Your AI budget just went uncapped.

Between the highs and lows of the World Cup last week (that’s far too many years of hurt now!), I helped a friend last weekend re-model the commercial framework for one of her vendors. She’d just been offered a renewal quote and found the shape of the contract had materially changed: same vendor, same product surface, but the pricing had moved from a flat platform fee to a consumption model. Per-seat had become per-call. The number at the bottom was now a range, not a fixed value.

This is happening quietly across the AI stack right now. Vendors are rebuilding their commercial model to match their own variable infrastructure costs, and they’re passing that variability straight through to the customer.

Most boards approved AI budgets assuming they were buying software with a fixed licence, the kind you sign in Q1 and know what you’ll spend in Q3. Consumption pricing doesn’t behave that way, because usage scales with adoption, and adoption is precisely what every executive has been telling their teams to drive.

What’s missing in almost every organisation is a functioning token-consumption governance layer that spells out who’s allowed to spend what, on which model, against which outcome, and with what approval threshold. Finance won’t own it, procurement won’t touch it, and the AI team quite reasonably doesn’t want to be the one saying no.

An uncapped bill isn’t really a pricing surprise so much as an accountability gap the new contract has quietly made visible.