It’s more than ‘where it runs a lot’ - it’s also where someone can make money by it being more efficient.
So for instance, JavaScript interpreter in a browser is meh unless the person doing the work is hired by someone who needs their SPA to run faster.
So a mega corp looking for someone with a lot of experience making their software more efficient (think a large operator like MS Azure , Google, AWS), especially if there are supply chain issues making it hard or expensive to scale up their existing hardware. Also high frequency trading where nano seconds matter.
Or older legacy systems (folks still using mainframes?) where the costs of switching to a new system in compliance and interoperability mean they’re trying to wring ever cycle out of the existing system.
I've been doing consulting for that kind of thing for running cost as fast as possible on "embedded" (raspberry pi-level) hw
I like the libraries that ship with toolchains for using this stuff. Anyone shipping silicon will have toolchain(s), either built in house or via consultancies.
Cutting cycles off integer divide is a waste of time for one application and a serious win if it's the integer divide the compiler injects into all applications.
Further from the machine you get language runtimes (e.g. libc, libm) then application libraries (blas, lapack, mpi) where it starts to merge with ML or HPC tooling.