if Apple wanted a much more powerful ARM based CPU there are already ARM-based chip that blow M1 completely out of the water, but use a ton of power doing so. Apple is not pushhing the limits of ARM, they seem to be more concerned with using little power and battery life.
...actually, one of the selling points of the other high-power ARM-based CPUs is that they
do use less power than Intel equivalents - which is important if you've got racks of the things sweating away in a large data center. However, they're also designed for the server market, and
especially areas where Linux is widely adopted - Linux/Unix has run on ARM since forever and there's a culture of processor-independence amongst Linux developers, so it's pretty well supported. Meanwhile, although ARM is getting popular in supercomputing, that often means that the ARM is basically a control processor for custom vector processors and suchlike, which would be pretty useless for general computing.
Remember that Apple Silicon isn't
just about the ARM instruction set - it's about GPUs, "neural engines", hardware codec acceleration etc. designed specifically for MacOS applications (plus ISTR there's some secret sauce in there that makes x86 code translated by Rosetta more efficient) and a chunk of their performance advantage comes from having on-package RAM and on-die GPU with very short connection distances. The ARM server chips from Amazon et. al. won't be any good for that. AFAIK, unlike most of those other chips, the M1 isn't even using an off-the-peg ARM core design.
Nobody knows how much of Apple's Mac income comes from the Mac Pro/Xeon end of the market - probably it is the MacBooks that are their cash cow, and making better Macbooks will be the focus of M1 development. The Mac Pro can be served by chaining together lots of M1/M2s in parallel - after all, the big difference between the $6000 and $25,000 Xeon Mac Pros is having lots of CPU and GPU cores which aren't
that different from the core-i ones (and are actually sometimes slower, per-core).
