The misinformation and assumptions on this thread are funny as many are incorrect. Azure, or 'The Cloud' will use up to 11 severs to render the structure of the city - which is large and destructable. The your local computer will decompress , like it does when you watch a movie off Youtube or Netflix, and apply your prescpective and other details. It doesn't use that much bandwith at 2-4 down and 1 up.
Here is a demo from a few years ago from what Microsoft and Nvidia were working on together.