At least with AWS, the model is relatively familiar, which makes the failure modes predictable. This seems to be breaking new ground, and is likely to fail in New and Interesting ways.