I have not been able to workout if AI means demand for compute for most people will mean it's infinite since you can have AI to just work on many different problems and many different possible solutions in parallel. Or if for most people (developers included) once AI is good enough for most tasks people will reach a limit of things they could want to do and find some limit of compute enough and if so where would this limit be.

I am talking about average users and developers, there will always be tasks that want as much compute as possible I suppose but these will be limited and cash constrained as they where in the past, I can imagine most developers being satisfied with finite compute and not wanting any more for their AI agents and not needing any more intelligence for their LLM's so no more training apart from distilling and making models more efficent.

The thing I don't know is if where AI is basically good enough for largely all possible things people could want to do, if the lack of demand for training coupled with the satiated desire for compute will mean hardware falls massively in price as well as local AI getting better and cheaper to run. It feels as if some of these data centers are not going to find have the issue of massive over supply.