AI pricing gets harder as token use becomes unpredictable

AI companies and software vendors are struggling to set prices for services built on large language models because the underlying economics remain volatile. Tokens, the units used to process prompts and responses, are difficult to predict because small changes in prompts can alter outputs, different models behave differently, and the same request may not always produce the same result.

The challenge is growing as businesses adopt agentic systems that use multiple AI agents to make decisions and perform tasks. Goldman Sachs forecasts external token consumption will increase 24 times between 2026 and 2030 to 120 quadrillion tokens a month, while companies often discover how much they have used only when budgets run out or bills arrive.

Executives cited several ways to contain costs, including choosing models more carefully, writing more precise prompts and avoiding overly broad rollouts. Pricing remains unsettled for vendors building AI into products, with options including higher standard fees, results-based charging or bundles tied to incidents. Variable costs from large model providers could still disrupt those plans, making budgeting harder for corporate customers.