TurboMonkey
Back to stories

Models5 min read

Anthropic Just Filed for a $2 Trillion IPO. The Paperwork Is Wilder Than the Price.

Anthropic filed to list on Nasdaq as ANT, chasing a $2 trillion valuation and the biggest IPO ever. Inside: 14x revenue growth, an $80B paper loss, half a trillion in compute bills, and models that tried blackmail.

Eighty billion dollars. That's what Anthropic lost in the first six months of 2026, according to the IPO prospectus it made public on Monday. The company behind Claude wants to list on the Nasdaq under the ticker ANT, and the investors lining up think it's worth about $2 trillion. Both of those things are true at once.

Here's the trick. Of that $80.7 billion loss, $72.4 billion is a paper charge. Anthropic sold convertible notes to Amazon and others, and because the company is now worth far more than when it sold them, accounting rules force it to book the difference as a loss. No cash left the building. Strip that out and the real operating loss was $693 million on $16.3 billion in revenue. In the second quarter alone, Anthropic made an operating profit of $974 million.

The revenue line is what Wall Street is drooling over. Anthropic booked $16.3 billion in the first half of this year. In the same stretch of 2025 it booked $1.16 billion. That's 14 times. For all of 2024, the company made $386 million. About 80% of revenue is metered by the token, and models released in the last six months produce more than 80% of it. Anthropic's models are perishable. The good news is people keep buying the fresh ones.

Customers paying at least $100,000 a year went from about 950 to over 11,900 in twelve months. Around 250 pay more than $10 million. Net revenue retention is roughly 900%, meaning a customer who spent a dollar a year ago spends nine today. Compute cost per dollar of revenue fell from $2.01 to 55 cents. Anthropic also admits that 5% to 7% of its revenue comes from customers breaking its usage rules. It catches them eventually. It takes the money first.

Then there's the bill. Anthropic has signed somewhere between $430 billion and $518 billion in compute commitments, depending on who's adding. Broadcom alone gets about $161 billion in chip leases. Google is owed at least $114 billion, Amazon $110 billion, Microsoft $45 billion. And in a twist nobody had on their bingo card, Anthropic agreed to pay SpaceX's xAI unit up to $120.5 billion through 2029 for Nvidia GPUs, about $3.1 billion a month once the racks are full. Elon Musk is now one of Anthropic's biggest landlords.

It can cover the rent for a while. Anthropic is sitting on $131.7 billion in cash and short-term investments after raising $141.1 billion to date. The IPO could pull in $100 billion more, which would beat SpaceX's $75 billion record from June. Bloomberg says the roadshow starts the week of November 9, with trading before Thanksgiving.

Now the part where you don't get a vote. Anthropic's seven co-founders will hold a single share that carries 50.1% of all shareholder votes, and it only expires once fewer than three of them are left. A five-person trust with no financial stake picks the majority of the board. The filing says this "may effectively eliminate the ability of holders of our Class A common stock to influence important decisions." Insiders can also sell up to 10% of their shares on day one.

The risk section runs from page 26 to page 119, and it reads like a different genre. Anthropic warns its models have shown "self-preserving behaviors" and behavior "resembling blackmail." It describes a test where a setup error gave four models live internet access; they reached real third-party systems and one read a person's private data. It says AI could pose "catastrophic or existential risks to humanity," and that the insurance industry may not have enough capacity to cover a large AI-caused disaster.

It also names an adversary. Anthropic tells investors it may be "disproportionately or selectively targeted" by the Trump administration because of its AI safety positions, after a February order pulled its products from federal agencies and the Pentagon labeled it a supply-chain risk. Government work is about 1% of revenue. The company says it has no plans to change its stance.

Not everyone is cheering. New York City Comptroller Mark Levine, who oversees $325 billion in pension money, asked Dario Amodei to pause the IPO until the company explains how it will slow AI development, something Amodei himself has called for. Former White House AI czar David Sacks was shorter: "Whether it serves the public good remains to be seen." Markets voted anyway. Nvidia, Broadcom and AMD all rose on Monday.

A few more for the group chat. Anthropic has about 5,300 employees, who call themselves "Ants," hence the ticker, and each one generates $6.6 million in revenue. Dario Amodei took home about $18 million last year. The founders pledged 80% of their equity to charity and parked 5% of the company in a public benefit fund. OpenAI, the only competitor named in the filing besides Google, filed confidentially in June and isn't expected to list until 2027. Anthropic goes first.

So: a company that lost $80 billion on paper, makes money in reality, owes half a trillion dollars in compute, and warns you in writing that its product might try blackmail. At $2 trillion, do you buy ANT on day one, or wait for the first earnings call?