Power is the constraint.

Chips got faster. The grid did not.

Semiconductors improve on a schedule the industry has spent sixty years learning to compress. A better process arrives, a design shrinks, a fab retools. The cycle is measured in months to a few years, and the whole apparatus is built to shorten it.

Electrical infrastructure improves on a completely different schedule, governed by things software has no purchase on: permitting, right-of-way, environmental review, transformer lead times, the physical labour of stringing conductor across land somebody else owns. Nobody has found a way to compress that cycle, and it is not for lack of wanting to.

You cannot ship a gigawatt

This is the sentence the whole argument rests on. A chip is a small dense object; you can put it on a plane. Electricity at industrial scale is not a thing that can be moved to where it is wanted. It has to be generated near the load, or carried there over infrastructure that took years to build and is already carrying something else.

Two systems, then, moving at wildly different speeds, wired in series. When you put a fast process and a slow process in series, the slow one sets the rate. That is not an opinion about energy policy. It is what a bottleneck is.

And the bottleneck has moved from silicon to the grid.