The Token Tax

Meta employees burned through 73.7 trillion tokens in just 30 days — tracked on an internal leaderboard called "Claudeonomics." At Claude's current API rates, that's roughly $1.1 billion per month in token costs, putting Meta on track to spend billions on internal AI in 2026 alone.

AI pricing looks cheap at the start, but per-token billing can turn into massive enterprise costs as usage spreads across customers and employees. Companies pay for AI by the word — every word the AI reads and writes is called a "token," and you're charged for each one.

Here's the paradox: even though the price per token keeps dropping, total AI bills keep growing because everyone in the company (including agents, your digital employees) is using it more and more.

This paper shows why metered pricing becomes unsustainable at enterprise scale and makes the economic case for moving beyond rental models. Paying by the word doesn't make sense when AI becomes part of everything you do. It makes sense to "own" your AI (intelligence), not rent it.

—

A six-paper series on Private AI

Intelligence Unshared: The AI Sovereignty Papers

2 of 6: The Token Tax