cloud SLAs are a joke. That's the point. If a cloud outage takes your application down, the appropriate parties to blame are those responsible for application design and operations architecture. You can't just take an app designed for a datacenter and naively move it to amazon and expect HA. You need to design for failure and automate the hell out of provisioning. The ideal is to either load balance across multiple providers or automate the living hell out of provisioning and backup until you can reliably get DR times in the 5 minute range.