Exactly the same conclusions I came to- working at NERSC/LBL some 15 years ago before leaving for Google and convincing google that HPC in the cloud was a big market (it took Urs and his folks too long to realize that, which is why I'm a fan of AWS).<p>The worst part is all that money spent on supercomputers benefits abotu 5-6 sophistcated users and the vast majority of scientists can't take advantage of them, but if the money had just been giving in the form of cloud credits, the scientists could easily do their work in the cloud.