Fine. It’s a bubble.
I’ll give you that. The valuations are insane. OpenAI at $150 billion. Anthropic raising at $60 billion. Nvidia’s market cap swinging by hundreds of billions on a single earnings call. The infrastructure spend looks like the late ’90s fiber buildout. Journalists are writing “AI bubble” pieces with the same confidence they wrote “crypto bubble” pieces and “dot-com bubble” pieces. The pattern recognition is satisfying. The takes are easy.
Now tell me how it pops.
This is the question where everyone goes quiet. I’ve asked it dozens of times — to journalists, to investors, to the people most loudly declaring the bubble. The conversation always follows the same arc: confident declaration that we’re in a bubble, vigorous agreement with historical parallels, and then a long silence when asked for the specific mechanism by which this particular bubble deflates.
The only answer anyone ever produces is: “OpenAI is going to fail.”
OK. Let’s take that seriously. What does OpenAI failing actually look like?
The best historical analogy is Nokia. At peak, Nokia had $50 billion in revenue and 1.1 billion active users — 40% of the global mobile phone market. They saw the smartphone transition coming. They’d been making smartphones since the late ’90s. The failure was organizational: platform complacency on Symbian, a matrix structure where middle management could veto anything threatening their fiefdom, and finally the Elop decision to bet on Windows Phone instead of Android. The $7.2 billion Microsoft acquisition was the death certificate.
BlackBerry hit $20 billion in revenue. They had a genuine security moat that evaporated when Exchange ActiveSync made BES irrelevant. They thought they were a hardware company when they were actually a secure communications platform (and they are now). BBM had 100 million viral users and could have been WhatsApp — they held it hostage to drive hardware sales until it was too late.
Both companies failed. Both were massive. Both had the kind of revenue and user base that looked like an unassailable moat.
And mobile didn’t pop.
This is the part the bubble talkers consistently skip. Nokia’s failure didn’t unwind the smartphone market. It created a vacuum that Apple and Android filled. BlackBerry’s collapse didn’t prove that mobile enterprise was a bubble. It proved that BlackBerry was the wrong vehicle for a real transformation. The technology wave continued. The companies shuffled. The market grew.
So when someone says “the AI bubble will pop because OpenAI will fail,” what they’re actually describing is not a bubble popping. They’re describing a market correction within a growing technology category — which is what happened with Nokia, what happened with BlackBerry, what happened with Yahoo, what happened with MySpace, and what happened with dozens of dot-com companies that died while the internet continued to eat the world.
If OpenAI fails, the most likely mode is the Yahoo path: not a dramatic collapse but a slow fade into irrelevance through a thousand mediocre product extensions. ChatGPT becomes a utility everyone uses but nobody pays premium for. Enterprise goes to companies with better compliance stories. The consumer product goes ad-supported. Revenue grows but margins compress. The valuation becomes unjustifiable. They never die — they just stop mattering.
And Anthropic takes the enterprise layer. And Google continues to dominate science and engineering while growing their core businesses with AI. And the infrastructure keeps getting built. And the models keep getting better. And the “bubble” keeps expanding because it’s not a bubble — it’s a platform shift with a handful of overvalued companies riding it.
Bubbles have popping mechanisms.
The dot-com bubble popped because companies had no revenue and were valued on page views — when the capital markets tightened, the gap between valuation and revenue had no bridge. The housing bubble popped because the underlying assets were fraudulently rated — when defaults cascaded, the derivatives built on top unwound catastrophically. In both cases, the mechanism was specific: a structural gap between what was claimed and what existed, revealed by a trigger event.
What is the structural gap in AI? The models work. ChatGPT has hundreds of millions of users who come back every day because it’s useful. GitHub Copilot is writing real code in real companies. The enterprise contracts are real revenue, not eyeball-based projections. The infrastructure spend is building real capacity that will be used. You can argue the valuations are ahead of the revenue, and you’d be right. But “the valuations are ahead of the revenue” describes every high-growth technology company in history. That’s not a bubble. That’s a growth bet.
For the AI bubble to pop the way people seem to mean when they say “bubble,” you’d need one of:
The models stop improving. If scaling laws break and the next generation of models isn’t meaningfully better, the capital thesis collapses. This is the real risk — and it’s a technical question, not a market one.
A better paradigm emerges that makes current AI irrelevant. Not better models — a fundamentally different approach that obsoletes transformer-based systems the way smartphones obsoleted feature phones. Nobody has proposed a credible candidate for this.
Regulation kills the market. A coordinated global regulatory action that makes training or deploying frontier models illegal or uneconomic. Possible but requires an alignment of political will that doesn’t currently exist in any major economy.
None of these are what the bubble talkers are describing.
They’re describing what they feel based on a shallow understanding of mechanics (in my opinion, well this entire post is my opinion). They’re describing a feeling that the enthusiasm is excessive, which it might be, but enthusiasm is not a bubble mechanism. Overvaluation is not a bubble mechanism. A bubble mechanism is the specific structural failure that causes the gap between perceived and actual value to close violently.
When someone tells you AI is a bubble, ask them one question: how does it pop? If the answer is “OpenAI will fail,” remind them that Nokia failed and mobile didn’t pop. If the answer is “the valuations will correct,” tell them that’s called “the stock market,” not a bubble. If they don’t have an answer, what they’re actually telling you is that they have a feeling, and feelings don’t pop bubbles.
Only mechanisms do. And nobody has described one.