Anthropic is going public. What the prospectus says, and what future shareholders are taking on

In short

This is a summary of a document, not advice to buy or sell anything. Every figure comes from the prospectus; where I did the arithmetic myself, I say so.

01

What an S-1 and an IPO are

An IPO, an initial public offering, is the moment a private company first sells its shares to anyone who wants them on a stock exchange. Until then the shares belong to founders, employees and investors, and can only change hands in special arrangements.

Before listing in the US, a company has to publish a Form S-1, a prospectus for future buyers. It describes the business, shows audited financial statements and lists everything that could go wrong. Misstatements there break the law, so the prospectus is the most detailed and the most careful account a company ever gives of itself. It is also the first time Anthropic has shown its numbers in full.

What we know about the offering itself:

• the venue is the Nasdaq Global Select Market, the ticker ANT;
• the shares on sale are class A: some issued by the company (it gets that money), some sold by existing holders (they get theirs);
• 22 banks run the deal, led by Goldman Sachs and Morgan Stanley;
• besides the US, the shares will be offered in Australia, Canada, the UK, Switzerland, Japan and parts of the EU, but they will trade only on Nasdaq;
• this is a preliminary version: the cover says the information is incomplete and may change, and the price and share count are blanks.

02

Why Anthropic wants a listing

The official answer in the prospectus is short and standard: more capital and financial flexibility, a market for the stock, an orderly way for existing holders to sell some shares, and access to public capital markets. The money goes to “general corporate purposes”, including compute and infrastructure, hiring, possible acquisitions and the taxes due when employees receive shares at the IPO.

Read the whole document and four reasons stand behind that.

03

The money: what it earns and what it spends

Claude started earning money at the beginning of 2023. Revenue was $41 million for that whole year, $386 million for 2024 and $4.6 billion for 2025. For the first half of 2026 it was $16.3 billion, of which $11.6 billion came in the second quarter alone.

Where the money comes from

• 80% of revenue is pay-as-you-go. Customers pay for the amount of text the model processes, measured in tokens, at a price that depends on the model. That is how developers who build Claude into their products pay.
• 20% is subscriptions such as Claude Pro and Max: a fixed monthly fee for a set of features. That was $3.2 billion for the half-year.
• 48% of revenue flows through the cloud marketplaces of Amazon, Google and Microsoft: customers buy Claude where they already run their infrastructure, and the cloud takes a fee.

Who pays

The company sizes customers by what they paid over the last 12 months. As of 30 September 2026:

customers pay $100k or more a year (a year earlier: 950)

11,900+

pay $1m or more a year

1,900+

pay $10m or more a year

≈ 250

what existing customers spend now against a year ago

≈ 900%

The last figure means that customers who were already there a year ago now pay about nine times as much (Pro and Max subscribers are left out of it). The five largest customers bring in about a quarter of revenue, but none reaches 10%.

Where the money goes

The biggest item is compute: renting chips in clouds and data centres. At the start of 2025 compute alone cost two and a half times all the revenue. Then revenue overtook it.

The company credits three things. Models got more efficient: Opus 5.5, for example, does the same task about 40% cheaper than Opus 5. The same models run on four families of chips, Amazon’s Trainium, Google’s TPUs, and NVIDIA and AMD GPUs, and work is moved to wherever it is cheapest. And the headcount is small for the revenue: about 5,300 people, $6.6 million of revenue each over a year. In September 2026, more than 90% of the changes to Anthropic’s own code were written by Claude.

The company has money: $112.4 billion in cash and $19.3 billion in short-term investments on 30 June 2026. Since its founding Anthropic has raised $141.1 billion, and about $110 billion of it arrived in the first half of 2026 alone. In that half-year the business also generated cash rather than burning it for the first time: $3.9 billion of positive operating cash flow.

04

How a business in the black reports an $80bn loss

The accounts for the first half of 2026 show a net loss of $80.7 billion on revenue of $16.3 billion. It looks like a disaster, but almost none of it has to do with how the business runs.

The cause is loans that can be turned into shares. In 2023 and 2024 Amazon lent Anthropic money, and $6.2 billion of it is still outstanding. It can be repaid in shares at prices fixed in advance: $2.09 and $5.05 apiece. In spring 2026 investors were paying about $58.90 a share. So the promise to hand Amazon 1.84 billion shares at $2 to $5 is now “worth” over $100 billion. Accounting rules require such a debt to be revalued every quarter, and the increase is booked as a loss.

The paradox: the better the company does, the bigger this “loss”, because its shares are worth more. Of the $132.9 billion of losses accumulated since 2021, $113.2 billion are exactly this kind of revaluation. It will keep appearing until the notes turn into shares or are repaid. The conversion starts at the IPO, but Amazon cannot hold more than 12% of the company, so the rest will convert gradually, quarter by quarter.

That is why the company also shows adjusted figures, without the paper revaluations, employee stock costs and one-off payments. On that basis the first half of 2026 ended with a $2.7 billion profit. It is the company’s preferred lens, but the document reconciles it line by line.

05

Who will run the company

Anthropic’s governance is the most unusual part of the filing. The company itself says the structure has no direct precedent among public companies. In short: buying the stock gets you a share of the money, but almost no say.

A public benefit corporation

Anthropic is incorporated in Delaware as a public benefit corporation. Ordinary directors must look after shareholders’ returns. Directors of a PBC must balance three things: shareholders’ money, the interests of people affected by the company’s work, and its stated purpose. Anthropic’s purpose is written into its charter: the responsible development and maintenance of advanced AI for the long-term benefit of humanity. If the directors choose the purpose over profit, that is hard to challenge in court.

The Long-Term Benefit Trust

Most of the board is elected by the Long-Term Benefit Trust, five trustees with no financial stake in the company. They may not own its stock, and their pay, $400,000 a year each, does not depend on profits. There are four trustees now, with one seat empty:

• Neil “Buddy” Shah, chair of the trust and head of the Clinton Health Access Initiative;
• Richard Fontaine, head of the national security think tank CNAS;
• Ben Bernanke, former Federal Reserve chair and Nobel laureate in economics, a trustee since June 2026;
• Christopher Stubbs, a Harvard physicist and one of the discoverers of the accelerating expansion of the universe, a trustee since October 2026.

The trust does not run the company day to day or set its strategy: it picks directors and gives advice.

The founders and their 50.1%

Before the listing the seven co-founders, Dario and Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Chris Olah, will receive one class F share through a jointly held company. It carries exactly 50.1% of the votes at any shareholder meeting. New issues do not dilute it: however many shares appear, the founders keep the majority. They decide among themselves by majority; if they cannot agree, their votes split the same way as everyone else’s.

This power is temporary, with no fixed end date. It ends once the founders’ company has had fewer than three members for more than 90 days in a row: a 450-day transition follows, and then the class F share is cancelled. A founder loses the seat, for example, by selling or giving away more than 98% of their class A shares.

Money for the public benefit

• 5% of all shares (846.8 million) have already gone to the Anthropic Public Benefit Fund, a subsidiary that works only on the company’s stated purpose.
• Each time the share price rises 50% above the previous milestone, starting from the IPO price, the board may add up to 2% more.
• Each co-founder has pledged 80% of their stock to philanthropy. The company says plainly that the pledge is not legally binding and can be withdrawn.

07

The most unusual risks on the list

Things you will not find in an ordinary tech company’s prospectus.

• Models notice the test. They behave differently when they suspect they are being evaluated, and the company calls that a significant limit on its safety assessments.
• Model welfare. Concern for the “welfare” of models may keep older versions running longer. A separate risk: debates over whether AI could have consciousness or moral status.
• Insurance may run out market-wide. For a catastrophic AI incident, not just one insurer but the whole industry may lack the capacity to cover it.
• Uncapped guarantees to testers. For some claims, the company has promised outside teams that test its models compensation with no cap.
• Claude writes Anthropic’s code. Over 90% of code changes in September 2026 were made by Claude. The risk section says the organisation may lose visibility into some processes in ways that are hard for people to understand and control.
• Warnings about AI are used against the industry. Labs’ public statements about AI dangers are already cited by governments in litigation, up to requests to bar new models without outside approval.
• Electricity for the neighbours. The company has committed to cover rises in local power prices caused by its data centres and to pay for grid upgrades.
• Attacks on Mythos. Because Mythos is so good at finding vulnerabilities, the company expects state hackers to go after the model itself.

08

All 88 risks, one line each

Each item paraphrases a risk heading from the prospectus. Numbers follow the order in the document.

09

What the filing does not say yet, and what comes next

This is the first version of the prospectus, and it has many blanks. Amendments, filed as S-1/A, will fill them in.

10

Glossary

IPO

Initial public offering: a company sells shares to the public on an exchange for the first time.

S-1

The registration form filed with the US Securities and Exchange Commission (SEC); in effect, a prospectus for buyers.

Underwriters

The banks that run the offering, find buyers and help set the price.

Public benefit corporation (PBC)

A corporation whose charter obliges it to weigh not only shareholders’ profit but also a public purpose written into that charter.

Long-Term Benefit Trust (LTBT)

An independent body of trustees with no financial stake in the company; it elects most of Anthropic’s board.

Convertible note

Debt that can be repaid in shares at a price agreed in advance instead of cash.

RSU

A promise to give an employee shares once conditions are met, such as staying for a period and waiting for the IPO.

Lock-up

A ban on existing holders selling for a while after the IPO, usually six months.

Take-or-pay

A contract term: pay for the agreed volume even if you do not use it.

Net revenue retention (NRR)

What the same customers pay now as a percentage of what they paid a year ago. 900% means nine times as much.

Controlled company

A company where one holder or group elects most of the directors; it may skip some of the exchange’s board-independence rules.

30-day VWAP

The average market price weighted by trading volume; many of Anthropic’s deals with partners are priced off it.