Let me say the quiet part first.
Yes, it's a bubble.
I run an AI company. I've spent three years managing a portfolio through cycles. And I'm telling you the thing most founders are too invested — literally — to admit: the market has priced AI like it already won, when most of it hasn't even shipped a return yet.
That's not a doomer take. It's just what a bubble looks like from the inside.
What a bubble actually is
People hear "bubble" and think "fake." Wrong.
A bubble isn't a lie. A bubble is a truth that got too expensive.
The internet was real. It still 10x'd, then lost 80% of its value, then 100x'd again. Railways in 1840s Britain were real — they rewired the entire economy — and they still bankrupted a generation of investors on the way there. Fiber in 1999 was real. We're literally still using cable that got laid during the crash.
The technology being world-changing and the valuations being insane are not opposites. They're roommates. They show up to every mania together.
So when someone tells you "AI can't be a bubble because look how powerful it is" — that's not a counterargument. That's the setup for one.
The numbers that don't sleep at night
Here's what actually keeps me up.
The spend and the return aren't in the same room.
Hyperscalers are pouring hundreds of billions into data centers and GPUs on the promise of demand. Meanwhile, a widely-cited MIT study last year found that ~95% of enterprise generative-AI pilots produced no measurable P&L impact. Ninety-five percent. Read that twice.
The money is running in circles.
Nvidia invests in the labs. The labs use that money to buy Nvidia chips. Cloud providers sign compute deals with the same labs they're funding. It's an ecosystem financing its own demand. When the customer, the supplier, and the investor are the same three companies wearing different hats — that's not a market. That's a group chat.
The whole index is leaning on a handful of names.
Strip out the AI mega-caps and a lot of "the market is up" evaporates. Concentration like that isn't strength. It's a single point of failure with great PR.
None of this means the tech is fake. It means the price of believing in it got detached from the proof it's delivering.
That gap — belief minus proof — is the bubble. Always has been.
Why I'm still building in it
Now here's where I lose the doomers.
I'm not selling. I'm building faster.
Because if you actually understand bubbles instead of just fearing them, you learn the most important rule: bubbles overbuild the infrastructure the next decade runs on.
The 1999 fiber glut is why your broadband got cheap in 2005. The overbuilt railways became the network that made everything after them possible. The money that gets torched in a mania doesn't vanish — it gets converted into rails that survivors ride for free.
Right now the world is subsidizing GPUs, models, and compute at a loss so I can build on top of them cheaply. Founders who see a bubble and run are misreading the moment. The move isn't to avoid the bubble. It's to be the thing still standing when the price catches down to reality.
Amazon didn't win because it dodged the dot-com crash. It won because it survived it.
The only two questions that matter
Forget "is AI a bubble." That's a comment-section question. Here are the two that decide whether you make it out:
1. Am I selling the narrative, or building the fundamentals?
Companies priced on story die when the story cools. Companies priced on revenue, retention, and real usage just get cheaper to acquire customers. Build the boring version. The boring version survives.
2. Do I need the mania to be right, or do I win either way?
If your entire thesis needs valuations to keep going up forever, you don't have a business. You have a lottery ticket with a nice landing page.
I want to be the guy who's building something that works whether the market is euphoric or terrified. Because both are coming. Probably in that order.
The bottom line
AI is a bubble.
The tech is also the most important thing built in my lifetime.
Those sentences are best friends. Anyone telling you to pick one is selling you something — usually their own book.
So don't run. Don't cheerlead. Do the third thing almost nobody does:
Stay clear-eyed while everyone else picks a side, and build something that's still there when the tide goes out.
Bubbles don't punish the people who see them coming.
They punish the ones who forgot they were standing in one.