I've been reading Boom by Byrne Hobart and Tobias Huber (Ben Thompson did a long interview with Hobart on Stratechery (if you want the audio version of the argument) and it reframed how I think about the current AI spending wave. The book splits bubbles into two types:
Mean-reversion bubbles money piles into something that already exists, prices detach from reality, crash, nothing left behind. Housing 2008. Tulips. The crater kind.
Inflection bubbles money piles into something that bets the world works differently going forward. Amazon wasn't a better bookstore. It was a categorically new thing. The investors looked insane by the standards of 1997. They were right about 2010.
The dot-com crash is the cleanest example of an inflection bubble working as intended. Telecom companies borrowed insane amounts and laid fiber optic cable nobody needed. Then they went bankrupt. But the cable stayed. And because bankrupt companies built it, the internet was essentially free. The bubble funded the future and then got out of the way.
So here's the actual question about AI:
Google, Amazon, Microsoft, and Meta are on track to spend close to $700 billion on AI infrastructure in 2026 nearly double last year. That gap between what's being spent and what's being earned is real and large.
But Hobart and Huber's deeper argument is that stagnation is more dangerous than a bubble. Progress has been quietly slowing since the 70s breakthroughs are rarer, more expensive, harder. Bubbles are sometimes the only force strong enough to override the collective risk aversion that stops necessary things from being built. The honest question isn't whether AI is a bubble. It probably is. The question is which type.
Does AI produce something categorically new or is it a faster, more expensive version of software we already had? If it's the former, the infrastructure survives the crash and becomes the foundation for whatever comes next, the way fiber became the internet. If it's the latter, we get the crater.
History only tells you which kind it was after the fact.
What do you think inflection or mean-reversion? And what would actually convince you either way?
[link] [comments]