People are often asking if (or confidently claiming that…) AI is a bubble. It seems hard to justify these massive investments -- trillions of dollars -- for a technology so unproven to deliver value.

I strongly suspect the answer is “No, AI is not a bubble.” I think of a bubble as a case where the fundamentals aren’t sound. But I believe in the power of AI… not necessarily today’s AI, but Real AI. And I think it’s about even odds that we will get to Real AI this decade, at which point it could take everyone’s jobs, in which case owners of AI companies would get fabulously rich, while everyone else would… something or another… that part’s still a bit unclear… Die? Live off handouts? Serve as status symbols for feudal overlords?

But anyways: Investing in AI is clearly a good bet if it might be your only ticket to having any money in a few years after making money from working stops being a thing. So the fundamentals seem really strong.

There is, of course, the risk that current approaches to AI are a dead-end and fundamental breakthroughs are needed. And then there’s the risk that AI kills everyone instead of just taking their jobs. In those cases the value of AI companies might collapse at some point, but I don’t think we can confidently rule out the “AI owners do great, everyone else gets screwed” future.

And again, speaking of capital-F Fundamentals: In principle, investments don’t need to pay out any time soon. Even if we don’t get Real AI for several decades, if the companies working on it today are still the ones doing it at that time, well… that’s still a huge payout!

I think most of the view of it as a bubble relies on wishful thinking. I also wish we didn’t have to worry that AI might be about to make humans obsolete. That sucks. I don’t even want AI to be able to write better songs than me (but if anyone knows how I can use AI to help make it easier to get the songs I’ve written out of my head and into proper audio “recordings”, please let me know)! I hope AI hits a wall, for so many reasons. And I hope we don’t let AI companies get away with stealing the entire cultural legacy of humanity; that we force them to get meaningful consent from people whose data they use; and that we set up a system that gives human creators enduring power over AI, and not just temporary compensation (also, if anyone knows a way I can use AI in a way that doesn’t feel a bit like stealing, please let me know)!

But even if AI isn’t a bubble, maybe it’s a house of cards. A delicate balancing act where everything needs to go just right or it will all come crashing down. There definitely seems to be a series of big bets being made and a feedback loop where loans are leading to growth financing loans leading to growth financing… you get the picture. Welcome to the modern economy, where “how much money exists” is largely a philosophical question. And even if everyone in this game believes in the fundamentals of AI, I expect that particular investments are being made with the expectation of a quick turn-around, and if that doesn’t pan out… the whole thing can come crashing down.

A bubble keeps inflating until it’s so large that the walls become too thin and it pops. A house of cards can just keep growing without limit if the people building it are skillful enough. A gust of wind could take it down, but really all it takes is finding the right card -- one low on the stack -- and pushing hard.

If AI really is a house of cards, and investments are dependent on things going to plan, then any major disruption to the plan could cause it to collapse, significantly slowing down the race to develop superintelligence. An investor gets cold feet. Datacenter projects get cancelled due to local resistance. A strike delays the shipment of critical components or resources. Regulations or legal challenges block an expected entry into a new market.

This wouldn’t stop the race, but it could help buy time and move resources away from the destructive and dangerous activities of frontier AI development and deployment and into more productive pursuits. I continue to believe in the fundamental power of Real AI, and sadly, I think AI companies have a good chance of delivering on it soon. But buying time could be really valuable. Every day, more people become more aware of the flaw in the fundamentals that makes AI a bad investment for humanity: all the money in the world won’t matter if we’re all dead.