[krastanov]

I would suggest attempting to learn the general principles behind the one way to do ML/cloud/whatever you have studied. That type of breadth of knowledge is the useful one. It lets you gain further knowledge much more efficiently. As I mentioned above, to great extent breadth of knowledge is about learning how to be an efficient learner.

[heavenlyblue]

I would say the most important knowledge is the one that allows you to ask the questions of "why is it still not done?" or stating "well, they did it, but it won't work well in situation X" and then either finding the products that do exactly that, or knowing that in a few years there will be an industry providing exactly that kind of service.

Saying that you can't move between cloud providers because you need to learn something new probably implies you don't understand the fundamentals constraining the design of e.g. databases provided by cloud.