The shift from tokenmaxxing to efficiency is going to break a lot of AI pricing models

For two years the assumption behind almost every AI product was that usage only goes up. You priced per token or per seat, watched consumption climb, and growth took care of itself. This week that assumption started cracking in public. UBS put a number on it in a report a few days ago. Around 60% of enterprises have already put guardrails on their AI spend. They found individual users burning up to $35,000 a month and some teams running 200% over their token quotas. Nobody is quitting AI. They are getting ruthless about how they use it: routing easy tasks to cheaper models, pooling tokens, capping heavy users, and drifting toward open-weight models for anything that does not strictly need a frontier model. This is a problem for anyone who built a product on top of these models. Most of them copied their pricing straight from the providers, per token or per request. That works while the buyer wants more. It turns against you the moment the buyer's main goal is to use less. You end up charging for the exact thing your customer just hired a team to shrink. The products that come out of this stronger probably stop billing for consumption and start billing for outcomes. Cost per resolved ticket. Cost per shipped PR. Cost per closed lead. The tokens become an input cost you manage in the background, the way a SaaS company manages its AWS bill, instead of the headline number on the invoice. It also changes the product itself. If buyers now reward efficiency, then routing, cost visibility, and outcome-based pricing stop being back-office concerns and start being features. if efficiency is the new default instinct, is usage-based pricing still the right model, or is it quietly on the way out?