventually convert into Anthropic shares, rather than money the company spent running its business. Strip that away and the operating picture is still severe: after setting aside the fundraising-related write-downs, losses exceeded $8 billion.

The cost engine is compute. Last year's compute and infrastructure bill came to $7.33 billion — triple the 2024 level and more than half of $12.65 billion in total operating expenses. It ended the year with $20.28 billion in cash, cash equivalents and short-term investments.

The $518 billion obligation

The number that reframes the entire AI trade is not the loss — it is the commitment. The prospectus lays out $518 billion in cloud, computing and infrastructure obligations coming due in future years, according to Reuters' review. Dow Jones Newswires, which also reported on the filing, put it the same way: the spending plan stands despite the losses.

Read it plainly: a company with $4.6 billion in annual revenue has obligated itself to more than a hundred times that revenue in future infrastructure spending. The filing's own risk disclosures sharpen the point. Nearly a quarter of Anthropic's 2025 revenue came from just two customers, and the company warned that many of its largest clients are not locked into long-term contracts and could cut or stop spending, Reuters reported. Contractual obligations on one side; concentrated, uncontracted revenue on the other.

A prospectus that warns about its own product

The filing does not hide what the technology can do. Reuters reported that the prospectus confronts evidence from Anthropic's own research that increasingly autonomous AI models can behave in unexpected and harmful ways — including sabotaging code, assisting fraud and manipulating information in controlled tests. The Financial Times, which also reviewed the filing, found risk factors running to roughly a third of the prospectus, according to TechCrunch's account of its reporting.

That tension runs through the company's public posture. Chief executive Dario Amodei has called for the global AI community to slow the pace of releasing new capabilities. Yet Anthropic rolled out its new Opus 5.5 model last week, countering the momentum of OpenAI's GPT-6 Astra launch ahead of its own expected IPO. The filing, per Reuters' review, says the company has already restricted or delayed some capabilities for safety reasons — including choosing not to build commercially attractive image and video generation models, directing that compute toward research and safety instead.

The governance: public money, private control

If the numbers are the headline, the governance is the fine print that matters most to future shareholders. A second Reuters piece, published Monday, details a novel “Founder LLC” structure: a vehicle made up initially of Anthropic's seven co-founders — people the filing calls “distinctly equipped to be stewards of our mission” — that will direct a single Class F share holding a 50.1% vote on key corporate matters, including the election of some directors.

The arithmetic for public investors is blunt. Class A common stock will carry one vote per share, but the filing's own language warns the capital structure could diminish everyday investors' influence. Strategic partners get minimal voting rights. The company will remain a Delaware public benefit corporation, formally letting leaders balance investor interests against humanity's — and the filing warns that this setup could produce decisions “that may conflict with” shareholders' short-, medium- or long-term financial interests and business performance.

The remaining check on the founders is unusual too. Four of the board's directors will be elected not by shareholders but by a Long-Term Benefit Trust whose current trustees include former Federal Reserve Chair Ben Bernanke and national-security expert Richard Fontaine. The founders' supervoting power begins to sunset only when two or fewer co-founders remain. For the record: Amodei made nearly $18 million in 2025, his sister and company president Daniela Amodei made $16.4 million, and all seven co-founders pledged to dedicate 80% of their personal Anthropic equity to charitable causes.

Why the market gets a vote anyway

Whatever the founders control, the market sets the price. A public sale valuing Anthropic above $2 trillion would be worth more than double the $965 billion estimated mark from May — and it would arrive as the AI trade faces its stiffest test in years. SpaceX's recent blockbuster listing valued that company at $1.77 trillion; its shares jumped 19% on debut day in June to $160, then faded to around $147. AI and chip stocks have sold off recently, and Reuters noted the Anthropic sale will test whether enthusiasm for the trade survives greater scrutiny.

The calendar is part of the story. The listing is now expected after November's midterm elections, pushed back from earlier September-or-October expectations, according to the reporting. OpenAI — founded by the same milieu Anthropic's founders left in 2020 over governance and safety disagreements — confidentially filed for its own IPO in June and is expected to list by early 2027. Whoever goes first sets the reference price for every AI lab that follows, a dynamic analysts have flagged for years.

The backdrop is friction, not harmony. Amazon and Google are simultaneously Anthropic's strategic partners, its investors, and its cloud suppliers. Amodei has clashed with the White House over how Anthropic's tools may be used — a dispute that led to a temporary Pentagon blacklisting, later blocked by a federal judge.

What to watch next

Three things. First, the document itself: as of Monday night, no public registration statement had appeared in the regulator's database, so every figure in this story traces to journalists' review of a confidential draft — a caveat that stays attached until a public filing lands. Second, the price: whether investors will pay more than $2 trillion for a company that lost $42 billion last year and warns, in its own filing, that its leadership may act against their financial interests. Third, the funding path for $518 billion in obligations — in a market suddenly asking harder questions about the AI trade.