As I mentioned earlier, it uses a "rolling with additional batch" auto-scaling policy, which takes longer to "deploy", but ensures higher uptime and faster recovery if anything goes wrong with the "new" deployment. Here is the .ebextensions file in which you can change this behavior.