Now, get this, when it comes to a LOCAL-LLM System that runs at Performances comparable to «Max/Pro/Enterprise/etc.» subscription-plans, you need a LOT of VRAM and GPU; if you're able to connect multiple high-end computer-systems/servers that power your Local-LLM infrastructure or architecture, sort of like what this guy did, although he says it's still slow...
https://www.reddit.com/media?url=https%3A%2F%2Fpreview.redd.it%2Fmy-little-decentralized-locallama-setup-216gb-vram-v0-o1o7ekxycp5g1.jpeg%3Fwidth%3D1080%26crop%3Dsmart%26auto%3Dwebp%26s%3Dd4d91f4d81048c0a080bd9973ad4ea17b6a249ea
(Full thread at https://www.reddit.com/r/LocalLLaMA/comments/1pg8ix9/my_little_decentralized_locallama_setup_216gb_vram/ for more details/comments)
...then you can work/interact with your own A.I. without needing to pay subscription-fees!

I saw how insane the prices are for VRAM or even just RAM in general and my auto-perception immediately told me that this is being done deliberately to «price-fix» and the there's this...

Sam Altman, I must ask, what in the world is he currently doing and thinking!? O_O
https://www.mooreslawisdead.com/post/sam-altman-s-dirty-dram-deal
Time-Stamp: 20251208T10:40Z

PewDiePie built his own $20,000 AI PC:

The data centers used for the public use essentially the same technology. That technology consumes a lot of power, and electricity becomes more and more expensive. Therefore the market demands better technology that consumes less power. Analog processors that work 1,000 times faster than GPUs and self-sufficient robots are right around the corner.

They can't prevent decentralized AI. It's already happening.