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

Meta is shifting from encouraging heavy AI usage to actively controlling it. The company sent a memo to approximately 6,000 employees announcing new restrictions on internal AI token consumption.

According to The Information, the move reflects growing concern over runaway costs. Internal AI usage is projected to cost Meta billions of dollars in 2026. Starting in 2027, the company will implement formal governance: department budgets, individual usage limits, and a centralized AI Gateway dashboard for real-time spending visibility.

---

From “AI-Driven Impact” to Token Minimization

The policy marks a sharp reversal in Meta’s internal AI culture. Just months earlier, in November 2025, the company declared “AI-driven impact” a core performance expectation for employees. In spring 2026, it even ran an internal leaderboard called Claudeonomics, ranking the top 250 employees by the volume of tokens they consumed using Anthropic’s Claude.

The results were predictable — and expensive. Employees began spinning up parallel AI agents purely to inflate their rankings. In just 30 days, total token usage across the company jumped from **60.2 trillion to 73.7 trillion**. Meta quickly removed the leaderboard.

The company has now learned Goodhart’s Law the hard way: “When a measure becomes a target, it ceases to be a good measure.”

---

New Rules and Internal Alternatives

Going forward, Meta will push engineers toward its own coding assistant, MetaCode, and away from external models such as Claude. The new AI Gateway system will give managers visibility into who is spending what and allow the company to enforce hard limits.

The memo reportedly frames the changes as necessary for “sustainable” AI adoption rather than a retreat from the technology. However, the shift from celebrating maximum usage to imposing budgets highlights the tension between Meta’s famous “move fast” culture and the reality of spiraling inference costs.

Also read:

• Second Circuit Upholds Sam Bankman-Fried’s 25-Year Sentence as Elon Musk Becomes World’s First Trillionaire
• iFixit Teardown Confirms: Trump Mobile T1 Is Just a Gold-Painted HTC U24 Pro
• ECB Moves to Rein In Revolut’s “Self-Guided Missiles” in Europe
• Meta Smart Glasses App Contains Fully Built — But Currently Dormant — On-Device Facial Recognition System

---

Why This Matters

Large language model inference remains extremely expensive at scale. Even a company as wealthy as Meta is feeling the pressure as thousands of employees integrate AI into daily workflows. Other big tech firms are likely watching closely — many have similar internal AI enthusiasm programs that could soon face the same cost reckoning.

Meta’s experiment shows both the power and the pitfalls of turning AI usage into a performance metric. After watching token consumption explode under gamification, the company is now trying to thread the needle: keep innovation high while keeping costs under control.

The era of unrestricted “use as much AI as you want” inside Meta appears to be ending. From 2027 onward, every prompt may come with a price tag — and a watchful eye from the AI Gateway.