What Would It Look Like If the AI Bubble Popped?

yawpitch

My guess is what it looks like is an enormous glut of HPC compute suddenly becoming available to science at virtually no cost at precisely the moment we’ve gutted the entire concept of science.

PeterStuer

It will not pop before the risk is sold to retail investors at a huge profit.

snackerblues

We may be finding out come Monday premarket

bko

The problem with these kinds of analysis is that they're surface level. They criticize "number go up" but rely on it to make comparisons to other bubble events. For instance, AI infra buildout is like 2008 because number went up in both. For instance: > So the speculative discourse only works as long as investors subsidize the use of the technology. When that subsidy stops, these AI firms have to actually deliver value, or customers won’t buy it. Are investors subsidizing the technology? There's upfront build out, but Anthropic is profitable and I believe the big labs are profitable on inference. In comparison consider the real estate bubble in 2008. Why was real estate going up so much? Was there a surge of people coming into the country driving up demand? Were people using these houses? No it was purely inflationary driven by cheap money and financial engineering. It also relies on arguments that this technology is all speculative, like one day we'll figure out how to use AI and demand will be high. But the demand is already there. Everybody is using AI, revenues are insane, it's the fastest growing product in history. It spans consumers and increasingly businesses. The comparison to dot-com crash is also superficial. Was dot-com a bubble? I don't know, if you were transported to peak 2000 hype, would you argue "you guys are in a bubble, and the internet impact on the economy will be no greater than that of the fax machine"? No, of course not. In this case the supply outstripped demand. The build out and hype was early. But with AI you're seeing real usage. It's not contagion. People are using this technology, not because it's cool and (apart from a few examples) because their bosses are forcing them to hit token metrics, but because it's actually useful and people are finding more uses every day. It's just lazy. There is real risk that the build out is too much. But to make that argument you would have to say that the model intelligence would asymptote or become increasingly expensive such that its not worth it. Or that demand for broad intelligence is capped somehow. But saying we're in a bubble just because number goes up is not a strong argument.

libertine

If it's just going to mean cheaper hardware, cheaper access to models and cheaper energy prices, it's going to be all right I think! I very much doubt the general public will accept a any sort of government bailout - people are completely done with the rug pulls, the only worrying sign is the new cult standard of people that just blindly accept whatever their leader states. But in a moment of crisis it might reactivate normal people back into voting. Cause if it doesn't, it might be the final nail on some democracies as we know. So yeah, either not such a big deal or the final nail for some decades.

b3ing

I think open source AI could cause it. I hope local AI takes over but this government will probably ban it.

k310

Don't overthink it. Government exists to protect the wealthy at the expense of the many. The wealthy will be bailed out. Wilhoit's law. “Conservatism consists of exactly one proposition, to wit: There must be in-groups whom the law protects but does not bind, alongside out-groups whom the law binds but does not protect.” We slobs will pay for the bailout. Big Brother needs these data centers for mass surveillance, either directly or by proxy e.g. Google et.al.) https://pylimitics.net/wilhoits-law/

nsonha

Did the last AI bubble even pop? Why would I expect this one, arguably more groundbreaking wave of changes to "pop"?