Put differently, you can happily compute away for a week on a few 100MB of data in ML. That stuff needs to be scaled, too!