Two years of AI adoption ran on pilot budgets and enthusiasm. The invoices are now landing, and they are priced in a unit most operators had never heard of when they signed: the token.
Usage-based AI pricing has a property flat SaaS never had. It compounds with success. The more an AI feature gets used, the bigger the bill, and the companies that rolled AI into real workflows in 2024 and 2025 are discovering that the line item grows faster than the headcount it replaced. Finance teams that once asked whether AI works are asking what it costs per call, per customer, per workflow.
You can hear the shift happening. Token talk has climbed every month since December and now surfaces in one business conversation in seven. "Token usage" is up 1,227% since December. "Token consumption" is up 983% and has risen every single month without a miss. "Inference cost" is up 638%. Engineering vocabulary is showing up in the middle of pipeline reviews and hiring discussions, because the bill stopped being an engineering problem.
The tell is who is talking. The VC world leads, as it usually does: 28% of VC interviews in the last six months touch token economics, up from 11%. But in general business conversation the subject went from 1.3% to 6.4% over the same split, a fivefold jump. When operators adopt billing vocabulary, the spend has become real enough to argue about.
The responses are already visible. Model routing, sending cheap work to cheap models and hard work to expensive ones, was an infrastructure trick a year ago; mentions are up 148% since February. And the cost conversation is folding into a larger one: "pricing pressure" has risen five straight months in the same conversations, "restructuring" too, while the growth-era vocabulary of product-market fit and ideal customer profiles has been falling since March. Companies are moving from what to build to what it costs to run.
The AI budget conversation of 2024 was about permission. The one starting now is about the invoice.
Data: share of episodes mentioning each term, by publish month, across ~24,000 business podcast transcripts, July 2025 through August 2026. Trend runs measured on 3-month smoothed monthly reach.
Talk of token usage, token consumption, and inference cost has risen every month since December. Among investors it now touches 28% of conversations; in general business conversation it jumped fivefold, from 1.3% to 6.4%, in six months.