💸 Is AI More Expensive Than Employees?#
Microsoft just canceled most of its Claude Code licenses for a surprising reason: it was too expensive. 😳
📰 The facts:
- Microsoft cancels Claude Code licenses (used by thousands of engineers) and moves to GitHub Copilot CLI
- Uber burned through its entire 2026 AI coding tools budget in just four months
- Bryan Catanzaro (Nvidia VP): “The cost of compute is far beyond the costs of the employees”
🤔 The token paradox: Individual tokens are getting cheaper, but AI agents consume exponentially more tokens per task. Goldman Sachs projects a 24-fold increase in token consumption by 2030.
📊 What Gartner says: By 2030, inference on large models will cost 90% less… but companies won’t notice it on their bills. Agentic consumption grows faster than unit prices fall.
⚠️ The uncomfortable conclusion: The promise of replacing or augmenting human labor with AI is proving more expensive than forecast. The productivity gains are real, but they come with an unexpected bill.
💡 Explanation in a nutshell#
Imagine hiring a highly efficient employee who works at record speed, but needs enormous amounts of electricity to operate. That electricity (AI “tokens”) is getting cheaper, but the employee uses it so fast that your bill goes up anyway. That’s exactly what’s happening with AI at large companies: more efficiency, but also more consumption — and the tab is a surprise.
More information at the link 👇
