All news

Anthropic's S-1 moves the margin fight from compute to distribution

Cobalt marbles roll down a wooden chute and squeeze through a brass turnstile toward an empty bowl

Compute fell from 177% to 48% of Anthropic's quarterly revenue in a year, while fees to the clouds that resell Claude became its fastest-growing cost.

Anthropic spent $1.77 on compute for every dollar of revenue in the second quarter of 2025. A year later it spent 48 cents. The public S-1 it filed on 26 October to list on Nasdaq as ANT shows first-half revenue of $16.3 billion, against $1.2 billion a year earlier, and a second-quarter operating profit of $974 million.

The filing settles one argument and opens another. Five months after the confidential draft that put frontier AI pricing on a path to public markets, the numbers in the prospectus show that a frontier model can pay for its own compute at scale. The margin freed from compute does not all stay with the lab, though. The cost line now growing fastest is distribution, the fees Anthropic pays Amazon, Google and Microsoft for selling Claude through their cloud marketplaces. The next contest over AI margin runs through the checkout, not the data centre.

The mechanics sit in the prospectus. Marketplace sales made up 11% of revenue in 2023, 47% in 2025 and 48% in the first half of 2026. The platforms bill the customer, collect the cash and pass it on net of their fee. Anthropic books the gross amount as revenue and records the cut under sales, marketing and partnerships, a line that rose from 10% of revenue to 19% in the first half year on year, mostly on channel fees. The same three companies also sell Anthropic its compute, with commitments to Google Cloud and AWS of at least $110 billion each. They earn on the way in and again on the way out, the position their contracted backlog already pointed to in the spring.

The lesson carries down the stack. An AI-native application that reaches enterprise budgets through a marketplace should treat the fee as cost of goods, not marketing, because at scale it behaves like one. A gross margin quoted before channel fees is a margin the company will not keep. For Fanaura, a family office investing in AI-native companies, cost of distribution now belongs next to cost of inference in every model, and the companies Fanaura looks for are those whose direct relationships with customers grow faster than their resold revenue.

The strongest objection is that the clouds sell reach, not a toll. Large enterprises hold committed cloud budgets, and a marketplace listing lets them spend that money on Claude without a new procurement cycle. More than 1,900 customers now pay Anthropic over $1 million a year. On this reading the fee is a customer acquisition cost that buys accounts no direct sales team could open as fast.

The signal that settles it is the marketplace share itself. It quadrupled in two years and then flattened at 47 to 48%. If Anthropic's first reports as a listed company show that share falling while the partnerships line drops back toward a tenth of revenue, the lab has taken its checkout back. If both hold, the clouds have turned AI's best new margin into their own.

For founders

Building an AI-native product that sells directly to enterprise customers? Send us the project.

Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.

Apply in five minutes
Subscribe

Get the news by email

One short note each weekday. No marketing, and you can unsubscribe at any time.