Heavy ML workloads make this more worthwhile since you get to design it to squeeze value out of every facet. For a basic web server and database it’s definitely overkill and something like a colocation makes much more sense