Meta’s Adam Mosseri suggested that limits on employees’ AI token spending could become necessary within the next year or two. Speaking on Lenny’s Podcast, the Instagram head said that a strong engineer’s token burn rate might eventually match their full cost of employment, making spending caps a likely requirement.
AI token costs have become a pressing issue across the tech industry in recent weeks. Meta reportedly shut down an internal leaderboard tracking employee AI token usage after costs put the company on track to spend billions of dollars in 2026. Mosseri is not alone in flagging the issue; Uber reportedly exceeded its full-year AI coding budget by April, while Microsoft reportedly discontinued Claude Code licenses in favor of consolidating engineers around its own Copilot CLI tool.
Mosseri compared token budgets to other constrained resources like compute capacity, labeling budgets, and payroll, saying caps would need to scale with how much a company trusts an employee to use tokens in an ROI-positive way. He noted Meta currently has no token caps in place, but expects future pricing competition among AI model makers to eventually drive costs down. For now, he said, curbing wasteful practices like the leaderboard has helped rein in spending.
