Google recommends 64GiB to build the Android kernel. That's a thing you could technically do at home. And if you want to do anything else at the same time, you're gonna need to go to 128.
Funny you should say that... my entire career has been built on embedded Android and I've built a lot of images from "scratch" (or as close as you get with random SoC supplier provided garbage tacked on)
The first time I built an AOSP image from scratch was on some dinky office workstation that had been freshly upgraded with a whopping 16GBs so you wouldn't come back in the next morning to a random OOM
These days I get to open a PR and some random monster of a machine on a build farm does the heavy lifting, but I can still say from a lot of experience that 64GB is truly more than plenty for Android OS builds and definitely won't be what keeps you from doing other stuff... IO and CPU usage might make it an interesting proposition, but not RAM.
When Google says 64GB it's for the whole machine: the build process will use a lot of it when configured properly, but not so much that you can't run anything else
(Also again, Android builds are a perfect example of where a remote machine pays off in spades. The spin up is such a small fraction of what's needed you don't have to start messing with containerization if you don't want to, and you can get access to some insanely large instance for a few hours then stop paying for it as soon as it's done.
It just seems unlikely to have a task that warrants that much RAM that isn't just about throwing resources at an otherwise "stateless" task that's a great match for cloud compute)
