Well it all started with usual SV style "growth hacking"(price dumping as a SaaS) of "gather users now, monetize later" by OpenAI - which works only if you attain virtual monopoly(basically dominance over segment of a market, with competition not really competing with you) over segment of the market.

The problem is no one attained that position, price expectations are set and it turns out that wishful thinking of reducing costs of running the models by orders of magnitude wasn't fruitful.

Is AI useful? of course.

Are the real costs of it justified? in most cases no.