The emperor has no clothes, and it turns out the tailors are charging overtime.

Microsoft is quietly canceling most of its Claude Code licenses just six months after rolling them out to employees - a move that speaks volumes about the economics of AI in practice versus AI in PowerPoint presentations. The company that's betting its future on AI apparently can't justify the cost of its own AI coding tools.

Let that irony sink in.

This isn't just Microsoft. Uber's CTO reported burning through the company's entire 2026 AI coding tool budget in just four months. An Nvidia executive - at a company literally selling the GPUs that power AI - admitted "for my team, the cost of compute is far beyond the costs of the employees."

These aren't anti-AI Luddites. These are companies at the cutting edge of the technology, with every incentive to make it work. And they're discovering that at scale, the math doesn't add up.

Here's what the AI hype cycle promised: tools that would multiply developer productivity by 10x, eliminate grunt work, and let companies do more with fewer people. The ROI seemed obvious. Of course you'd pay for a tool that makes developers more productive.

But there was always a question nobody wanted to ask too loudly: what if the tools cost more than the productivity gains they generate?

Turns out that's not a hypothetical. It's the reality these companies are living with.

The problem is that AI inference isn't like traditional software licensing. Every query burns compute. Every autocomplete suggestion hits an API. Every code generation request racks up costs. And at the scale of thousands of developers making hundreds of requests per day, those costs multiply fast.

Traditional software has beautiful economics: build once, sell infinite copies at near-zero marginal cost. AI tools have the opposite profile: every use incurs real, ongoing compute costs. You can't just sell more seats to improve your margins - each additional user increases your infrastructure burden.

Microsoft's solution? Push employees from Claude Code to GitHub Copilot - which Microsoft owns and can therefore run at internal cost. That's not a vote of confidence in AI economics. That's vertical integration to make the unit economics survivable.

The implications here are bigger than just coding assistants. If AI tools can't pay for themselves in one of their most promising use cases - helping developers write code - where exactly is the business case compelling?

Customer service chatbots? Maybe, if you're replacing expensive call centers. But most chatbots are supplementing human support rather than replacing it, because customers get frustrated with AI limitations and demand human escalation.

Content creation? Sure, if you don't care much about quality or originality. But quality content still requires human oversight, editing, and strategic direction - which means you're paying for AI and people.

Data analysis? Again, the AI can generate insights, but someone needs to validate them, contextualize them, and make decisions based on them.

The pattern is clear: AI is great at augmentation but expensive at automation. It can make skilled workers more productive, but it can't yet replace them at a lower cost.

This doesn't mean AI is useless. It means the use cases that make economic sense are narrower than the industry wants to admit. And it means a lot of companies are about to discover that the AI tools they adopted enthusiastically are too expensive to scale.

We're heading into the hangover phase of the AI hype cycle. The demos were impressive. The potential is real. But the business case? That's still very much a work in progress.

The technology is impressive. The question is whether anyone can afford it.