

TLDW: cooling’s fine if you use starlink V2 size and power (which is not very suitable for AI ‘datacentre’ use) because it works already.
This is not about a few huge datacentres, it’s about a million small ones. There’s 99 problems with this (see Kessler syndrome !, radiation, …), cooling isn’t (much of) one.
Doesn’t matter anyway, it just has to be vaguely plausible for a stock IPO pump (and dump) scheme while sweeping all that xAI debt under the rug.







Simplest put, a fedora immutable usually keeps two images, the one you’ll boot into next reboot, and the one you’re running. If a rpm-ostree update hasn’t been run it’ll be the one you’re running and the last one. My bazzite (heavier than silverblue I guess) images are ~ 14Gb, you need room for three (the two you’re using and room for downloading the next) plus 3% of your hard drive because fedora says so, so 3*14 = 42 + .03 * 240 = 42 + 7.2 = 49.2 =~ 50Gb.
Wait a sec, when I actually do a
sudo du -sh /sysroot/ostree/deploy/fedora/deploy/*I get 14Gb for my previous one and 2.1Gb for my current one, so there’s some diff black magic fuckery (ostree chunking) going on, which makes sense because it’s not taking that long to download. So 50Gb would be super safe, you might get away with 25 depending on how different the two images are (i.e. how much has been updated), but updating to the next major fedora version (e.g. 42->43) would be iffy.
Upshot is, it shouldn’t have filled to 90-something in the first place (maybe before ostree chunking, but even then), but if you end up with a lot of entries in your GRUB they’re all taking a notable chunk of space and you’ll need to purge some.