Meta May Cap Engineer AI Spend Within 2 Years as Token Costs Near Salaries
Updated
Updated · TechCrunch · Jul 14
Meta May Cap Engineer AI Spend Within 2 Years as Token Costs Near Salaries
2 articles · Updated · TechCrunch · Jul 14
Summary
Adam Mosseri said Meta may need per-engineer AI token caps within one to two years if a strong engineer’s usage cost approaches their salary.
Meta has no token caps now, but Mosseri said token budgets will have to be managed like payroll or other operating expenses and allocated by expected ROI.
Billions of dollars in projected 2026 AI costs already pushed Meta to shut an internal token-spend leaderboard, which Mosseri called one of the “silly things” inflating usage.
The cost pressure is spreading across tech: Uber burned through its 2026 AI coding budget by April, while Microsoft canceled Claude Code licenses and steered engineers to Copilot CLI.
Mosseri expects model providers to eventually cut prices in a competitive pricing war, but said companies first need to curb waste because “token incinerators” create little value.