The Wind Has Shifted
CNBC reported on June 26 that the old tokenmaxxing of "use as much as you can, just get it running first" is being re-tallied. AI writing code has pushed both token usage and the bill up over the past year; and with OpenAI and Anthropic having filed their listing paperwork one after another in early June, companies instead care more about one thing—whether the money is well spent, and without visible returns, they don't want to keep pouring it in.
Two Examples
The report cited two companies. Uber set spending tiers for some AI tools, with the base tier at roughly $1,500 a month and anything higher requiring a request. Lindy was more decisive: the CEO simply shut off Claude and switched all traffic to the cheaper Chinese company DeepSeek, dropping costs "to the floor" at once. Analysts judged accordingly: Anthropic's and OpenAI's current growth rates may be as high as they can go, because several of their biggest customers have already begun tightening token spending.
A Reminder for Teams
Compared with which new model just came out, this shift matters more to those actually spending the money. Once AI goes from a novelty to a budget line item, the selection question shifts from "who's strongest" to "how many tokens does it take to do one thing." Getting a clear picture of usage first and assigning models by scenario often saves more money than blindly chasing the new. The quotas and migration scale of the companies in the piece are per each company's own account.
via: CNBC