Even Meta also wants 6,000 employees to use AI sparingly; usage doesn't equal effective output.

Meta is planning to spend hundreds of billions of dollars on data centers while simultaneously sending a memo to nearly 6,000 employees asking them to save on internal AI costs approaching billions of dollars.
(Previous: Tom Lee praises Ethereum Foundation: infrastructure is key to financial upgrade, EF is playing a critical role in the policy market)
(Background: Japan announces 1 trillion yen: deploy 10 million AI robots across 18 industries by 2040 to solve labor shortage)

Table of Contents

Toggle

• "Claudeonomics" Leaderboard and 73.7 Trillion Tokens
• Dismantle the Leaderboard, Introduce Dashboards, Direct to Meta's Own MetaCode
• Usage Does Not Equal Impact

Meta plans to invest $600 billion in data centers by 2028, with AI infrastructure spending reaching up to $135 billion this year. But ironically, an internal memo from Meta warns nearly 6,000 employees that the company's AI usage costs are approaching tens of billions of dollars and asks everyone to "use it sparingly."

"Claudeonomics" Leaderboard and 73.7 Trillion Tokens

It is understood that over the past 30 days, Meta employees consumed 73.7 trillion tokens through internal tools. This number is fully recorded on an internal leaderboard called "Claudeonomics," where anyone can see which team or individual used the most, and the rankings even became a hot topic during internal coffee breaks.

The leaderboard was supposed to improve efficiency but turned into a competition game—the higher the usage, the more prestige, while actual output was rarely calculated.

A culture of "tokenmaxxing" also emerged among employees, who frantically boosted their usage to rank high on the leaderboard rather than actually doing better work. Managers also struggled to see how much each team actually spent or what output they gained, and the gap between usage and output grew wider. Moreover, most spending went to third-party services, especially Anthropic's Claude, which was the main source of this cost explosion.

Dismantle the Leaderboard, Introduce Dashboards, Direct to Meta's Own MetaCode

Faced with this situation, Meta's first step is to stop the bleeding: immediately dismantle the "Claudeonomics" leaderboard, so usage is no longer a trophy in internal competition, and remove the stage for comparison mentality.

In the coming weeks, the company will launch an "AI Gateway" dashboard to track each team's usage and costs in real time, automatically alerting managers when anomalies spike, allowing them to see where the money is going for the first time. By early 2027, Meta plans to implement formal token budgets and quota systems for each team, turning "use it sparingly" from a slogan into a data-backed rule.

More notably, on the strategic side: Meta plans to gradually guide employees toward its own coding assistant, MetaCode, replacing external tools like Claude. This move serves two purposes: saving API fees paid to third parties and allowing its own product to "eat its own dog food," being refined through internal real-world use.

Usage Does Not Equal Impact

Meta's CTO Andrew Bosworth also sent another memo around the same time, speaking more bluntly: "No one should use AI just for the sake of using AI." He further emphasized that all actions do not equal progress, and simply looking at token usage is not a measure of impact in any form.

The illusion created by the leaderboard—"more usage equals more contribution"—was even unacceptable to the company's top technical leadership.

This anxiety is not unique to Meta. Uber burned through its entire annual AI coding tool budget in just four months, and now caps each employee's spending on each tool at $1,500 per month. A KPMG survey shows that only 26% of companies have full visibility into their AI costs, with most realizing the situation is out of control only when the bill hits, often too late to reel it back.