I don't understand why most of the comment here are negative. On the bright side, being able to train a model in any sort of cloud (doesn't matter if it is Azure, Google, AWS, or IBM) and then directly create CoreML models, and iterate from model feedback, is a pretty special and unique position. It's a hybrid edge + cloud approach to running A.I. for end applications, and I do see it to benefit developers who want to roll their own models and create new applications particularly to computer vision, photography, and videos.