Meta Cracks Down on Employee AI Spending as Costs Head Toward Billions

Meta is instituting new controls on its internal artificial intelligence spending, a move prompted by concerns over escalating costs that are projected to reach billions of dollars in 2026. This decision represents a notable reversal from the company's prior stance, which had actively encouraged widespread AI tool adoption among its workforce. Previously, in November 2025, Meta had even designated "AI-driven impact" as a core performance metric for employees. In an effort to foster engagement, the company launched an internal leaderboard in spring 2026, dubbed "Claudeonomics," which ranked the top 250 employees based on their consumption of tokens using Anthropic's Claude. This initiative, however, had an unintended and costly consequence: employees began creating parallel AI agents solely to inflate their rankings. This led to a dramatic increase in total token usage across Meta, jumping from 60.2 trillion to 73.7 trillion in just 30 days, forcing the company to quickly remove the leaderboard. This experience highlighted "Goodhart's Law" for Meta, demonstrating that "When a measure becomes a target, it ceases to be a good measure." To address the burgeoning expenses, Meta has sent a memo to approximately 6,000 employees outlining the new restrictions. Looking ahead to 2027, the company plans to roll out a formal governance framework. This will include the establishment of departmental budgets for AI usage, the implementation of individual usage limits for employees, and the introduction of a centralized AI Gateway dashboard. This dashboard is intended to provide real-time visibility into AI spending, enabling more effective cost management. Furthermore, Meta will encourage engineers to utilize its proprietary coding assistant, MetaCode, as an alternative to external models like Claude. The new AI Gateway system will empower managers with insights into spending patterns and facilitate the enforcement of strict limits. While these changes are framed as necessary for "sustainable" AI adoption, they underscore the inherent tension between Meta's rapid innovation culture and the practical realities of managing spiraling inference costs.