One missing factor in the article is the source of energy. Free energy can be easily obtained from renewable sources like wind farms since as a result of wind variance only some fraction of generated power can be practically utilized. Variance is very different around the world making it difficult to centralize.
Though miners would then work only for 20% of their time making it feasible once electricity will be about 5 times more expensive than equipment.
Therefore the sooner we will reach state of the art equipment, the better. And for simple SHA it is much easier to reach the limits imposed by physics than for CPU.
I mean it is free for farm owners, since as of today they have nothing better to do with extra energy that is produced in good conditions. So it is basically wasted.
As of today that is true. But once we hit 10nm node the progress in terms of cost per transistor will probably cease dramatically. And it already slowed considerably.
Compare for example nvidia 460 gtx (2010) and 2014 760 gtx (2014). For same $ you get roughly 40% more transistors. Cards cost $200 and tdp is 130W which gives $450 at $.1/kWh in 4 years. So once cards with 2 times more transistors hit the shelves you would already spend 5 times more on electricity than card itself.
Though miners would then work only for 20% of their time making it feasible once electricity will be about 5 times more expensive than equipment.
Therefore the sooner we will reach state of the art equipment, the better. And for simple SHA it is much easier to reach the limits imposed by physics than for CPU.