Sonatype uses the Amazon Web Services Elastic Compute Cloud, better known as EC2, for all our training lab machines. Most of our students use a training workstation for 3-4 hours per class, and since we often have bursts of custom training activity it just wouldn't make sense for us to own real, physical hardware to support training. If we were maintaining our own hardware, we would have to either maintain enough capacity to teach several classes simultaneously, or we would have to limit the number of classes we deliver. For Sonatype's training effort, instantiating machines as they are needed is the right thing to do and is exactly what the cloud is designed for.
Sonatype has leveraged AWS's great set of command line APIs and written scripts on top of them. We just invoke "instantiate-lab-machines-small.sh 10" and poof, we have a set of pristine machines for the students, all running VNC as a service (for remote desktop access) and pre-setup with the latest version of Maven, Eclipse, and the m2eclipse plugin. No teardown. No hardware maintenance. No hassle. We even have a subsequent script that lists out the machines hostnames for each copy-and-paste into the email that goes out to the students.
While our overhead for training infrastructure is smaller than it has ever been, our use of AWS is about much more than just the reduction in cost associated with cloud-based hardware, we're much more agile because it takes us 10 minutes to create infrastructure. I don't even want to imagine owning these lab machines. As our training offerings expand, we'll be using more and more of these lab machines. I can foresee a day when we'll literally have a set of AWS machines up around the clock to serve our training offerings. Until then, it's pay by the hour -- the raison d'etre for cloud based offerings.