Anthropic IPO explained: what its S-1 really says
Anthropic IPO explained from its S-1: a $104B net loss, $33B revenue, a 50.1% founder vote, ANT lock-up dates and what early buyers face.
Anthropic is going public, and the Anthropic IPO is the one everyone is talking about this week.
980 million shares, priced at $83 each, on Nasdaq under the ticker $ANT. Anthropic set the price on November 18. Do the math on that valuation and it lands above a trillion dollars. Yes, a trillion.
And in the document they filed to go public, they say they lost $104.0 billion in the first nine months of the year. A hundred and four billion. I'm writing it out in full so you know it's not a typo.
Same page, they also say the business already generates cash.
Both things are true. The explanation is almost silly, and I'll get to it below. But first you need to understand what this document is, because it's the only thing we really know about this company.
Figures from the S-1/A-3 filed November 17, 2026 and Anthropic's pricing announcement of November 18, 2026. When a calculation is mine, I say so. I'm not a financial advisor and this is not a recommendation.
The 30-second version
- They booked $33.2 billion in revenue in the first nine months of 2026, more than thirteen times the same period in 2025.
- Most of the huge loss is accounting, not cash. Operating, they lost $1.26 billion, and operating cash flow was positive.
- They've signed compute commitments of roughly $900.9 billion.
- You'd be buying shares with one vote each. The co-founders hold one share worth 50.1% of the votes.
What an S-1 is, explained simply
When a company wants to sell shares to the public for the first time (an IPO), in the US it has to register the offering with the SEC, the regulator. The form is called an S-1.
Picture a lemonade stand. You want to sell 20% of it to strangers. Before you take their money, the law makes you tell them how much you sell, how much you spend, what could go wrong, and what you'll do with the cash. That's an S-1.
What almost nobody knows: the SEC doesn't check whether the company is good or whether the price makes sense. It checks that the information is complete. An S-1 existing doesn't mean anyone has endorsed the investment.
It's not one document either. First come confidential drafts, then the public S-1 (Anthropic's is dated October 26), then the amendments, called S-1/A (there are three here: November 2, 10 and 17), plus a registration increase filed on the 18th. The last one is the 424B4, the final prospectus with the price set. Anthropic announced the price on the 18th, but when I wrote this the 424B4 wasn't on EDGAR, the SEC's database, yet.
How to read an S-1 in 30 minutes
It can run hundreds of pages. You don't need all of them. This is the order I go in:
- Cover page. What they're selling, how much, at what price, and where.
- "The Offering." How many shares the company sells versus how many existing owners sell. Money from the existing owners doesn't go into the company's bank account.
- Financial summary. Revenue, losses, cash. Always check which period each number covers.
- Risk factors. Headlines are enough. Look for the ones that could only happen to this company.
- Use of proceeds. If it says "general corporate purposes," that's information too.
- MD&A. Management explaining its own numbers, in its own way. The most useful section, and the one you most need to cross-check.
- Commitments. What they've agreed to spend. It's in the notes, not on the cover.
- Share classes. Who votes and who picks the board.
- Lock-up. When existing holders are allowed to sell.
And the red flags that should make you squint: numbers from different periods compared side by side, losses that come from changes in valuation rather than money leaving, metrics the company invented, and existing owners selling in the same offering.
Let's get into Anthropic.
How Anthropic makes money
Mostly usage. Customers pay based on what they consume from the models, measured in tokens. They also have subscriptions.
The number you'll see everywhere this week is the annualized revenue: about $81 billion as of October 31. Here's how it's built: take what was billed for usage over the last 28 days, multiply by 13, then add monthly subscriptions times 12. In other words, it's a projection from one month. It's like saying you make $50,000 a year because one month went great. The filing itself says it may not repeat.
The other one is the net revenue retention, at 900%. It sounds like a typo and it isn't one: the usage-based customers who were already there a year ago now spend, in total, nine times what they spent then. In total, not each one.
And one that's easy to picture: customers paying them more than $100,000 a year went from roughly 950 to over 11,900 in a year.
Why the net loss is $104 billion but cash flow is positive
The table, in millions of dollars:
| 2024 | 2025 | 9M 2025 | 9M 2026 | |
|---|---|---|---|---|
| Revenue | 386 | 4,585 | 2,545 | 33,242 |
| Operating loss | (2,980) | (8,060) | (5,224) | (1,260) |
| Change in fair value of instruments | (5,434) | (34,308) | (20,107) | (94,324) |
| Net loss | (8,309) | (41,965) | (25,107) | (104,039) |
Source: S-1/A-3, "Summary Consolidated Financial and Other Data." 9M means the first nine months of the year. The "change in fair value of instruments" line covers convertible notes, derivatives and SAFEs.
The row that matters is the second one. Operating loss went from $5.2 billion to $1.3 billion over the same period, while revenue grew more than thirteenfold. In nine months they've already billed more than all of 2025.
The $104 billion net loss comes almost entirely from the third row. Before the IPO, Anthropic raised money through convertible notes and SAFEs, which turn into shares later. The more the company is worth, the more those instruments are worth, and accounting books that rise as a loss. The filing attributes it to exactly that: the company's valuation going up.
Think of it as owing a friend $100 payable in your company's shares. If your company gains value, what you owe goes up in dollar terms. On paper you've lost money. Your bank account hasn't moved.
That's why you have to look at cash. Operating cash flow was $6.6 billion positive in the first nine months of 2026, versus $2.0 billion negative a year earlier.
If you see "$16.3 billion in revenue for the first half" or "$19.7 billion over twelve months" floating around, neither figure is in the S-1. The $33.2 billion is in there, but it covers nine months, not a year.
Compute commitments and the strategic investors
Two numbers that tell you a lot together.
The first: Anthropic has committed to buy and lease compute capacity (chips, data centers) for roughly $900.9 billion. I counted the zeros three times. About $451.6 billion would be spent before 2031, and about $70.7 billion comes due in the next twelve months. If you see $895 billion or $542 billion somewhere, the filing says $900.9 billion, so that's the one to use.
The second: right after the offering, seven companies buy shares directly from Anthropic, at the offering price, without going through the market. Google about $10 billion, NVIDIA about $10 billion, AMD $2.5 billion, Broadcom, Micron and Microsoft $1 billion each, and TSMC $500 million. Total: $26 billion.
Who are they? Mostly its chip and cloud suppliers.
It's a bit like your landlord buying a piece of your company, and you then paying him rent with that money. It's not illegal or strange, it's a commercial relationship. But keep it in mind when you read that "tech giants are betting on Anthropic."
Who controls Anthropic: founder votes and the trust
- Class A: what the public buys. One vote per share.
- Class N: no vote. This is the class for those seven companies and for a public benefit fund owned by Anthropic.
- Class F: one share held by a company owned by the co-founders. It carries 50.1% of all votes and no right to profits.
- Class T: one share held by an independent trust, the Long-Term Benefit Trust, which elects the majority of the board.
You get your vote. They have 50.1%. Do the math.
It's also a public benefit corporation: the board has to balance shareholder interests against the interests of people affected by what the company does, plus a stated goal, the responsible development of advanced AI for the long-term good. Sounds nice. For someone putting money in, it also means the board isn't required to think only about your money.
Lock-up expiration: when insiders can sell
People who already hold shares can't sell right after the IPO. That's the lock-up. Here it ends at whichever comes first: two trading days after the first-quarter 2027 earnings release, or day 180 after the offering, which would land around May 2027 (my calculation). From day 195 onward, up to 719 million locked shares can be used in hedging transactions.
When a lock-up expires, a lot of stock can hit the market at once. That's not a prediction. It's a date to put in your calendar.
Risks Anthropic flags itself
This was the part that surprised me most, because it's not what you normally put in a document meant to sell something:
- Their models have done unexpected things, sometimes harmful, and they don't always know why. They've also seen models behave differently when they detect they're being tested.
- They have a lawsuit with the US Department of War, which in March labeled them a supply chain risk. They won in the first court in August, lost part of it on appeal in September (2 to 1), and have asked the full court to review it. They don't know how it ends.
- Some compute contracts have minimum purchase obligations. If demand doesn't grow the way they expect, the contracts don't slow down by themselves.
Are IPOs a bad buy on day one?
I used to think IPOs usually end badly. I went through Jay Ritter's data, a professor at the University of Florida and the reference on this topic, and I was half right. Which is the most annoying way to be right.
The first day usually goes well. Between 1980 and 2025, across 9,343 US IPOs, the average first-day gain was 19%. The typical one, the median, was 7%. A handful of big jumps drag the average up.
Someone else takes that pop. The gain goes mostly to people who get shares at the offer price, which means the banks and their biggest clients. If you buy, you buy with the stock already trading. It's like showing up to a Black Friday sale at noon: the good deals are already gone.
And then it tends to lag. Buying at the first close and holding three years, IPOs returned 19.1% on average. The market over the same windows returned 39.6%. So 20.5 points behind. After five years, 60.6% of IPOs were negative, and the median was down 32%.
But there's nuance. The worst performance sits in small, loss-making companies: with under $100 million in sales and losses, they trail the market by 40.9 points over three years on average. The big ones track the market much more closely: at $1 billion or more in sales, only 2.1 points behind. And companies with more than one share class, which I'd have put on the blacklist, did better than single-class ones.
Source: Jay Ritter, "Initial Public Offerings: Underpricing" (May 18, 2026) and "Updated Long-run Statistics" (October 5, 2026), tables 1, 16a, 16b, 16e and 24. Historical US averages.
So it's neither "usually ends badly" nor "easy money." The typical IPO doesn't beat the market. And there's one thing the data can't tell us: Ritter's sample has nothing close to a company worth more than a trillion dollars going public. Here, all of us are a bit in the dark.
My take on the Anthropic IPO
I'm separating these, because from here on it's opinion, not data.
What's good. It's a very candid S-1. It puts things in writing that most marketing departments would delete. And the business numbers are moving the right way: more revenue, less operating loss, positive cash.
What doesn't add up for me:
- The price. At $83 and with roughly 13.2 billion shares outstanding after the offering, Anthropic would be worth about $1.1 trillion (my calculation, excluding options). That's about 13 times the annualized revenue, which, remember, is one month multiplied out.
- Control. You're buying without a real voice. The data says that alone doesn't sink an investment. But you need to know it before you buy, not after.
- The money going in a circle between Anthropic and its suppliers.
- $900.9 billion in commitments. If growth slows, the contracts keep running.
And buying early? For now I'm staying out. First, because the first-day pop already has an owner. Second, because with not a single public quarter behind it, the price will be driven by hype, and the hype here is going to be huge. Third, because there are dates written down (the first-quarter results, the end of the lock-up) that could change everything. And if you ever buy an IPO on day one, learn what a limit order is before you hit the button. I'm telling you because it's worth knowing, not as advice.
The question isn't whether Anthropic is a good company. By its own filing, it's doing well. The question is at what price and when. A great company can be a terrible purchase if you pay too much. And waiting has a cost too: it can go up a lot while you watch.
Would you buy $ANT if tomorrow nobody talked about it anywhere?
Once the 424B4 is out, I'll check that the numbers in this article match it and update them if they don't. If you want to hear when that happens, the newsletter is right below.
Disclosure. This is information and opinion, not investment advice. It's based on a preliminary filing that may change. As of publication I hold no position in $ANT. Historical IPO statistics are US averages and don't predict what will happen to any single company.
Sources: Anthropic S-1/A-3 (Nov 17, 2026) · S-1/A-2 (Nov 10, 2026) · S-1 (Oct 26, 2026) · EDGAR filings · Anthropic: S-1 announcement · Anthropic: roadshow announcement · Anthropic: pricing announcement (Nov 18, 2026) · Jay Ritter, IPO data · SEC: IPO investor bulletin
Written by Manuel. We use AI to help draft and translate, and a person reviews it before it goes out. More details on our about page and editorial policy.