TL;DR: Electrical power is a general purpose consumer good vs PoW mining equipment is a single purpose consumer good. Hence the mining equipment rent is the barrier to entry, given if you invest in power generation capital you could use the power for a different purpose.

It is true that electrical power is created by utilizing capital equipment, and the capital rent + labor of generating such power is the basis for the "Power Cost" component of the ideal miner competition profit equation.

But... electrical power is a general consumer good that can be used for many things, so investing in the capital to create it is not a very risky endeavor.

On the other hand, Bitcoin mining equipment capital is an EXTEREMELY specific kind of capital that only has exactly one use: efficiently/competitively mining a coin that has a particular PoW algorithm. Hence investing in bitcoin mining equipment is a more risky endeavor than power generation capital. Such a risk is a barrier to entry, and it is the barrier that is most considered when an entity considers mining Bitcoins.

Mature Arithmetic Logic Unit (ALU) bound PoW algorithms lacking new attacks (cryptographic definition) can only be out-dated by more efficient, more general purpose (less specific case proprietary) transistor fabrication technology.

Memory Latency bound PoW algorithms lacking new attacks (cryptographic definition) have the risk of being encumbered by all sorts of physical hardware patent inventions. This is because latency has significantly more room for such specific-to-PoW non-general purpose inventions... beyond additional patents relating to memory technology on top of ALU patents. Patents, I should point out, either cause the price of capital equipment to increase or enforce a monopoly on the capital... neither of which are desirable.

The capital maturity outlook of memory latency bound algorithms is also significantly worse than ALU bound... due to all of the expected future patent-able optimizations that could improve memory latency. Hence investing in memory latency bound mining equipment is even riskier because of the likeliness of a new patented optimization making your capital non-competitive, and given its specific nature, worthless.

This discussion brings me to a new insight. We have said that some places have "cheaper" power than others, due to the non-durable nature of electrical power. With the existence of Bitcoin, given other cost factors being less significant, Bitcoin causes all sources of power everywhere to be more equal in price at a particular time.

Now you might argue that memory latency bound PoW algorithms result in the mining capital component being the larger component than the electricity component being a good thing because: then mining would be less local to otherwise untapped (cheap) power sources. The problem with this is that as the mining capital matures (as all the optimizations are found, and the patents run out), we go strait back to the power cost being the largest component... and we had to suffer all the years of various entities unpredictably attaining a monopoly on mining in order to get there.