AI and Machine Learning are changing society. This is also true in science where these techniques have promise to change how things are done: making science more productive and yielding new discoveries. In this talk we will review some recent developments in applying Machine Learning to Science, particularly in US Department of Energy (DoE) science areas such as Earth Science, High Energy Particle Physics and Materials. This deep learning revolution has been enabled by improvements in computing, and further advancements will mean increasing computational demands. So this talk will also describe developments in Deep Learning at Supercomputing Scale including tools, techniques and hardware, particularly those deployed at NERSC, the production supercomputing center for the DoE Office of Science.