Raw LLM Responses

I keep posting this question under videos like this but this far, as far as I know, the topic was never addressed: So we know that AI can come in 2 forms. 1:Local AI and 2:remote AI (think ChatGPT) Option 1 needs local computing power, local database and some crazy degree of security, potentially entirely separated from the internet. Option 2 is built to service the requests of millions of people across the globe. As such it needs massive computing power, massive databases in the form of server farms, and supposedly increased security (but we all know they do the bare minimum, with all the data leaks) However both of the option have something in common. They need power and cooling. If you deploy AI locally to everywhere, that will massively increase power consumption on the local infrastructure. Think back of the times when there were brownouts and blackouts simply because all the people were trying to turn on the AC at the same time. Think back also to the time when the charging of electric vehicles was forbidden to put less strain on the grid. My point is that scaling up AI to be everywhere is not feasible, unless companies pushing AI want to invest into replacing and massively upgrading the infrastructure everywhere. With remote AI, the building of infrastructure might be simpler, because they only need government to give them a plot of land where all the requirements can be built in one go, but that is also placing all the eggs in one basket situation. If anything fails or gets sabotaged there, everyone that was relying on the AI there is coming to a grinding halt. Either way, upscaling AI on a global scale is not feasible with the current infrastructure.