In the past year we've literally seen a 100x increase in network speed. We wont be seeing anything like that ever again, but neither are we anywhere close to a ceiling at todays btc exchange rate.
Bitfury recently announced they secured funds to expand their mine to 100MW. With their current chips, that means a few 100PH. With the 0.2J/GH they claim for their 28nm chip by the end of this year (0.1J/GH next year), that means ~500PH and possibly ~1EH next year. Thats BF alone.
KnC have said they will gun for 20% of the network, up from <4%. In the past 24 hr they appear to have doubled their hashrate, so I guess thats starting. Then there is AM sitting on a huge pile of chips that will find their way to the network one way or another, Spondoolies 28nm, etc.
I posted this chart over a year ago:
and got mostly ridiculed both for the conclusions and the assumptions, which turned out to be pretty darn accurate (considering the >100x increase it predicted). But to predict the next few years, they are outdated given that efficiency will likely be 3-5x better than that, and price per GH will drop below $0.2/GH according to at least spondoolies. So lets see what happens then:
If you pick a conservative electricity cost $8c/KwH and double BF and SP most optimistic efficiency projections from 0.1J/GH to 0.2, you end up somewhere around 1700PH. If you expect bitcoin to rise in value, you can increase those numbers proportionally.