The first time I heard the term tokenmaxxing, I was not surprised. After all, the article was about Meta, the company that renamed itself for its last windmill.
The practice was exactly what it sounds like: AI transformation measured in tokens burned, with per-employee spend reportedly reaching fifty thousand dollars a year at list prices and an internal leaderboard celebrating whoever consumed the most. Meta was the loudest, but I was shocked by how many otherwise sober enterprises followed suit.
To get employees using AI, enterprises everywhere made it clear they were watching AI usage. The thinking was that more tokens meant more adoption, more adoption meant more transformation, and so the metric of success became consumption itself.
It did not last, even at Meta. After reportedly realizing that internal AI costs were on track for billions, the leaderboard came down, and the memos went out. Tokenmaxxing died the way it was always going to die: after the bill.
After tokenmaxxing, companies went two ways. The first already has a name and a consulting practice: tokenomics. Deloitte has a CFO's guide. The FinOps Foundation, which officially launched a spin-off organization called the Tokenomics Foundation Tuesday, calls the token the atomic unit of AI value.
Calling the token the atomic unit of AI value is like calling the gallon the atomic unit of value for milk. The gallon measures volume. It tells you nothing about what the milk is worth. The token measures consumption, and consumption is cost, not value. Optimizing the token bill is still reading the bill. A cheaper bill is not a measurement.
