AWS recently announced they will be moving to a per-second billing model for several services, most prominently their EC2 compute and EBS storage services. Previously they used a per-hour billing model, whereby if you launched an instance you were charged for a full hour minimum, even if you shut it down 10 minutes later. The changes will go into effect on October 2nd, and also includes the EMR and Batch services. For now only Linux instances are included.
This change eliminates the need for strategic planning of instance usage that some customers and applications were doing to ensure that each instance was used in complete, hourly blocks. As mentioned in the blog post, it moves EC2 and EBS in the direction of Lambda, in the sense that you now truly pay only for what you use. I think this will be big news and savings for those customers and applications using ephemeral instances to do on-demand or realtime compute processing, because it allows you to optimize for time rather than cost.