In ML, "scalablity" is usually achieved through better algorithms which stream data/use approximations, such that resource usage is bounded