Research & Analysis
The cost of using AI is falling faster than the cost of adopting it
Running a model has become cheap. Changing how an organisation works around it has not - and that gap explains most of what businesses are experiencing.
What happened
The price of a model response kept falling
Across the major providers, the published price of generating text with a comparable model fell steeply over two years. Most coverage treats this as a story about technology getting cheaper. That is accurate, and incomplete.
Illustrative data
Using AI got cheaper; adopting it did not
Why it matters
The bottleneck moved inside the organisation
When use is cheap, the constraint becomes the work around it: cleaning data, redesigning a process, training people, and deciding who is accountable for an automated decision. Surveys that ask only whether a firm “uses AI” miss this entirely.
The question is no longer whether a business can afford the model. It is whether it can afford to change around it.
What is actually changing
Budgets are shifting from licences to people
The evidence points to spending moving towards integration and skills rather than software seats. That trend is early and uneven, and the data behind it is thin.
What we don't know yet
Adoption costs are self-reported in small surveys; there is no consistent public series. We will revise this episode if one appears, and note the change.
Data & sources
- Provider price lists, archived monthly, 2024–2026 · vendor documentation
- Peer-reviewed study on organisational AI adoption · journal article
- National business survey, AI module · official statistics
- Further sources listed in the published episode record when the series is typeset.