You are generally right. The data is a big difference. Managing it, it's too big for git etc is a massive problem.

ML also has a distinction between the system for training a model and the system that uses it, but that's just two projects to manage