Before anything: I believe that AI is currently a bubble. Despite it growing in use day after day, it still isn't a full replaceable tool for most tasks. It feels intelligent, but it is still clunky and unreliable for a lot of things. I'm not pretending to understand more than the others, but I have noticed a pattern that has been proving itself true for a while.
I believe that AI has reached to a point where it became too present to be a common bubble. The triggers for it are becoming more and more hard to achieve in the close future. The big companies are pumping money into eachother, but in the end they are actually *gambling* against eachother. They are putting money on the game, hoping to snatch away from the first one who falls on the race.
Take OpenAI as an example. Had they declared a poor financial state by 2023, the bubble would had bursted far more quickly, because at the time they were pretty much the only big AI company. If they lose relevance nowadays, we still have other companies on the game. There will be a moment of fear, but it won't last long. For example, Gemini isn't the only Google product (and I argue that it isn't even their biggest spending). GOGL would take a hit, but there still would have reason to buy their stocks.
On regard to the GPU producers, the main difference between the dot com bubble and this one is that now they have many other uses for it. Companies need computing power for LLMs, but they also need it for cloud storage and many other uses. Besides, every AI company is racing for AGI. When this happens, I seriously believe the bubble could be called down.
My main point on this all is that it *may* happen, but now a lot of things should happen in sequence to trigger the bubble. And even if it is the case, it wouldn't pop suddenly. People are aware enough of it, and the tech stocks are extremely sensible (just look at Intel this week), so basically the market is correcting itself enough to avoid an uncontrolled overvaluation. It's precisely the fear that is regulating the size of the bubble. No one is buying NVDA like there is no tomorrow, like they did with Cisco in 2001.
And even considering all that, and even if the bubble bursts suddenly, it's not science fiction to think that by 2040 and onwards AI will be thousands of times more powerful.