Skip to content

Glossary entry · Cost

AI usage cap

A limit on how much AI a person, team or API key may consume in a period, set in money, tokens or credits, after which use stops, slows or needs approval. Also called a token budget or AI allowance. A cap controls the size of the bill; it does not say which part of the bill produced value.

Why it matters

Caps are now a common response to AI spend that has outrun its budget. Set per person, they bind only on the heaviest users, who may also be the people getting the most from the tool, while low-value use below the line carries on. Budgets set by use case, with exemptions earned by evidence, control cost without making that trade blind.

A usage cap can be set at several levels: a daily or monthly allowance per person, a pooled budget for a team or cost centre, or a hard limit on an API key or session. The economics differ. A pooled team budget absorbs the natural spread between light and heavy users. A per-person cap set near typical use cuts the heaviest users first, whatever their use is worth.

Caps still have a place as a backstop against runaway agents and as a defence against cost-exhaustion attacks. As the main budget control, they are a cost decision standing in for a value decision. Token Budgets Are Not a Value Strategy works through an example and sets out the alternatives: budgets by use case, routing before rationing, cost per outcome and attribution coverage.