I suspect the cutting down in RAM and now GPU production we've been seeing is not to make gaming more expensive. If anything they wish to make gaming cheaper to sate the masses. No, to me they want to make running local LLMs cost prohibitive. If they can siphon all LLM usage to controlled centralized entities, the deeply personal data they extract can be identified to you with certainty.

If you talk to an LLM in even a short amount of time with no mind to the privacy implications, it's not a stretch to say that these models can accurately predict your thoughts, feelings, tastes, manner of speech etc. They can know you better than your own mother or spouse.

Private LLM usage is their enemy. I would easily expect for them to crack down on it from all fronts.