It would take an electric signal about 40ns to travel from the center to one corner and back, and a signal certainly needs to do more than that in a clock cycle, so the physical limit on its operational frequency would be well below 25 kHz. If you want comparable computing power to a regular M1, you'd need to build a tower of them at least 100,000 high. On the plus side, you could build a city on the stack and no one would need heating in the winter. The cluster should be powerful enough to coordinate the traffic lights. Synergy!

> The cluster should be powerful enough to coordinate the traffic lights.

This probably says more about how bloaty modern software development is, than the capabilities of your hypothetical macro-M1. You wouldn't need a cluster. Just one should do. The entire traffic light system of San Francisco ran on a few PDP-8 controllers, once. About as fast as the macro M1 at 25 kHz. (Even at 25 kHz, with parallel execution, it'd still be completing hundreds of thousands of 64-bit-data instructions per second, running circles around something like the original IBM PC or even the early Macintosh.)

With hardware so cheap and software stacks so deep these days, we often grossly overestimate the number of cycles really necessary to complete a task.

Well I assumed it would be an electron app because getting it done with a shedload of giant 555 timers wouldn't be as funny :)

40ns = 25MHz.

Oops. It's 40ms!

That's 25Hz, isn't it? Better stay of the eggnog until Christmas Day.