You’ve felt it, haven’t you? That quiet unease every time another company announces a multi-billion dollar AI investment. The numbers stopped making sense months ago.
The AI industry is currently spending roughly $200 billion a year. The revenue base? About $30 billion. That gap isn’t a rounding error — it’s a chasm. And everyone in the industry knows it. They’re just hoping you don’t do the math.
But here’s the thing that nobody’s telling you: the scariest outcome isn’t a crash. A crash would be clean. Painful, but clean. Markets would correct, capital would reallocate, and we’d all move on. No, the real nightmare is something else entirely.
The worst thing that can happen to a bubble isn’t that it bursts — it’s that it never does, and instead slowly bleeds everyone dry for a decade.
Think about it. The current justification for the spending gap is simple: AI is infrastructure. Like railroads in the 1860s, like telecom fiber in the 1990s. You build the tracks before you know exactly what will ride on them. Fair argument. Maybe even a correct one.
But railroads had a physical destination. Fiber optic cables carried the internet that actually arrived. The question hanging over every data center being built right now is brutally simple: what if the demand never shows up at the scale everyone assumed?
I’m not talking about AI failing. ChatGPT works. Claude works. The technology is real. But there’s a massive difference between ‘this technology works’ and ‘this technology generates $200 billion in annual revenue.’ That gap is where careers will be made and destroyed.
Consider what’s actually happening on the ground. Startups are building AI wrappers around API calls and calling it a business. Enterprises are running pilots that never reach production. The $30 billion in revenue? A significant chunk comes from other AI companies buying compute from each other. It’s a circular economy where everyone’s the customer and nobody’s the end user.
When the revenue of an industry is largely the industry selling to itself, you’re not looking at a market — you’re looking at a confidence game with extra steps.
Now, here’s where it gets genuinely interesting — and where most analysts get it wrong. They frame this as either ‘historic bubble about to pop’ or ‘visionary infrastructure buildout.’ Binary. Clean. Wrong.
The most likely outcome is the messy middle. AI doesn’t crash spectacularly. It just… underdelivers. Year after year. The models get incrementally better but never quite reach the transformative threshold everyone promised. Enterprise adoption crawls instead of leaps. The $200 billion in annual spend slowly shrinks to $150 billion, then $100 billion, as investors lose patience. Not a crash — a slow deflation.
And that’s the real danger. Because in a crash, people panic and act. In stagnation, they rationalize and wait. Companies keep their AI teams for another year. Investors hold their positions for another quarter. Founders pivot instead of shutting down. Everyone keeps telling themselves the breakthrough is just around the corner.
If you’re an investor, the implication is clear: stop asking ‘is AI real?’ and start asking ‘does the revenue timeline match the capital timeline?’ If you’re a founder, the question is whether your business model assumes a level of AI capability that may not arrive for five more years — or ever. If you’re an employee, you need to know whether your company’s AI strategy is a bet on transformation or just FOMO spending that will get cut the moment the board gets nervous.
The bubble isn’t irrational. It’s a rational bet on an irrational timeline. And that’s exactly what makes it so dangerous — because everyone involved is making a reasonable decision that collectively leads nowhere.
Every major technology transition had a period of irrational exuberance followed by a reckoning. The internet had it. Mobile had it. Cloud had it. AI will have it too. The only question is what shape the reckoning takes.
My bet? Not a bang. A whimper. A long, expensive, career-defining whimper. The kind that doesn’t make headlines but quietly reshapes industries while everyone’s still waiting for the explosion.
Prepare accordingly.
FAQ
Q: But isn't massive upfront spending normal for infrastructure buildouts like railroads or telecom?
A: Yes, but those had clear demand trajectories. Railroads connected physical cities that already existed. Fiber carried internet traffic that was already growing. AI infrastructure is being built for demand that is largely hypothetical and, in many cases, circular — AI companies selling to other AI companies. The analogy breaks down when the end-user revenue doesn't materialize at scale.
Q: So should I exit AI entirely if I'm an investor or founder?
A: No. The technology is real and useful. The practical implication is to scrutinize the gap between your revenue timeline and your capital burn timeline. If your business model requires AI capabilities that don't exist yet, you're making a timing bet, not a technology bet. Size your exposure accordingly.
Q: Isn't this just the same doom-and-gloom every skeptic has been saying since ChatGPT launched?
A: The contrarian take isn't that AI is overhyped — it's that the most likely outcome isn't a dramatic crash but a boring, expensive stagnation. Most doomers predict a pop. I'm predicting the opposite: no pop, just a slow bleed that's harder to recognize and slower to correct. That's actually worse.