Anthropic's S-1 Deep Dive: Huge Growth, a Giant Paper Loss, and a Trillion Dollar IPO

On Monday, October 26, Anthropic did something it has managed to avoid for five years: it showed everyone its numbers. The company behind the Claude AI models filed its S-1 with the SEC, the long prospectus every company has to publish before it can sell shares to the public. A confidential draft went to regulators back on June 1, but this is the first version anyone can read, and it runs to roughly 240,000 words.

These filings usually follow a script. The growth story goes up front, the risks get buried in legal language, and the reader nods along. This one is different, and not only because of the size of the numbers. The risk section alone runs to more than 90 pages and warns, in plain words, that AI could pose catastrophic or existential risks to humanity. That is the company selling the shares saying that about its own product.

What happened

Anthropic, formally Anthropic PBC, has applied to list its Class A common stock on the Nasdaq Global Select Market under the ticker ANT. Goldman Sachs (GS) sits in the lead position among 24 underwriters, with Morgan Stanley (MS) named second. Both the company and some existing shareholders will sell shares, and Anthropic receives nothing from the shares its existing holders sell.

Plenty is still blank. The filing gives no number of shares, no price range and no listing date, and the registration statement has not yet become effective, which means no shares can be sold until the SEC says so. Reporting from Reuters and the Wall Street Journal points to a roadshow starting around November 9 and trading in the week of November 16, with Nov. 19 or 20 mentioned by Fox Business. Treat those dates as reporting, not promises, because timing on a deal this size can slip.

Why it matters

Until now, the only way most people could own a piece of a frontier AI lab was indirectly, through Microsoft (MSFT), Alphabet (GOOGL) or Amazon (AMZN). Anthropic is the first pure AI lab of this scale to come to market, so its IPO is the first real test of what public investors will pay for this kind of business. The answer will set the reference price for everyone behind it in the queue, OpenAI included.

The scale is hard to overstate. If the deal prices anywhere near the range bankers and investors are discussing, between $1.5 trillion and $2 trillion according to Reuters, it would rank among the largest IPOs in history and could raise around $100 billion. For context, Anthropic's last private round in May 2026 raised $65 billion at a $965 billion valuation. None of the IPO figures are in the filing itself, because the price section is still blank, but the direction of travel is clear.

The numbers that matter

The headline pair looks like a misprint: $16.3 billion of revenue in six months, and a net loss of $80.7 billion over the same period. Both are real, and the second one needs explaining.

Most of that loss never left the building. About $72.4 billion of it is a non-cash accounting charge on convertible notes and similar instruments. A convertible note is a loan that can turn into shares later. When the value of a company rises fast, accounting rules make the company record the rising paper value of those instruments as a loss. Anthropic's valuation rose very fast, so the paper loss is enormous. Underneath it, the operating loss for the half was $693 million, and the second quarter on its own produced an operating profit of $974 million, the first in the company's history. Operating cash flow for the half was positive at $3.93 billion.

A few more figures deserve attention. Gross margin, what is left of revenue after the direct cost of serving customers, came in around 85% for the first half, because cost of revenue was $2.38 billion. That number flatters the business a little: most of the computing bill sits in research and development instead, since the same machines that serve customers also train the next models. Total spending on compute and infrastructure was $8.95 billion in the half. About 80% of revenue is usage based, meaning customers pay per token, the small units of text the models read and write, and subscriptions make up the rest. The company says it has more than 11,900 customers spending at least $100,000 a year, and around 250 spending at least $10 million.

Anthropic has also told investors that third quarter revenue reached $16.8 billion with adjusted profitability. That quarter is not part of the audited figures in the S-1, which run through June 30, so file it under company communication rather than filed financials.

Who owns it, and who controls it

Amazon and Google are not just Anthropic's landlords for computing power, they are two of its biggest shareholders. Bloomberg's reading of the filing puts Amazon's economic stake at roughly 9% before its remaining notes convert, with $6.2 billion of convertible notes still outstanding at June 30. Those notes convert into non-voting shares at the IPO, and Amazon's total holding is capped at 12% unless Amazon agrees to lift the cap. Google's economic stake is at least 6.6% on the same basis, capped at 14.99%, and Google holds an option to invest up to another $10 billion in shares at the IPO price alongside the offering. Each of the seven co-founders holds roughly 240 million shares.

Control is a separate question from ownership, and here the answer is blunt. A single special share held by an entity owned by the seven co-founders will carry 50.1% of the voting power on most shareholder matters. A separate body, the Long-Term Benefit Trust, five trustees with no financial stake in the company, will elect a majority of the board. Anthropic is a public benefit corporation, a legal form that asks directors to balance shareholder returns against the company's stated purpose, the responsible development of advanced AI for the long-term benefit of humanity. People who invest in the IPO get Class A shares with one vote each, and the filing warns in so many words that this structure may effectively eliminate their ability to influence important decisions, including a sale of the company.

The bull case

The optimistic reading starts with growth that has almost no precedent at this scale. Revenue went from $41 million in 2023 to $16.3 billion in a single half of 2026, and the second quarter alone was bigger than everything before it combined. Customers are not just trying the product, they are expanding: the company reports net revenue retention around 900%, which means existing customers spend many times more than they did a year earlier. The first operating profit has already happened, a year or two ahead of when sceptics expected it, and the balance sheet carries about $131.7 billion in cash and short-term investments, so this is not a company listing because it is running out of money. If AI spending by businesses keeps compounding at anything like this pace, a very large valuation can be argued for with a straight face.

The bear case

The cautious reading starts with concentration. About 20% of first half revenue came from just two customers, and almost half of revenue arrives through the cloud marketplaces of Amazon, Google and Microsoft. Those three are simultaneously Anthropic's biggest suppliers of computing power, major shareholders, distribution channels and direct competitors. The filing says moving its workloads elsewhere would be costly and slow.

Then there are the commitments. Anthropic has signed compute and infrastructure deals that reporting puts at more than $500 billion over the coming years, and the filed figures include about $265.6 billion of non-cancelable purchase commitments at June 30, up from $54.6 billion six months earlier. Revenue that is 80% usage based can fall quickly if customers cut back, while those bills cannot. The company itself says it expects its growth rate to decline over time, possibly sharply, and warns that margins may compress when large blocks of new computing capacity switch on. Add the unusual items: the company estimates that 5 to 7 cents of every revenue dollar comes from users who break its rules and whom it intends to remove, and the risk section describes test incidents in which its own models gained unauthorised access to third party systems. Finally, public shareholders pay full price for a company they will have almost no say in running.

What to expect from the IPO

The mechanics from here are standard, even if the size is not. Expect an updated filing with the share count and a price range, then a roadshow, the period where management pitches large investors, reported to start around November 9. Pricing follows, then trading, with the week of November 16 the window most reports point to. The gap between the $1.5 trillion some investors are leaning toward and the $2 trillion in early headlines is the negotiation to watch: it decides whether this is remembered as the deal that cleared easily or the deal that tested the market's limit.

The lock-up, the period during which insiders cannot sell, is staggered rather than a single 180 day cliff. Some employees and other non-founder insiders can sell up to 10% of their holdings from the first day of trading, with further releases tied to the publication of full year 2026 and first quarter 2027 results, or 180 days, whichever comes first. Founders and the largest holders wait longer, up to 365 days. That spreads the extra supply of shares across months instead of dropping it on the market in one day.

What happens after the debut matters more than the debut itself. The first one or two earnings reports as a public company will show whether growth anywhere near this pace survives contact with public markets, and whether the computing commitments start to look like an advantage or a weight. Either way, the books are open now. For a company built on the promise of transparency about AI, that feels appropriate.

This article is for information purposes only and is not financial advice.

Cover: Anthropic logo via Wikimedia Commons. Portrait of Dario Amodei by TechCrunch (CC BY 2.0), via Wikimedia Commons.