The cost of custom software collapsed. Delivery models did not.

Estimates of the AI services market range from roughly $14 billion to $589 billion in 2026 depending entirely on where the boundary is drawn. The narrow figure counts advisory work alone. The broad one counts every integration and managed-service engagement that touches a model. The gap between those two numbers is the actual business: implementation, not advice.

Demand is concentrated at the top. Global systems integrators capture the largest programmes, and they are structured to sell time. Beneath them sits a long tail of boutiques competing on rate rather than outcome. Very little of the market is organised around a fixed-price, fixed-duration deliverable — which is precisely what an AI-native delivery team can now offer without losing money.

The constraint on enterprise AI adoption is no longer model capability or budget. It is the gap between a workflow a company knows is broken and a working replacement for it. Most organisations never cross that gap because the traditional quote — six figures, two quarters — cannot be justified for a single process.

Collapse the quote and the addressable set of problems expands enormously. A process that was never worth a $200,000 engagement is worth a five-day fixed-price build, and a company that buys one of those will buy several. The economics only work if the delivery team has already built the underlying patterns — identity, tenancy, orchestration, deployment — and is reusing rather than rewriting them.