The idea is that your services run on remote systems without regard for what those systems are (as a VM, docker image, etc.) Your architecture is decoupled from theirs - you can run on an Amazon host one week, and a server in your closet the next.

And as a bonus, systems hosted this way are often harder to scrape as they’re all structured differently. Additionally, you can (and should!) take additional measures to protect your data from your provider - something that just can’t be done when the provider controls the data architecture.