Although organofconti has done a good job of predicting short term difficulty changes, I think it's unrealistic to extrapolate the short term trends far into the future.

As this graph shows, there have been a few different regimes in the historical difficulty trends.

1. Only CPU mining (steady growth)
2. Initial GPU mining (sharp difficulty spike)
3. GPU mining common, FPGA mining available (steady growth)
4. Initial ASIC mining (sharp spike) - The current regime

At each new technological step, the hash rate per dollar spent increased by an order of magnitude. FPGAs don't count as a technological leap, because they only incrementally improved the hash rate per dollar of TCO.

Once ASICs become common, I expect the difficulty growth to level off to a steady rate, until the next major technological leap which increases hash rate by 10x per dollar (or more). ASICs 2.0 will probably not be that leap.

Based on that, I think it's quite reasonable to predict the following regimes:

5. ASIC mining common (steady growth)
6. Initial New Technology (sharp spike)

I think Regime 4 is likely to last through the summer. It's impossible to say how long Regime 5 will last, but I think first generation ASICs will remain profitable for the duration of Regime 5 and into the early stages of Regime 6, in the same way that GPUs from 2011 are still profitable.