Re: "(if there is a case where you can’t do this, I would be interested to hear it)"

Sure, cross-compilation exists; Macross being perhaps my favorite named compiler for such things (the authors, paying homage to the Japanese animation series which kicked off in the 1980s).

OpenBSD developers AFAIK, tend to eschew cross compilation (unlike, say NetBSD) because sometimes bugs are found on the architectures using their own toolchains that are not otherwise encountered and it is absolutely worth remediating such things.

It also used to be a thing, maybe forgotten, decades ago, that worthwhile systems were typically "self hosting" to varying degrees, insomuch as for example: if I build an operating system, I can rebuild it, from source, and every piece, on the same system.

Many things? Have lost that. As far as I know, the L4 family of microkernels, for example, can only be built from other operating systems' and are not self hosting. That, to me, is deeply distressing. The sort of thing I imagine would have received failing marks in classes decades ago.

These days? A lot of software "engineers" seem to for example, test only in Chrome, and if it runs there? That's good enough for them.

I have, at one point in my career, been tasked with assisting teams involved in so-called "Software Escrow" where a build environment is essentially recreated down to the hardware, and then placed into a vault. Undertaking such tasks is nontrivial, and costly even for those which undergo such processes, and many more development processes never even begin to try to embark in such realms.

Reproducible builds are still a rarity in our field; and a more worthwhile goal.

Having the privilege of a fast machine, to deploy onto slow machines? Should be acknowledged as a privilege. Helping those with slower machines, be self sufficient, is a more benevolent objective.