The previous escalation of hash power per new version of Bitmain miner would seem to indicate that 14 -> 75 Th/s is... an extremely big jump but not entirely outside of the realm of possibility. S5 -> S7 = 410% increase, S7 -> S9 = 300% increase, S9 -> "Bigfoot" = 500% increase.
So it doesn't seem completely impossible. Also, they're about due for a new model, based on previous release schedules.
Past history of the speed increases and model releases is irrelevant. It is all about the current cutting-edge node size and for miners that is going to remain at 14/16nm for at least another year if not longer.
In the case of the higher nodes Bitmain and the others were using fab technology that was already a few years old and starting to mature. When the jump to 14/16nm happened that was NOT the case. Bitmain and BitFury jumped in at the bleeding-edge stage where production processes are still being refined. 14/16nm is
still a work in progress with usable chips per-wafer yields getting better but still a long ways from what they get from the bigger and now fully mature node sizes.
Reasons why the current miners will be around for a fair bit more have been pointed out in numerous other threads but in short it boils down to
Foundry production capacity. Bitmain, Canaan and all other miner ASIC chip makers are bottom of the queue with AMD, Apple, various GPU makers, and Cisco having first call and the lions share of capacity at TSMC and GloFo. Same with Samsung but they also make chips for their internal usage so guess who is first in line with them...
Yes 10nm is coming on line for cell phone processors but as with the drop to 14/16nm node the primary design goal mainly targets mobile devices and that means low power chips. As in 1-2watts dissipation at most. As was proven with the current chips, yes dropping node size reduces power and size per-gate allowing the placement of more circuits on a chip. Great as far as that goes
provided you keep the chip total power requirements very minimal. The problem comes with Bitmain et al trying to pack as many cores as possible into the chip and pushing tech that was intended for low power mobile devices to now work in the realm of more than 10x higher power. The result is what we see with the Bitmain gear: Very high chip temps causing erratic performance and high failure rates. Dropping to a lower size - once it is viable - will have the same result unless they under-utilize the die space (don't pack to the max with cores) to keep power/temps within reason and that rather defeats the whole point.