They didn't take any of the AppGet code, just the ideas.
While they would be doing that, Azure will continue to lag behind (still does not even offer T4) on GPU offerings in comparison to AWS and GCP, because there's no compelling reason to run custom deep learning workloads on Azure. Azure is loosing $$$ millions of potential revenue to that. It would be a totally different story if .NET folks would get full TensorFlow and/or PyTorch in place of partial ports available now.