> its calculations were "embarrassingly parallel," with no communication between nodes

That's probably the only type of process that would work in the cloud. Most HPC applications require lots of communication between nodes, so I don't think I would call this a proper supercomputer.