It honestly has a lot to do how many orders the various ASIC companies have and how quickly they can a) catch up on backlog, and b) fill new orders quickly.
My prediction goes well above 1000TH/s, but this takes into account:
--BFL catching up on all backlog (thousands of orders) and shipping in a timely manner (not this day-at-a-time nonsense).
--Avalon catching up on Batch 2 and 3, as well as disclosing any future plans.
--KnCMiner developing and shipping according to their schedule.
--BitFury (and other hardware developers) receive their chips (either Avalon or BFL), boards, and parts so they can complete their devices.
--ASICMiner continues to increase their deployed hardware (including sales) and discloses future plans as well.
--Other unknown or still in development projects / companies.
There's a lot of factors at play at this point..vastly more than any of us could have imagined just 12 months ago.
I've considered the "well above 1000 TH/s", but at some point there are lead times in play that even though well over 1000 TH/s have been sold, it takes quite a bit of time for them to come online. Interesting to note is that in one of my models I have a 50,000 TH/s by end of next year that is based on multiple chip vendors using the 22nm chip fab process. In many of my models the beyond 100k TH/s we start approaching saturation of the network and diminishing returns on profitability. Basically it would cost more to harvest a bitcoin when we start hitting the 100k TH/s plus than a return for the co-op miners.