With Nvidia, AMD, and MS pushing, most new releases will probably have an ARM release, and older games can run with translation. The main problematic niche is older releases that are very CPU bound (like some sim games).
Games are the most optimised code there is.
Translation layers might cut it with vintage games but not with anything modern.
They JIT the x86 code to arm and their codegen is good enough to maintain most of native performance. Each new release brings them closer to closing the gap.
Where's your cutoff for vintage?
Portal 2 is demanding enough of a real-time FPS to demonstrate the principle and FEX is only becoming more optimized. It won't be long until it's running AAA games on low settings.
They seem to care a great deal about their hardware and gamer teenagers also seem to spend alot more time analysing hardware and milliseconds performance advantages than anyone else. They literally care about mouse latency in milliseconds.
They spend a huge amount of time on YouTube and online forums talking gaming hardware. If the word on the street is that ARM is milliseconds slower than Intel then they won't buy it. There's lots of huge YouTube channels that do nothing but measure game performance on various hardware.
Gaming is no longer dominated by affluent teenagers. At least not to the degree it once was.
The gaming community may care about numbers but I don't know if most actually know what those numbers say.
Oof.
A lot of the people who only need those things have largely already moved away from the PC desktop. (Especially if we don't include "laptop" in the "Intel will be hard to displace on the desktop" statement, which I'm not sure if the OP meant to include or not. But even if we include laptops, a lot of people have already moved away to tablet or non-Wintel-land.)
The remaining things, though? Generally not highly abstracted. There's the highly-tuned-for-amd64 performance side of things, and then the legacy/slow dev cycle side. I'm skeptical on the legacy side that there's a good business to be built on trying to move those users to a different architecture. Everybody's special weird setup is gonna be different enough with lots of random one-off external hardware or other such requirements I don't see good universal solutions that could scale.
It's not a growth business anymore but it's decline is slowing, I think.
It also moved towards Docker and Rust.
Meanwhile, much legacy software that hasn't made this transition often comes from an age of much lower performance and so can afford the penalties of emulation against modern hardware. Other legacy products are continually aging out and being retired for other reasons.
There are always exceptions, but Intel doesn't have the nearly the security it once did.