Meta is pouring billions into AI infrastructure while simultaneously telling nearly 6,000 employees to curb internal AI spending. Internal memos reveal that in the past 30 days, Meta staff consumed 73.7 trillion tokens through internal tools, pushing costs toward billions of dollars.
The "Claudeonomics" Leaderboard and 73.7 Trillion Token Burn
An internal warning memo flagged runaway costs driven largely by third-party services, especially Anthropic's Claude. A leaderboard called "Claudeonomics" tracked token usage per team and individual, but instead of driving efficiency, it sparked a "tokenmaxxing" culture where high consumption became a status symbol. Actual output was rarely measured.
Immediate Measures: Dismantling Leaderboards, Deploying AI Gateway
Meta's first step was to remove the Claudeonomics leaderboard to kill the competition. Over the coming weeks, the company will roll out an "AI Gateway" dashboard offering real-time tracking of token consumption and spending, with automatic alerts for abnormal spikes. By early 2027, Meta plans to introduce formal token budgets and quotas for each team. A strategic shift is also underway: steering employees toward Meta's in-house coding assistant MetaCode to slash third-party API fees and dogfood the company's own product.
CTO: "Usage Does Not Equal Impact"
Meta's CTO Andrew Bosworth sent a separate memo with blunt language: "No one should use AI just for the sake of using AI." He stressed that token consumption is not a measure of progress or impact. Meta is not alone. Uber burned through a full year's AI tool budget in just four months and capped each employee at $1,500 per month. KPMG found only 26% of enterprises have full visibility into AI costs, leaving most to discover overspending only when bills arrive.