Comment by necovek

While I, for the most part, agree to this in principle, at the moment, general compute hardware production is relatively safe (or so it seems).

But when it comes to humongous costs, the fact that even "open source AI models" don't have their training data available (the actual "source") is one thing, but even if it was, it'd be impossible to retrain a model "at home". But if data was available, I am sure any of the existing free software foundations, or a new one, could rally users around sponsoring one DC.

We are back in the "mainframe" days where top-end compute is accessible only to few (with lots of money).