ML cares about computation times a bit, but very little about locality/decoupling, the stuff which makes it easy to scale.