Tokenomics: Why making AI pay is tricky

As artificial intelligence moves from experimental labs into everyday enterprise workflows, a fundamental question remains unresolved: how much should AI cost? Buyers of AI services are finding it difficult to rein in expenses, while providers grapple with setting prices that are both fair and sustainable.

Part of the problem is that the neat abstraction of tokens as a unit of AI work does not translate into neat, controllable bills. Costs can be fast-rising and highly variable, and traditional software billing models don't always apply to outputs whose utility can vary wildly from one use to the next. Vendors, meanwhile, are experimenting with pricing models that struggle to balance infrastructure realities with market expectations.

That tension is turning tokenomics from a back-end technical detail into a front-line budgeting and governance problem. When neither side can reliably predict what AI will cost, it complicates everything from procurement to questions about who ultimately captures the productivity gains AI promises. Without clear benchmarks, both buyers and sellers are operating in the dark, risking inefficiency and mistrust.