A leaked memo says Meta is building a central AI Gateway to track employee AI use and spending in real time, as internal token consumption surged and 2026 costs are projected in the tens of billions.
Meta is building a central AI Gateway to monitor employee AI usage and spending in real time, set budgets and cap token consumption, according to a leaked memo sent to about 6,000 employees. The document says internal token use reached 73.7 trillion over a 30-day period before an internal leaderboard was shut down, and Meta now expects employee AI usage alone to cost tens of billions of dollars in 2026. The system would give teams a view of their AI usage, trigger alerts when spending spikes unusually and roll out structured token management that is expected to be fully implemented by 2027. Meta is also preparing to steer employees away from external AI coding tools and toward its own assistant, MetaCode, previously called Devmate, in the coming weeks. The shift highlights a broader corporate effort to curb so-called tokenmaxxing, as companies including Amazon, Microsoft and Uber grapple with whether fast-growing AI token consumption is translating into measurable productivity gains.