I noticed that with AWS (I opened a free account to do some experimentation). How robust are the instances? If I have one running and it shuts down, how much do I lose?
(I'll talk in terms of AWS, as it is my preferred provider.)
The quick answer is "everything that you would normally expect stored on a server's local disk."
These are ephemeral instances. When they crash, they get terminated and removed, along with all the data that was on their local disk. Your EC2 cluster realizes the need for a new server and starts spinning one up to replace the one it just terminated. This new server is based off your machine image (AMI), which is a point-in-time, clean configuration for the box. Savvy shops will build custom deployment managers that will grab the latest version of deployed software so that AMIs don't have to change constantly.
You have to make a paradigm shift about where you store data. Local things that you want saved must be moved to cloud-based storage solutions - either cloud DB, or cloud disk. That's why I mentioned S3 earlier.
I don't know how this forum works under the hood, but you may need to rejigger some of the code to properly work with multiple ephemeral instances and centralized storage - I don't know. A google search reveals that a couple of providers have already put SMF solutions on AWS - Bitnami, Turnkey, and Webuzo. Start there and see if it prices in the range y'all are willing to pay. Bear in mind that with autoscaling, you may have fewer servers running during non-peak hours depending on your min/max server counts set in your autoscaling group.