The one area that few people in this particular debate topic focus on is the "left hand" of mining...
In mining, you need processing/calculating power, of course. But that's not the full equation. It's "how much power do we have" for "how much energy we put out." The only reason ASICs are superior to GPU or even CPU mining is because of their speed PER kilowatt-hour. You could theoretically have millions of CPUs still mining, but the power requirements make no financial sense.
The next step isn't about faster-than-ASIC, it's about less-power-than-ASIC.
I think instead of counting on faster and faster hashrates, you'll find more ASIC miners who switch to solar or wind energy for "free" energy.
To me, that will be where bitcoin mining finds its most profitable path -- no energy costs, high output.
This is the key to all high performance computing. Just look at the current top supercomputer ... 17 PetaFlops consuming 8MW.
Compare to human brain consuming 20 Watts. Estimates vary between 1 PetaFlop and 1000000 Petaflops to emulate (once we've got the software development cracked). So there is clearly some great scope for improvement, possibly using "WetWare" (real neurons genetically engineered), all the way to the "Computronium" beloved of the Singulantarians.
Not so sure about renewable/free energy though. Miners need to run 24*7 to stand any hope of ROI before they are outpaced by the difficulty spiral. Energy storage is expensive for intermittent renewable sources. Perhaps hydro or geothermal? Or for the space-cadets, put your rigs in orbit closer to the sun (latency may be a problem though, too many orphaned blocks).