The Missing Meter

Meta sent a memo to 6,000 employees last week. AI usage costs were approaching billions. They had an internal leaderboard called “Claudeonomics” tracking consumption. 73.7 trillion tokens in roughly 30 days. The leaderboard is gone now. A centralized monitoring platform replaced it.

Uber hit the same wall months earlier. Five thousand engineers burned the entire 2026 AI coding budget by April. Monthly costs per power user ran $500 to $2,000. The company responded with a $1,500 cap per tool.

Per-token prices have dropped 98% since 2022. Enterprise AI bills tripled in the same window. The gap between the cheapest and most expensive model is 4,500x. Agentic workflows consume 5 to 30 times more tokens than a chatbot interaction. Organizations budgeted for the chatbot. They deployed the agent. Nobody updated the forecast.

Cloud infrastructure did this exact thing between 2018 and 2022. Bills exploded not because the cloud was expensive, but because nobody governed how it got used. That era produced FinOps. Cost allocation, usage policies, tagging, spend alerts. None of it was hard. All of it was necessary.

The companies not in this story are the ones that built model routing, tiered access, and cost visibility before handing every employee a frontier model. They solved an operations problem. Everyone else solved a pricing problem that did not exist.