- Meta employees used 73.7 trillion AI tokens in one month.
- That cost $221 million for that month alone.
- $2.65 billion annualized
- Gamified internal culture (“tokenmaxxing”) caused runaway usage.
- Meta shut down the leaderboard, imposed budgets, and forced employees to use internal tools
🧩 What “AI token spend” actually means
AI tokens are the metered units used by large language models. Meta employees were using external AI tools heavily for:
- coding assistance
- research
- internal automation
- agentic workflows (multi-step AI tasks)
Because these tools charge per token, runaway usage created a massive bill.
💸 The core numbers (confirmed across multiple sources)
- 73.7 trillion tokens consumed in ~30 days → $221 million monthly cost at enterprise rates → $2.65 billion annualized
- Some internal estimates showed 60 trillion tokens in a different 30‑day window → $900 million at certain vendor list prices (higher per‑token rates)
- Per‑employee cost estimated at $50,000 per year for AI tokens alone (for employees using external tools heavily)
🔥 Why spending exploded: “Tokenmaxxing”
Meta accidentally gamified AI usage.
The Claudeonomics leaderboard
An employee built an internal dashboard ranking coworkers by token consumption:
- Titles like “Token Legend” and “Cache Wizard”
- Employees ran huge AI tasks just to climb the leaderboard
- Some agents ran for hours with no meaningful output
- Usage surged from 60T → 73.7T tokens in weeks
- Meta shut the leaderboard down within days of press coverage
This created a culture where employees used AI for the sake of using AI, not for productivity.
🧱 Meta’s response: strict controls & internal tools
Meta leadership intervened aggressively:
1. AI Gateway (new centralized dashboard)
Tracks token usage and spending in real time. Full token budgets roll out in 2027.
2. Mandatory shift to MetaCode
Employees are now pushed to use Meta’s internal coding assistant instead of Claude, Gemini, or other external tools. AI tool usage is now part of performance reviews.
3. Spending caps
Meta implemented hard caps on external AI usage to prevent runaway costs.
4. Vendor restrictions
Google reportedly cut Meta off from Gemini due to excessive usage, forcing Meta to reduce consumption.
🏗️ The bigger picture: Meta’s AI infrastructure spending
The $221M/month token bill is only a small part of Meta’s total AI investment:
- $125B–$145B projected AI infrastructure spending
- $600B in data center commitments by 2028
- Massive GPU purchases and Llama model development
Meta is positioning AI as the core of its future strategy across Facebook, Instagram, WhatsApp, and Reality Labs.
