I am pretty sure that the current demand of gpu's can pretty much eat the left idle time issue at major datacenters because of the AI craze.
Not that its good or bad tho but we could probably have something more akin to spot instances of gpu being given for gaming purposes.
I do see a lot of company are having GPU access costs per second/instant shutdown/restart I suppose but overall I agree
My brother recently came for the holidays and I played ps5 for the first time on his mac connected to his room 70-100 kms away and honestly, the biggest factor of latency was how far the wifi connection (which was his phone's carrier) and overall, it was a good enough experience but I only played mortal kombat for a few minutes :)