Trev is pushing it a bit with the 'a lot of 3rd party AAA games use custom engines' statement, but is clearly wrong about the 're-optimise/recode engines from scratch'. Actually, the 'from scratch' part is absolutely not true.
An anecdote: I'm currently porting an amd64 pet-project of mine to arm64, where I had gone overboard with x86 intrinsics - original code was really game-engine-level-optimised, but for generic x86*, meaning it does both SSE and AVX. Since I want to keep the same level of generic optimisation on arm64, I'm meticulously changing amd64 intrinsics into arm64 intrinsics (alternatively, I could be dropping intrinsics altogether and get the port done in a fraction of the time, but I don't want to do that). But the changes I'm doing for the arm64 port are exactly in those isolated places where I've used the intrinsics, and the translation of the intrinsics themselves is fairly automatic. Curiously enough, I even spotted the other day on the web a fellow developer who had developed an automatic sse->neon conversion header for a subset of sse intrinsics he needed, so he could basically continue writing in sse intrinsics and get neon code generated automagically. So what I'm saying is that for engine code, the porting effort to get something generically-optimised for amd64 to the same level of generic optimisation on arm64 is much smaller than what Trev makes it out to be. Last but not least, engine maintenance is much more compact and is done by much fewer people than the actual games using the engine.
To reiterate something which has been repeated ad nauseam over the past few days: as long as NX has the performance bracket, arm64 is not an issue.
* generic x86 as in not targeting a particular uarch.