Bubbles are a mechanism of innovation at times

The internet wiped out $5 trillion. Hundreds of companies raised fortunes on pitch decks with no path to profitability. Entire VC funds demolished. Nobody asked who would actually pay for any of it.

And yet.

The fiber laid by those bankrupt companies is what the internet runs on today. The server infrastructure, the protocols, the sheer volume of engineering talent that poured into figuring out how distributed systems at scale actually work. All of it survived the companies. Google, Amazon, and eBay were in that bubble too. They came out the other side because the technology actually worked. Everything built since runs on what the bubble paid for.

The bubble didn't prove the internet was fake. The bubble built the internet.

When a tech company folds, the technology doesn't disappear. The engineers move somewhere else and take everything they learned with them. The code gets acquired or open-sourced. The infrastructure gets absorbed. The cables stay in the ground. The company dies. The work doesn't.

Smart people watch the collapse and conclude the technology was oversold. That's the wrong lesson.

"Is this a bubble?" is almost always yes during a hype cycle. Early possibilities are legitimately exciting, returns get projected exponentially, capital floods in faster than anyone can deploy it sensibly, valuations detach from reality. None of that tells you whether the underlying technology works.

The question that matters is simpler: what's still standing when the companies fail?

If the answer is nothing, no infrastructure that survived, no protocol that got used, no behavior change that lasted, you're looking at pure speculation.

But when companies fail and the infrastructure remains, and other companies get built on that infrastructure, and those companies change how people live, that's a different story. The money was wasted, but not entirely. The wreckage funded the foundation.

I watch smart people make the same mistake repeatedly. Technology gets massively overhyped. Companies fail. The conclusion: the hype was wrong, therefore the technology was wrong, therefore their skepticism was correct all along.

They're conflating two things.

Being right that most companies in a hype cycle fail is not the same as being right about the underlying technology. You could have shorted every dot-com company in 1999, been right about all of them, and still completely missed that the internet was going to change everything. Correct on the individual bets. Wrong about what mattered.

The complaints I keep hearing about AI have the same structure. The models hallucinate. The products are unreliable. Companies are spending billions and struggling to show ROI. Valuations are disconnected from reality.

Some of that is probably true.

The compute infrastructure being built right now, the new chip architectures, the data centers, the models trained on basically everything humanity has written, none of that disappears when the hype settles. Companies that build in 2030 will take it for granted the same way we take fiber and cloud compute for granted.

They won't remember who went bankrupt building it.

The pattern is consistent enough that it almost feels like a rule. Real technology attracts speculative capital. The speculation builds the infrastructure the technology needs to actually work at scale. Most of the speculators lose. The technology wins anyway.

Whether AI follows the same arc isn't a debate about which companies survive the next funding crunch. It's a question about what's left when the first wave fails. What gets used? What gets built on? What do people actually pay for when the noise dies down?

Not the bubble. Not the failures. What's left.