That is where you are wrong. Chip design can and is already automated. Algorithm generation is also already automated. Even program synthesis (programming) is automated.
If we have a prediction market are you telling me you are willing to bet against AI being able to automate the mining industry within the next 20 years?
And my point about chips is there is a physical limit beyond which no chip can become more efficient. There will be a final ASIC design and once that design is found there will be no need to continue to innovate there because physics cannot be improved beyond the optimal design. This will happen for sure in our lifetime. So once the final design is produced it becomes a matter of how many chips you have as the "new currency" of POW rather than how efficient of a chipset you have. Even blogging on Steemit is not future proof because algorithms can blog too and soon AI will be able to do it better than people so why would I think I can blog for much longer? I see the writing on the wall.
References
Van Berkel, S., Turi, D., Pruteanu, A., & Dulman, S. (2012, July). Automatic discovery of algorithms for multi-agent systems. In Proceedings of the 14th annual conference companion on Genetic and evolutionary computation (pp. 337-344). ACM.
RE: In Defense of Consortium Blockchains