TechCrunch just reported about a new project called Scalr which provides interesting high availability capabilities for Xen-based on-demand virtual datacenter Amazon EC2.
Basically Scalr monitors a virtual farm performance and when it drops under a customizable threshold, because an existing virtual machine crashes or because there is an increase in a certain application workload, it spawns a new VM, configuring it as part of the farm.
Despite the project is still in early stage it gives an idea of how corporate virtual datacenter will be in the coming years: autonomous infrastructures able to react to unplanned faults or increasing resources demand without manual intervention.