anthropic
Indirect channels now carry 40% of Anthropic's ARR
Claude News
anthropicMore than 40% of Anthropic's ARR now comes through indirect channels, a line the company crossed in 2Q26. The figure comes from the Tokenomics Model thread published on X, which puts API and B2B revenue at the vast majority of net new ARR dollars across the labs.
At a glance
- Bedrock, Foundry and Gemini Agent Enterprise carry the indirect traffic, and API and B2B lines together account for the vast majority of net new ARR dollars being added at the labs.
- Depending on the deal the hyperscaler takes an IaaS fee, a revenue share, or both, while the lab remains seller of record and books the gross ARR figure.
- The revenue share lands in Sales and Marketing at the lab, so identical token spend produces different revenue and margin outcomes on each side of the deal.
The mix shift changes what an ARR number describes. A lab that acts as seller of record reports gross dollars while a share of the economics has already moved to the channel, which means headline growth and margin quality are drifting apart at the labs and, on the hyperscaler side, feeding a revenue line that appears to carry better incremental margin than raw infrastructure rental. Model comparisons built on a single blended monetization rate are likely to misstate both sides.
The split matters because indirect revenue is not monetized the way direct revenue is. Hyperscalers monetize their Token as a Service businesses at higher rates than their IaaS businesses, and the incremental margin that moves to them is booked at the lab as a Sales and Marketing cost, which pulls down EBIT margins.
The trade runs the other way on customer acquisition. The labs benefit from not deducting the cost of a large direct enterprise sales force from this revenue, and instead ride the distribution hyperscalers already hold through their cloud contracts with enterprises in the Global 2000.
Compute allocation follows the same logic. Indirect workloads run on hyperscaler megawatts while serving lab models, so capacity attribution in a given quarter only works with deal-level detail rather than a single blended assumption about where a lab's tokens are being served.
Where the deal-level numbers sit
The August update to the Tokenomics Model breaks out every hyperscaler and lab pairing separately: fee structure, gross versus net revenue, and megawatt allocations tied to each lab's own direct and indirect split and its revenue per megawatt. The thread does not disclose the individual fee terms, the dollar value of Anthropic's ARR, or how the 40% share divides between Bedrock, Foundry and Gemini Agent Enterprise.
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.
We only use your name and avatar from Google. We never store your email address.
