The software will not be randomly modified when you are least prepared to deal with productivity changes and you can avoid having to wait for GitHub to respond to outages (while with Azure you might have an SLA, or are simply running other parts of your key infrastructure there and thereby already have them as a point of failure in your workflow). I will also say that you might trust a generic computation and storage provider more than you trust a specific service to not be mining your data (for the common "I want to host it myself as I don't trust you to not steal my stuff" issue).