Ah. I feared this.
To make a very long story short, some of our applications / deploy scripts have hard coded IP addresses that they expect. This is done because our legacy bare-metal infrastructure does not have functioning DNS during certain parts of our deployment. Machines that come immediately after these early stages look for a host at a specific IP. Once the 0th and 1st stage infrastructure has been set up, DNS and auto-discovery/auto-configuration starts to work.
Basically, there is one role in my farm that has a single instance (min/max set to 1, manual scaling). This instance lives in a specific subnet within a VPC. Instances that are launched after it will attempt to use *this* instance for NAT/Bootstrapping/Etc. They do this w/o the use of DNS (hard coded IP). The ability to set the IP manually for this specific instance is desired.
Thanks.
-K