Meta's internal AI spending has spiraled to billions of dollars annually, forcing the company to abandon its freewheeling approach to AI token consumption. An internal memo to 6,000 employees signals a major operational shift starting in 2027.
The company plans to implement budgets, allocations, and a centralized dashboard called "AI Gateway" to govern how many tokens employees can consume. This represents a hard break from what insiders call "tokenmaxxing," the practice of maximizing AI model usage without cost constraints.
CTO Andrew Bosworth framed the shift bluntly in the memo: "All motion is not progress and token usage alone is not a measure of impact of any kind." The statement cuts to a core problem at scaling AI operations. Throwing compute at problems feels productive but wastes resources without generating proportional business value.
Meta's predicament mirrors a challenge facing every large organization deploying AI internally. Generative AI models consume expensive compute resources per interaction. Without governance, employees treating AI tools as free create runaway costs. A single complex query to a large language model can cost fractions of a cent, but thousands of employees running millions of queries daily adds up fast.
The "AI Gateway" system will function like an internal carbon accounting system for tokens. Teams will receive allocations. Departments that exhaust budgets face throttling or must justify additional spending. This creates visibility and accountability that didn't exist when AI access remained unlimited.
The move also signals maturity at Meta. Early AI adoption tends toward exuberance. Companies eventually confront the reality that compute scales linearly with usage but business impact often plateaus. Bosworth's memo suggests Meta leadership recognizes this inflection point.
Whether the token-rationing approach succeeds depends on implementation details. Too restrictive, and teams become innovation-averse. Too loose, and costs balloon again. Meta's challenge involves