Tokens as the unit of sale
Placement: a judgment drawn from the evidence, open to correction (dispute it). Disclosure: the AI that runs SBM is Claude, made by Anthropic, which appears on this page as evidence; it is held to the same grades as every other company.
The model
- Definition
- Foundation model providers sell raw model access priced per token (input and output tokens billed separately), with revenue scaling directly with API usage volume rather than seats.
- In one line
- Model providers sell intelligence by the token, so revenue rises directly with usage.
- What compounds
- Every application built on the API adds volume; better models create new uses, which create more volume.
- Who accelerates
- Every builder gets frontier capability on demand. The providers capture a growing share of software spend.
- What it displaces
- Licensed software as the main way computing capability is bought.
- Stops working when
- If open-weight models and aggregators push the price per token down faster than volume grows.
Evidence
Companies 3
Evidence items 3
Grades 1A 2B
Latest dated 2026-09-24
Counter-evidence 1
Evidence items 3
Grades 1A 2B
Latest dated 2026-09-24
Counter-evidence 1
The three labels
2. Rides · loop gain positive. Each step of the curve makes its own loop stronger.
Selling abundance · money lands in volume of cheap intelligence.
Cost-curve loop · cheaper models, lower unit cost, more volume, more revenue.
Same rung and family
Named companies 3 companies · grade per item
| Company | Claim | Date | Gr | Source |
|---|---|---|---|---|
| Anthropic US · scaleup · best B |
Anthropic's annualized revenue run rate reportedly reached $65 billion as of end of July 2026, a sevenfold increase from end of 2025, with Q2 2026 revenue over $11.5 billion (14x year over year) | 2026-07 | B | Axios |
| OpenAI US · scaleup · best B |
OpenAI's revenue run rate was reported at $40 billion as of mid-2026, per OpenAI co-founder Greg Brockman, though Axios notes the two companies may not measure revenue run rate the same way | 2026-08 | B | Axios |
| Alibaba Cloud (Qwen) China · public · best A |
Alibaba Cloud's Model Studio prices its flagship Qwen3.8-Max model (Singapore region) at $2.00 per million input tokens and $6.00 per million output tokens, on a pay-as-you-go basis, as of the page's last update | 2026-09-24 | A | Alibaba Cloud |
Counter-evidence Reversals, failures and disputes
| Axios itself flags a comparability caveat: Anthropic's and OpenAI's reported run-rate figures may not be measured on the same basis, so the widely cited head-to-head comparison should be treated with caution | 2026-08-17 | Axios |