I've read a handful of his LLM articles completely, last year and this year. What I'm getting from his articles (I think) is this:

The USA stock market is largely driven by a dozen of FAANG companies whose valuation is rising fast and who are investing tens of billions in the LLMs. And all those companies are mono-dependent on the Nvidia cards, whose valuation is skyrocketing based on those expenditures. So point 1 - if LLMs would be as profitable as expected, the whole FAANG top-10 will slow down or even drop.

Point 2 he makes - a lot of the LLM-only companies are deeply in debts, acquire more debt, and continue kicking repayment can down the road, all at the same time. So assuming (again) that LLMs are not as profitable as expected, some LLM-only companies can go bust.

I think he wasn't making a statement that LLMs are intrinsically bad or useless, at least in those articles I've seen. He was just saying that financing aspect if very iffy AND that current world IT sector is way to overdependent on LLMs.