Author falls into the same trap he talks about in the article. AI is not going away, we are not going back to the pre-AI world.
I can't predict when the shakeout will be, but I can predict that not every AI company is going to survive when it happens. The ones that do survive will be the ones that found a viable niche people are willing to pay for, just as the dot-com bubble bursting didn't kill Paypal, eBay, and so on. But there are definitely going to be some companies going bankrupt, that's pretty clear even at this point.
I'm juuust about old enough to remember the end of the Lisp Machine bubble (we had one or two at uni in the early 90s, and they were archaic by then). But obviously Lisp machines were the wrong way to go, even if they were a necessary step - obviously, hardware-mediated permanent object storage is the way forwards! POP! Ah, maybe not. Okay but can't you see we need to run all this on a massive transputer plane? POP! Oh. Okay how about this, we actually treat the microcode as the machine language, so the user-facing opcodes are like 256 bits long, and then we translate other instruction sets into that on the fly, like this - the Transmeta Crusoe! It's going to revolutionise everything! POP! Ah, what? Okay well how about...
And we're only up to the early 2000s.
It's bubbles, all the way back. Many of these things were indeed necessary steps - if only so We Learned Not To Do That Again - but ultimately are a footnote in history.
In 30 years' time people will have blog posts about how in the mid-2020s people had this thing where they used huge sheds full of graphics cards to run not-working-properly Boolean algebra to generate page after page after page of pictures of wonky-looking dogs and Santa Clauses, and we'll look at that with the same bemused nostalgia as we do with the line printer Snoopy calendars today.
The culture was nerdy, and the product promises were too abstract to make sense outside of Nerdania.
They were fundamentally different to the dot com bubble, which was hype-driven, back when "You can shop online!" was a novelty.
The current AI bubble is an interesting hybrid. The tech is wobbly research-grade, but it's been hyped by a cut-throat marketing engine aimed at very specific pain points - addictive social contact for younger proles, "auto-marketing team" for marketers, and "cut staffing and make more money" promises for management.
No, as it turns out.
But it was fun while it lasted.
You know what they say about when the taxi driver is giving you strong financial opinions.
The author is saying that those special purpose machines will age out quickly when the task of advanced computing shifts (again).
You seem to be making the assumption that "the huge neural networks they were designed for" are the only way to build AI. Things could shift under our feet again.
The author (and I) have seen too many people say that only to be proved very wrong shortly thereafter. This means that it doesn't quite have the logical force that one might think to assert that this time we have the approach right (ignore the previous 7 times somebody else said just the same thing).