Well fabs used to be hugely important, not only did each generation halve in linear size (4x in transistors per area), but each shrink was a big win on clock speed and power use. This revisions happened often, around 18 months.
These days the shrinks are smaller, i.e. 5nm -> 4nm -> 3nm, but each gen lasts longer, and provides very modest improvements in power and clock speed. They are also coming out in ever slower release cycles.
So now the competition has more time to catch up, and less of a disadvantage of they are a process behind. TSMC is currently leading, Apple, Nvidia, AMD, and others are bidding for the latest/greatest, while Samsung and Intel try to close the gap with their fabs.
Apple has an advantage of doing several generations in phones/tablets before bringing out the M1. Additionally they have an architecture license, so they do custom cores, not just what ARM is offering. This allowed them to tune their designs, use engineers from various companies they acquired to tune their chips, and get rid of the cruft, like 32 bit compatibility.
With all that said I expect Apples the perf/watt advantage to decrease over time. What does seem somewhat unique is they have build in a relatively small, power efficient, and inexpensive package (compared to similar functionality) 128, 256, 512, and 1024 bit wide memory interfaces. Sure you could build a dual socket Epyc with 16 dimms and likely burns north of 100s of watts and takes at least 1 rack unit, or you could buy a mbp m1 max. To match the M1 ultra you'd have to switch to some exotic CPUs that use HBM and sold by companies that typically send 3-6 sales people in suites before revealing their prices.
