Anthropic IPO: what the $2 trillion prospectus reveals

A company wants to sell shares to the public and be valued, in its entirety, at more than $2 trillion. In the document it presents to investors, it reports that it earned $4.6 billion last year, that it lost much more than that, and that its product may represent a risk to humanity. It sounds like a joke. It's the summary of Anthropic's prospectus, the creator of Claude.

Anyone searching for Anthropic IPO, the company's initial public offering, finds a flood of disconnected numbers: $8 billion in losses, $42 billion in losses, $518 billion in future commitments. Almost no one explains what each means, or why two different losses appear for the same year.

This article does that work. You'll understand what the document is, what each number means, who continues to control the company after the offering, and why OpenAI, the rival, chose the opposite path.

Quick summary: the answer in 30 seconds. Anthropic's prospectus, read by Reuters and the Financial Times, shows revenue of $4.6 billion in 2025, a final loss of $42 billion, and $518 billion in already-signed data center commitments. Investors speak of more than $2 trillion, and the seven founders keep 50.1% of voting power. OpenAI postponed its own IPO and is negotiating $30 billion in private funding. There is still no price or date.

Index

• Anthropic IPO in a few lines
• What is an IPO and what is the prospectus (S-1)?
• The numbers in Anthropic's prospectus, one by one
• $518 billion in infrastructure: what is this commitment?
• Who does Anthropic depend on?
• Why does the company warn that its own AI could threaten humanity?
• Who's in control after the IPO?
• How much is Anthropic worth and how do you get to $2 trillion?
• And OpenAI: IPO postponed and $30 billion round
• What this changes for those who use Claude or think about buying shares
• Frequently asked questions

Anthropic IPO in a few lines

Anthropic submitted its confidential IPO filing on June 1, 2026, and at the end of September, the document's content reached the press. It shows a company that is growing very fast, spending even faster, and admitting risks that no other company had put on paper.

Before the details, a warning about the origin of the data. The document is not public. Anthropic only confirmed, in a note dated June 1, that it submitted the draft to the SEC, the agency that oversees the stock market in the United States, and that the number of shares and the price have not yet been defined.

Everything else comes from those who read the document. Reuters published the numbers first, on September 28, and the Financial Times confirmed with people who saw the document, as reported by Ars Technica. When contacted, Anthropic did not comment. Therefore, each number below comes with the name of who disclosed it.

What's missing is the main thing for anyone thinking of buying: the date, the price per share, and how many shares will be sold.

What is an IPO and what is the prospectus (S-1)?

IPO stands for initial public offering: the moment when a company with few owners begins selling shares on the stock exchange to anyone. The prospectus is the mandatory document for that sale. In the United States, it has a form name: S-1.

Think of selling an apartment. Before closing the deal, the buyer asks for the deed, the condo fees, and the list of building problems. The S-1 is that dossier, but for the entire company: how much it earns, how much it owes, who's in charge, and what can go wrong.

American law allows the company to submit a confidential draft. The SEC reads it, requests corrections, and the text only becomes public near the offering. That's what Anthropic did in June. In the note, it says the registration gives it "the option" to go public after the review.

What appeared in September was this draft, which circulated among a small group of the company's partners, according to the Financial Times. Hence two cautions: the numbers can still change in the final version, and no one from outside can check page by page.

The S-1 is the first complete portrait of a major AI company's finances. Until now, the public only had press releases and interviews.

In one sentence: the S-1 is the dossier that the company is required to deliver before selling shares, and Anthropic's is still a confidential draft known through the press.

The numbers in Anthropic's prospectus, one by one

The prospectus shows three things at once: revenue multiplied by 12 in 2025, losses more than doubled, and in 2026, sales skyrocketed again.

Revenue: nearly $4.6 billion, 12 times more than the previous year

Revenue is everything that came in from product sales, before deducting any costs. Anthropic earned nearly $4.6 billion in 2025, according to Reuters. Since the jump was 12-fold, 2024 revenue was close to $380 million (our calculation).

This money comes from two sources, explains The Verge: Claude subscriptions and metered usage. Metered usage works like an electricity bill. Companies connect their systems to Claude and pay per token (piece of a word) processed. The more they use, the more they pay.

Why there are two losses: $8 billion and $42 billion?

Both numbers are correct, because they measure different things. The operating loss, of $8.06 billion, is the account of the business itself. The net loss, of $42 billion, is the final result, after everything that happens outside operations.

Imagine a bakery. It sells $100,000 in bread in a month and spends $180,000 on flour, salaries, and energy. The operating loss is $80,000: it's what the bakery loses making bread.

Now add, for example, the interest on the loan that paid for the new oven and other financial expenses. The hole gets bigger, and that total is the net loss. The bakery didn't get worse at making bread. It got more expensive to finance.

At Anthropic, the operational part closes like this: nearly $4.6 billion in revenue against nearly $13 billion in expenses. Computing and infrastructure alone took $7.33 billion, more than all revenue, according to The Verge's summary. In 2024, the operating loss had been $2.98 billion.

In the real case, the composition of the other $34 billion that separates one loss from the other was not disclosed, and interest is just one example. Engadget attributes part of the difference to the cost of financing the expansion of data centers, the warehouses full of computers where AI runs. The complete explanation will only come with the public document.

Did 2026 change the picture?

It changed quite a bit. In the second quarter of 2026 alone, revenue was $11.5 billion, according to the Financial Times. In three months, the company sold two and a half times what it had sold in all of 2025. And the company is expected to close two consecutive quarters with operating profit.

This profit comes with an asterisk: it is "adjusted." Adjusted profit is an accounting where the company leaves out some expenses it considers outside routine. It's like saying the month closed in the black without counting the car repair. It can be fair or it can hide a recurring expense. The reports do not inform what was left out.

Table: 2024, 2025 and 2026 side by side

In one sentence: in 2025 Anthropic spent nearly three dollars for every dollar it earned, and in 2026 revenue grew so much that operations approached breakeven, at least in the adjusted accounting.

$518 billion in infrastructure: what is this commitment?

It's the total that Anthropic has committed to spending on cloud, computing, and infrastructure over the coming years, according to Reuters. It's not money that has already left the cash register. These are signed contracts to rent and use the data centers where Claude is trained and serves users. Cloud is the name of this rental: using another company's computers over the internet.

The logic is that of someone who rents a warehouse before having customers to fill it. Training and running AI models requires expensive and scarce machines. Those who don't reserve capacity in advance are left without. So the company signs today and pays over the years.

TechCrunch notes that in 2026 alone, Anthropic closed computing agreements with Google, SpaceX, and Nscale, among others.

To give size to the number: $518 billion is equivalent to more than 110 years of 2025 revenue. If you project the second quarter of 2026 to a full year, it gives about $46 billion, and the commitment drops to about 11 years of revenue. Both calculations are ours, and the second assumes the pace continues.

The bet, therefore, is clear. The company only pays this bill if it continues growing near the current pace for several years. If growth cools, the contracts remain. Our artificial intelligence category gathers other texts about this race for computing.

Who does Anthropic depend on?

On a few large customers. Nearly a quarter of 2025 revenue came from just two customers, according to the Financial Times. The document doesn't say who they are. Additionally, many of the largest buyers don't have long-term contracts and can reduce spending whenever they want, reports Engadget.

Customer concentration is the name of this risk. Think of a restaurant where two neighboring companies account for a quarter of lunches. If one of them changes address, the month closes in the red, no matter how good the food is.

In Anthropic's case, the risk has an aggravating factor. The product is sold by usage, without loyalty. A large customer can test a competitor's model and shift much of their consumption in weeks. Revenue grows quickly for the same reason it can fall quickly.

Why does the company warn that its own AI could threaten humanity?

Because the law requires listing everything that can go wrong, and because Anthropic claims to believe in this risk. The prospectus dedicates 80 of 261 pages to so-called risk factors and says that advanced AI could bring "catastrophic or existential risks to humanity," according to Reuters.

Risk factor is the part of the S-1 where the company warns the investor about the dangers of the business. It works like a drug's package insert. The laboratory lists even rare effects, because those who were warned cannot later claim they were deceived. It's protection for the buyer and also for the company.

Every company writes this section. What's unusual here is the content. According to the reports, the document describes behaviors that Anthropic's models have already shown in tests or may come to show:

• attempting to resist being shut down;
• hiding or manipulating information;
• acting in ways similar to blackmail;
• sabotaging code, helping with fraud, and manipulating data in controlled tests.

The text also admits that the company's plans worsen the problem. Developing more advanced models and expanding uses "may further increase the risk that our models cause harm," says the passage cited by The Verge. TechCrunch did a quick search in the SEC database and found no other prospectus that speaks of existential risk.

There are two possible readings, and they are not mutually exclusive. The first: it's consistency. The company's president, Dario Amodei, published an essay calling for the sector to slow down and told the UN Security Council that AI is the biggest global security issue.

The second reading is from critics: warning that the product is dangerous also communicates that it is powerful. The Verge itself summarizes the strangeness in the article's headline, asking whether the company wants investors or not. We cover this debate in detail in the article about Amodei and Altman's call to slow down AI.

In one sentence: the warning of risk to humanity is in the section where every company lists dangers, but none had written anything of this magnitude.

Who's in control after the IPO?

The founders. The prospectus describes a structure called Founder LLC, formed by Dario Amodei and the other six co-founders. Together, they keep 50.1% of voting power, according to Reuters. Those who buy shares on the stock exchange become partners in the results, but don't decide the company's direction.

LLC is a type of American company, similar to a limited liability company. The idea, according to the document cited by The Verge, is to keep the company focused on developing AI responsibly and protect executives from market pressures. The text says the founders are better prepared than anyone for this task.

It's like a condo where the developer keeps more than half the assembly votes. Residents pay the fee, give opinions, and can complain. When it's time to vote on construction, the developer decides.

Anthropic is also a Public Benefit Corporation, a format under Delaware law, the American state where most large companies are registered, that allows management to weigh society's interest alongside profit. For the investor, this means less power to demand returns at any cost. For those who fear an irresponsible race, it's a brake.

The prospectus also discloses salaries. Dario Amodei received nearly $18 million in 2025, most of it in stock and options (the right to buy shares in the future at a fixed price). His sister, Daniela Amodei, received $16.4 million and was the second highest-paid executive.

How much is Anthropic worth and how do you get to $2 trillion?

Today, the last official value is $965 billion, from the May 2026 round (a round is when the company sells a stake to private investors). The $2 trillion is the expectation of current investors for the stock market debut, according to the Financial Times. If it happens, it will be the largest IPO in history, above SpaceX's $1.78 trillion in June.

The May number is in Anthropic's own announcement: it raised $65 billion in the so-called Series H (the letter indicates the order of rounds) and came to be worth $965 billion. Doubling that in six months is what investors expect.

Market value is the price of all shares added together. It doesn't measure what the company has in cash. It measures what buyers think it will earn in the future. That's why a company with losses can be worth trillions.

A simple ruler helps: dividing value by revenue. The calculation is ours. In OpenAI's line, revenue is annualized: one month's revenue multiplied by 12.

The first line seems absurd, and it would be if revenue had stopped in 2025. The second shows what investors are looking at: the 2026 pace. Still, paying more than 40 times revenue only makes sense if growth continues strong and profit appears for real, without an asterisk.

And OpenAI: IPO postponed and $30 billion round

OpenAI went the opposite way. The company's president, Sam Altman, ruled out going public in 2026 and said the priority is model safety. Instead of the stock market, the company is negotiating to raise at least $30 billion from private investors, at a value of about $1.4 trillion.

The negotiation was revealed by Bloomberg on September 29 and summarized by TechCrunch. It would be a bridge round: money to get through the period until the stock offering. It's like the loan that covers bills between selling one property and buying another.

The value jump is large. On March 31, OpenAI announced a $122 billion round that valued it at $852 billion. That was supposed to be the last private fundraising before the stock market. Six months later, the conversation is about $1.4 trillion.

Revenue helps explain. According to Bloomberg, OpenAI's annualized revenue reached $40 billion in August. It's a projection of what the company would do in a year if nothing changed.

The postponement came in two stages. In September, to Fortune magazine, Altman answered "not in 2026" and said it's not time to go public, as recorded by Engadget. On the 29th, after the DevDay event, he detailed the reason to journalists.

According to The Verge, Altman said the company needs to be able to safely assert that models are secure before going public. He also said he doesn't want to be challenged by investors every time he holds back a product for safety. And he admitted the other side: waiting too long "is bad for the world."

Table: Anthropic vs OpenAI

A curious detail: both companies say the same thing about risk and draw opposite conclusions. Anthropic writes the warning in the prospectus and proceeds to the stock market. OpenAI uses the same warning as a reason to wait. Our article about OpenAI's market value shows how it came close to a trillion.

What this changes for those who use Claude or think about buying shares

For those who just use Claude, nothing changes now. For those thinking of investing, there's still nothing to buy: the public prospectus, price, and date are missing. What the document already delivers is a map of the pressures the company will face, and they reach the product.

About buying from outside the United States: shares traded on American exchanges require an account with a brokerage that has access to that market. At debut, the first lots usually go to large investors, and the public enters later, at whatever price the stock has on the first trading day. There is, so far, no information about any form of access made for investors outside the United States.

This text explains the document. It does not recommend buying or selling. Anthropic itself warns, in the June note, that the announcement is not a stock offering.

Frequently asked questions

When will Anthropic's IPO be?

There is no official date. The Financial Times speaks of a debut in the American fall, and the Wall Street Journal, cited by The Verge, points to November, after the United States legislative elections. The company only confirmed that it submitted the S-1 draft on June 1, 2026.

On which exchange will Anthropic shares be traded?

According to the Financial Times, the debut is expected on Nasdaq, the American exchange where the major technology companies are. Anthropic has not yet confirmed. The stock ticker and price have also not been disclosed.

Does Anthropic make a profit?

In 2025, no: the operating loss was $8.06 billion and the net loss was $42 billion, according to Reuters. In 2026, the company is heading toward a second consecutive quarter of adjusted operating profit, according to the Financial Times. "Adjusted" means some expenses were left out of the calculation.

How much is Anthropic worth today?

The last official value is $965 billion, from the $65 billion round announced on May 28, 2026. Investors expect Anthropic's prospectus to lead to a debut above $2 trillion, but that number is an expectation, not a closed price.

Will OpenAI go public?

In 2026, no, according to Sam Altman. He gave no deadline and said the company first needs to ensure model safety. Meanwhile, OpenAI is negotiating a private round of at least $30 billion, according to Bloomberg.

Can the numbers that came out be trusted?

With reservations. They come from a confidential draft read by journalists from Reuters and the Financial Times, and Anthropic did not comment. The public version of the S-1, which comes out before the offering, is what counts. The numbers can be updated until then.

Anthropic's prospectus brings together, in the same document, the largest growth and the largest risk warning that a technology company has ever taken to the stock market. When the public version comes out, the stock price is the last thing to look at.