The machines at the very top are the only ones at the very top, yes. But there are dozens of manufacturers and fabs building useful ICs slightly shy of the bleeding edge. Lots of microcontrollers, general-purpose ICs, and special purpose ICs are still very frequently made on 22nm and 45nm scales.

And most of the hard trial-and-error discovery and experimentation has been done already, so it should not take 50 years to recover 50 years of historical progress. A process from 1970 can be done in the garage with 'just' a microscope and projector (and a lot of skill and hard work!): http://sam.zeloof.xyz/second-ic/

> We're infinitely more likely to lose the capability to make a modern CPU than lose the capability to know how to code in C.

I agree with this. I want to add that I think if the knowledge of modern CPUs is somehow lost, it won't be catastrophic, merely crippling, since there are literally tens of thousands of CS students every year learning how to build a CPU from electronic circuits.

We will revert to slow and bulky CPUs, be able to run C on them, and in due time rediscover and reëngineer miniaturisation, superscalar multithreading, etc.