Many companies in the past believed in token maxxing. Everyone’s default logic was: the more employees use it, the more AI-native the company is.

So Meta’s token leaderboard and Amazon’s KiroRank are ranked by employee token usage—like the line Naval said, “waste tokens, save time”—turning themselves into AI Native organizations.
But token consumption isn’t a linear cost; it can suddenly balloon in agentic workflows and multi-step research scenarios with long contexts.
Chinese and U.S. companies are all realizing they need to budget token consumption again, the same way they do cloud compute power consumption. Uber is assessing AI ROI internally, Amazon and Meta have shut down the leaderboards, Coinbase is asking for more use of cheaper models, and Shopify has set up a “circuit breaker” to cut off sudden personal spikes in token consumption.