Tech Companies Curb AI Token Burn After 60.2 Trillion-Token Sprees Drove Costs and Outages
Updated
Updated · InfoWorld · Aug 12
Tech Companies Curb AI Token Burn After 60.2 Trillion-Token Sprees Drove Costs and Outages
1 articles · Updated · InfoWorld · Aug 12
Summary
Meta’s internal Claudeonomics leaderboard pushed 85,000 employees to consume 60.2 trillion AI tokens in 30 days, rewarding top users with ranks such as “Session Immortal” based purely on token volume.
That metric fueled “tokenmaxxing” — developers ran autonomous agents and stuffed prompts to climb rankings — driving waste, shoddy code and, at Meta, reported site outages tied to careless AI-generated changes.
Amazon scrapped a similar Kiro IDE leaderboard within weeks, while Salesforce and DoorDash shifted from mandating AI use to rationing it; Uber reportedly exhausted its 2026 AI budget in the first quarter.
Code-quality data point to the same pattern: duplication jumped 81%, refactoring fell 70% versus 2022, and short-term code churn more than doubled to 7.1% in 2025.
Companies are now building AI finops guardrails because tokens are an input cost — with output tokens costing up to 5 times more — and raw usage metrics often miss AI’s highest-value work in debugging, design and analysis.