1. Mainframes are cheaper then distributed computing. We call them 'the cloud' these days.
If pointless things like solar windows become cheaper then utility solar, then utilities will just build farms of solar windows. You know what's easier then installing a solar window in a house wall? Installing one at ground level, without worrying about walls.
Instead of having a handyman drive for 50 minutes, and then spend two hours tearing apart my house to install 6 windows in it, they could have him install an array of 90 of them in one day at a utility site. Hell, he won't even need to make sure that they don't leak.
This was the argument against PCs too. And in the late 70s I was already using cloud computing (MIT lispms and PARC systems like grapevine) that were not using mainframes. The economics seemed clear, but a variety of path dependence and economy of scale issues meant the mainframe survived quite a while (and is still going strong, but isn't dominant). But we are doing more edge computing than ever before with supercomputers in every pocket, game consoles and the like.
I have also worked in distributed solar. That's why I understand that the existing utility model still has a long way to run. Where I disagree is that there might be an inherent advantage.
And mainframes will always be cheaper than distributed computing.
I believe the economics favor the utility approach now and for the immediate future but would be reluctant to put my money on always/never