The shift in strategy follows a period of unchecked experimentation that saw Meta on track to spend billions on AI by 2026. Mosseri, speaking on Lenny’s Podcast, noted that an engineer’s individual burn rate could soon rival their salary. To prevent waste, he envisions a future where access to these tools is tied to an engineer's ability to demonstrate an ROI-positive output. The company has already begun curbing inefficiencies, recently shuttering an internal leaderboard that tracked token usage, which Mosseri dismissed as a "token incinerator" that provided little actual value.
In section Startups & Technology
Meta eyes engineer-level AI token budgets to curb costs
Instagram head Adam Mosseri expects Meta to cap AI token consumption per engineer within two years, likening the processing costs of generative models to payroll or physical hardware. As internal spending on prompts and responses surges, the company is shifting toward managing AI resources with the same austerity as traditional operating expenses.
Meta is not alone in grappling with the financial reality of large-scale AI deployment. Uber exhausted its entire 2026 AI coding budget by April, while Microsoft recently canceled Claude Code licenses to consolidate its workforce around its proprietary Copilot CLI tool. While Meta currently imposes no hard caps on its staff, Mosseri anticipates that market competition will eventually drive down token prices as model providers compete for dominance. Until then, managing these costs remains a primary operational challenge, requiring the same rigorous deployment of capacity as allocating GPUs, storage, or headcount.
Comments (0)
No comments yet. Be the first!