Meta AI token usage inside the company has reached a jaw-dropping milestone: employees burned through 73.7 trillion tokens in roughly 30 days, a surge that reportedly costs Meta around $221 million a month, or nearly $2.65 billion a year, according to an internal memo first reported by The Information.
The explosive Meta AI token usage figure didn’t happen by accident. It was fueled by an internal leaderboard nicknamed “Claudeonomics” — a nod to Anthropic’s Claude, one of the most widely used AI tools among Meta’s roughly 6,000-employee test group. The leaderboard ranked employees and teams by token consumption, inadvertently incentivizing usage volume over productive output.
How Meta AI Token Usage Spiraled Out of Control
The roots of this crisis trace back to November, when Meta made demonstrating “AI-driven work results” a core performance expectation for 2026, dangling bonuses for top performers. That policy backfired almost immediately.
Instead of applying AI tools selectively where they added real value, many employees began chasing leaderboard rank instead of results. One internal dataset showed consumption rising from 60.2 trillion tokens in a 30-day window to 73.7 trillion before the company pulled the ranking altogether. Employees reportedly began “tokenmaxxing” — deliberately running multiple simultaneous AI tasks purely to climb the rankings rather than to do meaningful work.
Meta’s CTO Pushes Back on Runaway AI Spending
Meta Chief Technology Officer Andrew Bosworth didn’t hold back in an April memo obtained by The Information. “Nobody should be using AI tools just for the sake of using them. All motion is not progress and token usage alone is not a measure of impact of any kind,” he wrote, adding that tools should only be used when they genuinely help employees work better and faster.
Despite the warning, the numbers kept climbing — a clear sign that verbal guidance alone wasn’t enough to rein in the behavior driving Meta’s soaring AI token usage.
Meta’s Fix: Centralized Budgets and a New Tracking Dashboard
To get costs under control, Meta is rolling out a centralized monitoring system called “AI Gateway,” designed to track usage and spending across teams in real time, complete with automated alerts for unusual spikes. Full token budgets and formal allocations are expected to take effect in 2027.
The company is also nudging staff toward its own in-house coding assistant, MetaCode, in an effort to reduce reliance on third-party tools like Anthropic’s Claude — though outside tools reportedly remain available since Meta’s own models aren’t yet considered competitive at the frontier.
Meta Isn’t Alone in the AI Spending Crunch
This isn’t an isolated case. Rising Meta AI token usage mirrors a broader pattern among large tech companies racing to adopt AI at scale. Uber reportedly exhausted its entire 2026 AI budget by April after giving 5,000 engineers access to AI coding tools, and ServiceNow ran into a similar wall, prompting it to start tracking daily employee usage as well.
The tension is especially sharp at Meta, which has committed as much as $145 billion in capital expenditure this year, largely for AI infrastructure, even as it moves to clamp down on internal AI spending by its own workforce. For more background, see The Information’s original reporting and industry coverage from The Decoder.
What This Means Going Forward
The Meta AI token usage saga is a cautionary tale for any organization pushing rapid AI adoption without guardrails. When usage metrics become the goal instead of actual output, costs can spiral fast — and Meta’s $2.65 billion annualized bill is proof of just how fast.
As Meta builds out its AI Gateway dashboard and prepares formal 2027 budgets, expect other major tech employers to watch closely, since the balance between encouraging AI adoption and controlling runaway spending is quickly becoming one of the defining operational challenges of the AI era.
