If you are serious about building an operation at the scale you mentioned earlier you could certainly improve your power choices, you are taking a capital and 10-15% efficiency hit with what you choose.. If you are going to scale it up there are much better options...

For example I pulled together a build in a US data center for a customer last fall, about 150Th/s.. Coincidentally I was deploying within a week of one of Bitmain's US deployments (they have a few) and deployed the power for about 10K less in capital, and used 12% less power.. So much of a difference the data center execs came to me to help invalidate Bitmain's claims power wasn't being measured correctly.. Sure enough the PSUs Bitmain choose were poor..