The biggest IPO ever attempted is an AI lab, and it is in six weeks
Key takeaways
- Anthropic is reported to be targeting a November listing at roughly a 2 trillion dollar valuation, raising up to 100 billion dollars.
- The company closed a 65 billion dollar Series H in May 2026 at a 965 billion post-money valuation, so the figure has roughly doubled in four months.
- Anthropic has not confirmed the numbers, and most downstream coverage traces back to the same two sources.
- A public AI lab files quarterly, which would give the first audited view of what frontier training and inference actually cost.
Two trillion dollars, in November, from a company that did not exist ten years ago.
The Wall Street Journal reported on 18 September that Anthropic is moving ahead with a listing at roughly that valuation, raising up to 100 billion dollars. At anywhere near those numbers it would be the largest offering on record by a wide margin, and not a close one.
How the number got there
Anthropic closed a 65 billion dollar Series H in May 2026 at a 965 billion dollar post-money valuation. Four months later the reported target is a little over double that.
The revenue line moved just as fast. Investors quoted by the WSJ expect annualised revenue above 110 billion dollars by the end of this year, up from a 14 billion run rate in February. Whatever else is going on, that is not a story about a company raising against a promise.
One caveat that most coverage skips: Anthropic has not confirmed any of it. The aggregators repeating these figures are mostly tracing back to the same two sources, so treat the exact numbers as reported rather than settled.
Why a listing changes what anyone can know
Private AI labs disclose what suits them. Leaked decks, selective run-rate figures, the occasional benchmark. Everything the public knows about frontier economics arrives filtered, which is why OpenAI's projected 278 billion dollar cash burn only surfaced through a reported presentation rather than a filing.
A listed company files quarterly. Audited. For the first time there would be a real number for what training and serving a frontier model costs, sitting next to a real number for what customers pay. That is more informative than any benchmark table, including the ones we use when comparing frontier models for enterprise buyers.
It also puts a price on a question governments have been circling. Sovereign AI programmes are being budgeted right now against guesses about what a frontier lab costs to run. A public filing replaces the guess.
The part worth watching is not the valuation. It is whether the filing arrives at all, and what the first quarterly report says about margins once there is a regulator on the other side of the number.
This article is reporting, not investment advice.
