Meta Caps Employee AI Token Use as Costs Approach Billions

A leaderboard inside one of the biggest tech companies in the world quietly turned an AI rollout into an expense problem. According to The Information, Meta sent an internal memo to about 6,000 employees saying it would impose centralized controls on token usage, just weeks after pushing the same staff to use AI tools more aggressively in their work.

The numbers are the headline. Meta employees reportedly consumed 73.7 trillion tokens in roughly 30 days, a figure tracked on an internal leaderboard called Claudeonomics, a nod to Anthropic's Claude. The Decoder reports that staff competed for badges named Bronze, Silver, Gold, Platinum and Jade, plus titles like Session Immortal and Token Legend, with some employees instructing AI agents to run tasks in parallel mainly to inflate their numbers. The memo flagged an "exponential increase" in usage and warned the company is on track for billions in internal AI costs in 2026.

CTO Andrew Bosworth's follow-up memo was the part worth quoting. "Nobody should be using AI tools just for the sake of using them," he wrote. "All motion is not progress and token usage alone is not a measure of impact of any kind." To enforce that, Meta's product and engineering teams have built a central dashboard called AI Gateway that watches usage and spending in real time, with maximum quotas per team rolling in during 2026 and stricter budget allocations and tooling due by 2027. The company is also steering staff toward its own coding assistant, MetaCode, and away from external tools like Claude.

The honest caveat is that the reporting doesn't break out what fraction of that 73.7 trillion tokens was genuine engineering work versus gamed metrics, and it doesn't disclose Meta's contract terms with Anthropic, which is exactly where the billions number would actually live. Take the cost figure as reported, not settled.

What the story is really useful for is every other company quietly subsidizing its own employees' enthusiastic AI experimentation. If Meta is the one putting in gateways, real-time tracking and per-team caps, smaller teams without that telemetry probably have less visibility into their own AI bill than they think, and the leaderboard playbook is the cautionary tale to skip rather than copy.