Why read a prospectus
On 26 October Anthropic filed Form S-1 with the SEC, the US stock market regulator: the prospectus a company uses to ask to sell its shares to the public.1 In a sales pitch a company says what suits it. In a prospectus it has to list for future shareholders what could go wrong as well, because leaving out something material invites a lawsuit. It runs to more than three hundred pages, and not all of them please the people who signed it.
The document still has blanks in it. As of 30 October neither the price nor the number of shares has been set. What is known is that Anthropic wants to list Class A shares on Nasdaq under the symbol ANT. According to Bloomberg, meetings with large investors could begin in the week of 9 November, and according to the Wall Street Journal trading could start in the week of the 16th.2 These are reports, and the company has given no dates. The valuation is a press estimate too, between 1.8 and 2 trillion dollars, against 965 billion in the last private round, in May.3 The prospectus provides for public offerings outside the United States as well, including in some countries of the European Economic Area, without saying which.4 I am reporting news here: whether to buy is another matter, and not the business of this piece.
The accounts: record revenue, and a loss that needs careful reading
Revenue was 41 million dollars in 2023, 386 million in 2024 and 4.6 billion in 2025. In the first half of 2026 it reached 16.3 billion, against 1.2 billion a year earlier, a rise of 1,300%. On the same page Anthropic says it does not expect growth to continue at this pace, and admits it has struggled to forecast its own revenue.5
The net loss for the half was 80.7 billion, but 72.4 billion of that is an accounting item. Anthropic has also financed itself with convertible notes and similar instruments, money received today in exchange for shares tomorrow. On the balance sheet these are revalued when the company's worth goes up, and the increase appears as a loss without a dollar leaving the building. Another 8.8 billion comes from a similar mechanism on financing agreements with partners.6 The more Anthropic is worth, the deeper into the red its accounts go. The cost arrives in another form, because those instruments become shares and the cake is cut into more slices.
The net loss and how much of it comes from revaluing instruments
To understand the business you have to look at the operating loss: 693 million in the half, against 2.1 billion a year earlier. In the second quarter it turned into an operating profit of 974 million, and over the half operations generated 3.9 billion of cash. That result is incomplete, though. Shares promised to staff vest with the listing and have not yet weighed on the accounts. Had Anthropic listed on 30 June, it would have booked 2.9 billion of cost straight away, with another 23.7 billion to spread over a little more than two years.7
The computing bill
The heaviest figure sits in a note to the accounts. In the section on contractual commitments the amounts are still blank, but note 16 says that between 1 January and 1 June 2026 Anthropic signed contracts, mainly for computing and data centres, with minimum future payments of about 417 billion dollars. Note 17 adds about 245 billion signed after 30 June, part of it subject to contractual conditions.8 Fortune added up the eight main contracts described in the prospectus and reached 657 billion, of which 498 billion cannot be cancelled.3
These payments are spread over many years, up to twenty, so they should not be read as a single invoice. They do give the size of the bet. In the first half computing cost 8.9 billion, 55% of revenue, and at the end of June Anthropic held 131.7 billion in cash and short-term investments. The bill also reaches people living near data centres, the invisible weight of AI: the prospectus mentions local protests over the environment, land and water, and the company's pledge to cover electricity price rises caused by its sites.9
The bet on computing, in billions of dollars
The company spells out the risk itself. Revenue depends mainly on consumption and much of it is not tied to long contracts, while computing is committed for years and has to be paid for even if it sits idle. If growth did not keep pace, it warns, Anthropic might be unable to meet its commitments, with defaults spreading to other contracts and debts. Many of its largest customers, moreover, can cut their spending whenever they like.10
As for who sits on the other side of those contracts, what I wrote about who holds the keys applies. Amazon, Google and Microsoft are at once shareholders, suppliers, customers and competitors of Anthropic. Google has bought about 10 billion dollars of shares and can provide up to 30 billion in financing, while Anthropic has committed to spending at least 114.5 billion with Google.11
Buyers have little say
The shares on sale, Class A, carry one vote each. That looks normal, but the prospectus provides for five classes of shares. A single Class F share, held by a company of the seven co-founders, carries 50.1% of the votes and no right to a dollar of profit. The Long-Term Benefit Trust, a body that holds a power for a stated purpose, has five trustee seats, one of them now vacant, and holds Class T: as a rule it elects and removes four of the seven directors. The other three are voted on jointly by Classes A and F, which in practice means the founders.12
The prospectus says so plainly. Holders of Class A shares generally cannot affect the outcome of any vote, "regardless of how they vote", unless the founders fail to agree. Aswath Damodaran, who teaches finance at New York University and puts the value of Anthropic's equity at about 1.475 trillion dollars, writes that this structure strips investors of their power.13
The five share classes after the listing
| Class | Who holds it | What it votes on | Economic rights |
|---|---|---|---|
| A | Stock market buyers and current shareholders | One vote per share, but Class F already has the majority | Yes |
| N | Mainly strategic partners such as Amazon and Google | Nothing, except where the law requires | Yes |
| F | One share, held by the seven co-founders' company | 50.1% of the votes | No |
| T | The Long-Term Benefit Trust | Elects and removes the majority of the board | No |
| E | Employees, excluding the co-founders | A few matters listed in the prospectus | No |
The whole construction is meant to protect the mission from the market. Anthropic is a public benefit corporation (PBC): directors have to balance shareholders' returns, the interests of those affected by the company and a purpose written into its charter, the responsible development of AI for the benefit of humanity. Taken one at a time, the safeguards are thin. No rule requires the public benefit to be measured against an outside standard: the board sets the criteria, they will be qualitative, and the report comes every two years. The trustees appoint one another, and so far have chosen candidates in close collaboration with the company's top management. The trust gets its funds from Anthropic, 8.4 million dollars in the first nine months of 2026, and each trustee receives 400,000 dollars a year, fixed and unrelated to results, and may not hold the company's shares.14
Then there is the most quoted promise. The co-founders have pledged 80% of their personal equity in Anthropic to philanthropy. The prospectus describes the pledge as voluntary, revocable and not binding, and adds that it does not touch Class F: as they give, their economic stake will fall and their votes will stay the same. The Anthropic Public Benefit Fund, which receives shares equal to 5% of the capital, is a subsidiary too, and its first managers include Dario and Daniela Amodei and the chair of the trust.15
What Anthropic fears
Among the risk factors there are three that also concern people who will never buy a share.
The first is the US government. Anthropic bans the use of its models for fully autonomous weapons and mass surveillance. In early 2026 the Department of War, the Pentagon, suggested it could use the Defense Production Act, the law that gives the government special powers over industry, to force it to drop those limits. On 27 February the president ordered federal agencies to stop using its technology, and the Pentagon designated it a supply-chain risk. On 27 August a California court ruled that those measures were retaliation against free speech. On 25 September the Washington appeals court, by two votes to one, upheld instead the designation made under a different law, and Anthropic asked the full court to rehear the case. The company admits that keeping those limits has already had commercial consequences.16
Washington has another lever too. On 12 June the Department of Commerce imposed worldwide export restrictions on two models, Fable 5 and Mythos 5, and Anthropic switched them off for all customers until the order was lifted on 30 June.17 That decision made them vanish for people working in Europe too, another face of dependence on American technology.
The second is work. In May 2025 Dario Amodei told Axios that AI could wipe out about half of entry-level office jobs and push US unemployment to between 10% and 20% within five years.18 The prospectus treats the same subject as an investor would: if job displacement speeds up, governments may slow the adoption of AI, and the brakes on growth it lists include taxes and other measures aimed at AI companies or at job losses. It adds that the company's own statements on these matters may be used against it. Meanwhile it boasts of more than 6.6 million dollars of revenue per employee over the last twelve months.19 How many jobs are being lost to AI is still disputed, and the figures we have are partial.
The third is copyright. The settlement with authors in the Bartz case cost 1.5 billion dollars and was approved in July 2026. Lawsuits from music publishers and from Reddit remain open, and the prospectus notes that in some countries outside the United States fair use, on which Anthropic builds its defence, is generally not recognised.20 I have written about how the courts are moving in the piece on the first appeals ruling.
What I think, and what comes next
The growth is there and the bet is enormous. What the company is worth is for the market to decide. What interests me is who answers for this technology.
I am not moved by the shareholder with little say. On the future of AI I would trust a quarter on the stock market even less than a board of directors. But Anthropic's solution puts control in the hands of seven people and a trust that renews itself, with a public benefit that measures itself and promises that can be revoked. The prospectus itself admits possible conflicts when safety and business diverge.14 A technology of this weight needs a check that sits outside the company.
In the clash with the Pentagon Anthropic is right, and I say so knowing that it is my supplier. Weapons that choose their own targets and mass surveillance should not be built. That is exactly why a company's usage policy is not enough, least of all a company that, once listed, will answer to the stock market too. Those limits belong in a law, decided by a parliament and binding on every supplier.
The same goes for work. What the prospectus lists among its risks, a tax on AI companies or measures for people who lose their jobs, is a possible protection for working people. If a company warns its investors that displacement may speed up, governments already have a reason to decide now who will pay.
Before its debut Anthropic will have to fill in the blanks: price, number of shares, third-quarter accounts, the table of commitments. The second episode, out before the listing, will go into the detail of who is in charge and what remains of a public benefit corporation when it has to answer to the market. The third will follow the first day on Nasdaq: the price, the first session and what changes for people who use Claude.