I’m being pithy, but the real answer is that it’s still a very frothy space and the specific tools you should choose is an extremely nuanced decision based on the specifics of your business problem, data characteristics, organizational processes, etc.. So if you’re at the point where ML really matters—it’s a committed investment for some production critical use-case—you should be looking to hire someone with hard won experience.

If you’re just getting started and looking for proof of concepts, you’re probably over-thinking it. Choose what is easiest to get up and running which means please do not adopt two cloud providers. Then if it goes well, don’t over-extend yourself on the prototype infrastructure and take a pause to evaluate the specific needs for productionization. That will either be a good jumping off point to dive-deep on a few specific ML ops solutions out of the literal billions of garbage products out there that aren’t worth learning about. Or you might hire someone to point you at the right things.