Fantastic bit of work (that's a euphemism - I'm certain it was a lot of work) from Christian Belady at Microsoft, who "borders" the discussion regarding data center growth both globally and in the US.
Christian tries to do a couple of things in the paper: project data center growth based on foundational work from Jonathon Koomey and supported by the federal government, and further to assess the true capital requirements that this growth will require.
I think Christian has really nailed down some trends that affect both parts of the equation.
The first is that an increasing portion of data center growth will come from the cloud computing sector, in what I term "utility scale" increments. These are the multi-MW facilities where developers and operators are deploying leading-edge efficiency techniques, yielding PUE ratings just above unity.
Because these data centers make use of free-cooling, and often obviate much of the power conditioning and back-up systems, they also cost less than the typical $15 million per MW of IT load that is the accepted rule of thumb for new development.
Some of these techniques and measures will be adopted in more traditional enterprise data center builds, though we could predict that capital costs won't dip down into the $5 million per MW territory (maybe with adoption of container/modular systems, but that market is pretty nascent as yet).
In the final growth and investment curves that Christian has proposed (and he does so to begin a conversation), there is a short-term flattening perhaps for the next five years, but beyond then the growth rate might nominally return to what Jonathon projected: a doubling of energy use in the data center market every five years.
While I have been and continue to be an advocate for promoting the efficiency message in this segment, it's also clear that the other half of the conversation - the environmental consequences of this growth - has to be part of our discussion. Where the power comes from, even for highly efficient centers, will be what we're talking about in five years, so why not start planning for that now?