Part 2: The token shock.

Yesterday we asked who is going to foot the $2 trillion bill for the AI build-out. Today, we have our answer: corporate IT budgets. And they are already breaking. Welcome to Part 2. Today, we are talking about token shock.

The unit economics of enterprise AI are quietly failing. Flat-rate subscriptions are being replaced by token-based billing, and the costs are spiraling out of control.

Look at Uber

Uber's CTO recently announced that the company burned through its entire 2026 AI budget in just four months. The culprit? A surge in the use of Anthropic's agentic AI tool, Claude Code.

By the end of the first quarter, 92 percent of Uber's 12,000 software engineers were using the tool. This shift to agentic workflows requires massive amounts of tokens. Engineers were generating API costs between $500 and $2,000 per person, per month.

Across the industry, developers are running multiple AI agents in parallel. In extreme cases, a single employee has been reported to spend over $150,000 a month on AI tokens. Even Anthropic quietly doubled its own cost estimates for average Claude Code users.

The Realization

This leads to a terrifying realization for the enterprise. Bryan Catanzaro, VP of Applied Deep Learning at Nvidia, recently admitted that the cost of compute is far beyond the costs of the employees. If the cost of AI tokens reaches parity with the human labor it is supposed to automate, the business case collapses. You aren't automating work anymore. You are just rerouting your payroll to hyperscalers.

Companies are blowing through their budgets. But are they actually transforming the business?

Tomorrow in Part 3, we dive into a report from MIT which reveals a 95 percent failure rate for custom enterprise AI implementations.

Question for the network

What are you seeing on the ground? Are token costs starting to alarm your finance teams?

By Michael Lennard Gnaedinger. © 2026 Gnaedinger Consultancy. All rights reserved.