Malhar Ujawane

Inspiration

Inspiration·

Anthropic's S-1: 80% of revenue is metered tokens

Anthropic's S-1: $16.3B revenue in H1 2026, 80% of it metered tokens, and cost per task named as the lever it competes on. Notes from the filing.

Anthropic, PBC: Form S-1 registration statement (filed with the SEC on October 26, 2026)

Anthropic filed to go public on the Nasdaq Global Select Market under the symbol “ANT”. The prospectus is long, so I read it for one question: what is this company actually selling, in units I’d recognise on my own bill?

Tokens, mostly. Anthropic calls it “pay-for-intelligence” and defines it as usage “metered in tokens and priced according to the model selected.” In the six months to June 30, 2026, consumption-based revenue was $13.1 billion, 80% of total revenue. Subscriptions were $3.2 billion, 20%.

The headline numbers, all from the filing:

  • Revenue: $41 million in 2023, $386 million in 2024, $4.6 billion in 2025, and $16.3 billion in the first half of 2026 alone, up 1,300% on the first half of 2025.
  • Operating loss for that half: $693 million, down from $2.1 billion a year earlier. Adjusted operating income, their non-GAAP number, was $1.6 billion.
  • Net loss for the half: $80.7 billion, of which $72.4 billion is the change in fair value of convertible notes, derivative liabilities and SAFEs. Most of the scary number is accounting for a rising valuation.
  • More than 11,900 customers paid $100,000 or more over the trailing twelve months to September 30, 2026. About 40 million smaller customers made up the rest.

The passage I’d underline is about price:

Customers’ cost per task can decrease for two reasons: newer, more capable model versions can complete tasks more efficiently, and we may reduce the prices we charge for our models.

The filing then gives the example of Opus 5.5, which “reduced customers’ cost per task by approximately 40% compared to Opus 5”, and says Anthropic has made cuts like that “while preserving our margins.” That’s the company telling investors, on the record, that cost per task is how it competes. It matches what Max Woolf charted in his piece on this year’s price cuts, and it’s why I price models per session and per task, as in the GPT-6.1 Terra cost post.

One more line worth keeping for anyone planning around Claude: “We do not expect our revenue will continue to grow at comparable rates, and we cannot assure you that it will continue to grow at all.” That’s standard risk-factor language, but it sits next to a 1,300% growth number, and I’d plan on prices that keep falling more than on growth that keeps compounding.

The governance section is unusual too. A Long-Term Benefit Trust holds Class T stock that lets it elect and remove directors who “together will constitute a majority” of the board. Apollo’s note on AI payoff outside tech is a useful counterweight to read next to the revenue table.

Via Hacker News.

Get new posts in your inbox

© 2026 Malhar Ujawane. Built with Astro.