Large technology companies acquiring power stations to fuel their AI. Should entrepreneurs care?

As I was looking through the latest issue of The Frontier newsletter, one of the headlines that caught my attention was Google acquiring a stake in a nuclear fleet. I just read the summary, so don't take it from me that I know all the details. However, the point seems to be clear enough: the firms with the largest models are starting to look at electricity the same way airlines look at fuel.

For a single founder working on top of an existing API, that might seem remote. Nevertheless, I would say that it deserves some consideration. The cost of compute sets the floor on which any AI product's pricing will sit. If energy becomes a bottleneck, the cheapest tokens might become more expensive than cheaper rather than the other way around.

Now what I am doing is looking at cost per action, not just the total cost per month. It is a simple spreadsheet, but it has already affected which features I will make available for free.

Does anyone do this in their cost modeling of AI? And has anyone actually built something that runs better on small and inexpensive models rather than the largest?