Uber's COO just admitted what a lot of companies are thinking but won't say out loud: they blew through their full-year AI budget by April, and now they're questioning whether the return justifies the cost.This is the Solow Paradox all over again. In the 1980s, economist Robert Solow famously quipped that "you can see the computer age everywhere but in the productivity statistics." Now we're seeing AI everywhere except in the bottom line.The problem is token economics. Companies thought AI would be like cloud computing - expensive upfront, but costs would drop as you scaled. Instead, they're finding that AI tokens are more expensive than they anticipated, and usage grows faster than efficiency improves. Uber isn't alone. Microsoft and other companies are reporting similar sticker shock.Everyone promised AI would cut costs and boost efficiency. The pitch decks were beautiful. The demos were compelling. Then companies actually deployed it at scale and discovered that running AI models for millions of users costs real money - sometimes more than just hiring people to do the work.The technology works. The models are genuinely impressive. But here's the uncomfortable question nobody wants to ask: what if AI is just... expensive? What if the marginal cost of intelligence doesn't drop to near-zero like software did? What if we're in the early stages of a technology that's powerful but never becomes cheap enough to justify replacing humans at scale?Uber's CFO questioning the ROI isn't a bug - it's a feature. It's what responsible financial management looks like when you stop believing the hype and start looking at the spreadsheets.We've been here before with every major technology shift. The internet era had the dot-com crash. Cloud computing had years of skepticism before the economics worked. AI might follow the same pattern - real value, but only after we figure out how to make the economics actually work. Right now, we're still in the "spending more than we're making" phase.