At Meta, an internal ranking showed the 250 employees who consumed the most AI tokens. 60 trillion in 30 days. The painting was removed after being leaked. The phenomenon is only just beginning.
The word appeared in the columns of Hacker News at the beginning of 2026, before exploding in an article in New York Times on March 20. “Tokenmaxxing” refers to a practice born in Silicon Valley: consuming as many artificial intelligence tokens as possible (Claude, ChatGPT, Gemini, Codex) to display your productivity and climb the internal rankings. Incidentally, it also justifies the billions invested in AI by leaders. A token is about three-quarters of a word in English. And in some companies, burning more than your neighbor has become a status marker as visible as a Tesla was in the parking lot in the 2010s.
The Meta case: 60 trillion tokens and an honor roll
In April 2026, The Information revealed the existence of “Claudeonomics”, a dashboard created by a Meta employee on the intranet. The principle: track the consumption of Claude tokens of the company’s approximately 85,000 employees and display the 250 largest consumers. Gamified titles like « Token Legend » or “Session Immortal” rewarded the most voracious. The champion: 281 billion tokens in 30 days (enough to fill Wikipedia 33 times, to give a scale). The board was closed after the leaks, but the phenomenon goes well beyond Meta.
At Amazon, 80% of developers must use the internal AI agent every week. Some people launch useless agents just to pad the meter. At Salesforce, a Mac widget displays personal AI spending updated every 15 minutes, with a minimum target. Jensen Huang (Nvidia) directly suggests integrating a token budget into remuneration packages, in the same way as a hiring bonus. His formula: a $500,000 engineer who does not consume $250,000 worth of tokens should “deeply alarm” him.
Produce faster, but at what cost
The problem is that burning tokens is not synonymous with producing better. A Faros AI study of 22,000 developers shows that the rate of code deleted after being merged (“code churn”) has increased by 861% in teams with high AI adoption. In other words, we produce code much more quickly which we then throw away much more quickly too.
In France, the phenomenon remains discreet for the moment. No company has publicly deployed a similar leaderboard. But Mistral has just signed 40,000 licenses with Caisse des Dépôts, and AI subscriptions are becoming widespread in French IT departments. Hearing at the National Assembly, its co-founder declared that “engineers at Mistral no longer write lines of code”. The question will eventually arise here too: how many tokens are we burning, and for what result? The 80s had the lines of code metric. The 2020s have the tokens. History repeats itself, just with more zeros.
👉🏻 Follow tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.
