It then seems that X years down the road, mining will only be done by massive supercomputers. Or that seems to be the implication to me anyways. Could be wrong, but if the hashrate an difficulty go up like that, we should see ASICs pushed out as being unprofitable the same way that CPU mining has been pushed out, and the way that GPUs older than are not profitable now.
The question then seems to be about the initial cost of hardware? i.e. Are we assuming that the limiting factor will be electricity when the limiting factor in the future may be the cost of hardware?
Because of the scaling factor I don't think that HW cost will dominate. There will be such a huge market for mining gear that prices will drop. However: If a producer of top-notch mining gear can earn more money by mining than by selling his gear, the game could change. The bitcoin network would then become dominated by semiconductor-fabs. And knowing that such fabs are incredibly expensive to run this unfortunately means that bitcoin will be dominated by a handful of players (The end of all that's good about bitcoin).
One more thing: We didn't so far talk about the grey energy that needs to be spent before mining gear even starts to mine. That might at one point become important as life-cycles may be expected to be quite short for mining gear.