All good, as long as there is self-awareness that it's not built on a "need", but on a "want" basis.
It should be obvious to the reader that this is very much overkill, even for the stated goals of expandability and learning.
It should be obvious to the reader that this is very much overkill, even for the stated goals of expandability and learning.
Incidentally, this setup does leave me some significant headroom in terms of compute resources, but that's by design.
If it's inference, don't expect great performance or cost-effectiveness.
But if you learn a lot during it, I wish you all the best!