A bubble is a poor metaphor for AI because there is obvious real value in it; so when it pops it won't pop all the way, but investment is maybe getting too far ahead too quickly and there will be pull-backs. I don't know a better metaphor - train getting beyond tracks? leaning over the skis? nothing really works.
But it's not like the Web3 bubble where the underlying value was always suspect.
Just because guys like Bezos got a couple things right which translated to big $$ does not imply they'll be right ever again. AI is 100% not a bubble, if you think this way, you have not used it. What we have are amazing tools which will only improve over time.
Well, the tools may not be amazing for enough people to be worth the current price or may have arrived too early. At a certain price, everything can be a bubble.
2000 was still a bubble, even if the demise of tech was completely off.
LLMs are very powerful and machine learning will produce great things.
But the resource demands needed to generate current LLM quality are challenging and I do not see an efficient way forward.
This reminds me of the hydronic mining during the California gold rush. Not all that efficient.
It doesn't matter how useful the tools are if the companies providing them don't make enough money to justify their investment. Hundreds of billions of dollars a year are being spent building datacenters which produce only tens of billions of dollars in revenue, and those datacenters are depreciating assets. This cannot continue indefinitely. Is it reasonable to expect the AI industry to be produce revenue at the rate of a trillion dollars a year or so, within the next five years, as it will have to do to avoid a crash? I very much doubt it.