People don't usually buy this much compute for running a mail server or a media transcoder, so... probably AI? :)

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!