Definition
Cost per million tokens
The posting
Cost per million tokens is the unit price of inference, quoted per million input or output tokens. It is how inference spend is sized and metered, and for a paid product it is the basis on which that spend maps to cost of revenue rather than operating expense, making it the unit cost behind AI gross margin ASC 350-40-35.
Why it maps to COGS
When each customer request consumes a measurable number of tokens, the cost of serving is variable and real. Costing that consumption per million tokens lets you assign it to cost of revenue, so gross margin reflects the true marginal cost of the AI product.
Posts to
Primary sources
Ledger current as of 2026-07-24. A position and a citation, not accounting advice. See how we cite.