ML methods which you can scale are often either by accident or only become scalable after simplifying the model significantly.