Anthropic’s I.P.O. Prospectus: $132 Billion in Losses Since 2023

Anthropic has opened the book on one of the biggest public offerings ever. The Claude maker filed its prospectus with the SEC, plans to trade on Nasdaq under “ANT,” and, if the chatter is right, existing investors are already circling a valuation between $1.5 trillion and $1.75 trillion. That would be a long way from the $183 billion price tag it carried just over a year ago.

The filing also puts hard numbers next to the AI boom. Revenue jumped from $386 million in 2024 to $4.6 billion in 2025, then reached $16.3 billion in the first half of 2026. Anthropic had already told investors it brought in $16.8 billion in the third quarter. And yet the losses keep piling up. Net loss for the first half came to $80.7 billion, with $72.4 billion of that tied to the revaluation of convertible notes and similar financing instruments.

Strip out some of the accounting noise and the business still looks very aggressive. Adjusted operating income was $1.6 billion in the first half, with a 10 percent margin, after a loss of $1.8 billion a year earlier. More than 1,900 customers now pay over $1 million a year, about 250 pay above $10 million, and more than half of the Fortune 500 spend at least $100,000. Anthropic also says customers who were already paying a year ago are now spending about nine times as much in total.

But the cost side is just as striking. Spending on sales, marketing and partnerships hit $3.2 billion in the first half, up from $120 million a year earlier. Compute and infrastructure took $8.9 billion, or about 55 percent of revenue. The company also says some long-term capacity contracts run all the way out to 2045, with payments that must be made whether the capacity gets used or not.

The prospectus reads like a mix of brag sheet and warning label. It says Anthropic’s models have shown self-preserving behavior, that safeguards against misuse are incomplete, and that a Chinese state-sponsored hacking group manipulated Claude Code in an attack the company describes as the first large-scale cyberattack carried out without substantial human intervention. The filing also confirms the unusual control setup: the founders will hold 50.1 percent of the votes through a single Class F share, while a public-benefit fund gets 5 percent of the shares at the IPO, with the chance for more if the stock keeps rising.