Business Models and Pricing for AI

AI companies earn money in six main ways: they sell tokens through an API, subscriptions to consumers, seats to businesses, usage in the form of credits, actions or conversations, outcomes such as a resolved support ticket, and, increasingly, attention and commerce through ads and transaction fees inside assistants. The choice matters more than in ordinary software for two reasons. First, AI has a real marginal cost: every answer consumes compute (10), so a price that does not move with usage hands the vendor an open-ended bill. Second, AI increasingly does work that people used to do, so the natural reference price is not a software budget but a labour budget, and per-seat pricing, the standard of software-as-a-service for twenty years, shrinks when the software replaces the people who held the seats. This section maps the revenue models across the stack (chips and clouds aside), explains how each one handles variable cost and captures value, sets out the gross-margin structure from lab to application, examines the "wrapper" debate and the platform risk of building on someone else's model, and reviews the 2024–26 pricing experiments at OpenAI, Anthropic, Microsoft, Salesforce, Intercom, Cursor and others. A worked example prices one product four ways (token pass-through, seats, usage and outcomes) and stresses each against usage spikes, cheaper models, a reasoning upgrade and seat cuts; a checklist sets out questions to ask of any AI pricing model.