You could have it easier.
http://thenodepole.com/2014/02/06/knc-miner-to-build-10-mw-datacenter-in-the-node-pole/10MW = 10.000.000W => 1W per GH/s => 10.000.000 GH/s = 10.000 TH/s = 10 PH/s
3TH/s*3Batches*1200units=10.800 TH/s = 10.8 PH/s.
Those 800TH/s (~260 Neptunes) are whether canceled or they have somewhere else (current location?) ~125*4 boards (28nm) available.
Enough to cover all 3 batches I'd say.
And that's maybe the reason why the hosting is limited to 3TH/s even with the announcement they made 11% improvement on the chip-design.
I don't think the 10MW will do it, per my calculations a while back:
--snip--
If it were (not IS) 1kw per nov jup, then they could power probably 9500 Nov Jupiters (ISH, factoring in other equipment and losses etc).
9500x650GH = ~ (< engineering symbol for "roughly") 6000TH
KNC have sold 3600x3TH machines, which is 10800TH
Which is why I think they need another building.
They will also want to have some cloud hashing power for themselves/shite-minr to sell, so they might actually need one even larger building or even 2 additional buildings to cover it all.
*edit:
Can remember that there was some info about another 2 locations they will build as soon as this one has been finished.
This?
http://www.globalsecuritymag.com/KnC-Miner-to-build-10-MW,20140206,42773.html"KnC are also already in discussions with local authorities regarding the establishment of even larger facilities in the local area already later this spring, which would then bring even more business to the local community."