What's next in high-performance computing?

We can trace the evolution from Crays, to clusters, to supercomputing centers. But where does it go from here?