What kind of machine do you need to run all of this concurrently?
jellyfin is configured to not transcode anything. The vast bulk of my library is DVD/BluRay rips of my own creation and I just ripped them in the desired format in the first place. This could probably keep up with a single DVD-quality re-encode, I dunno about Blu-Ray (depending on config, perhaps), but I just have it serve the correct files in the first place.
There's a ~$125 5TB USB drive hanging off of it for the media storage, which syncthing syncs to another 5TB drive in the house. (I don't actually "back up" my media storage in the full sense; everything else is actually backed up in the full sense to S3 via restic.) The "contention" I mentioned above is because all the big data sets are mostly on that spinning-rust drive.
The Immich AI features worked fine on this, though it did take overnight to process my initial load of ~20 years of photos. However once it chewed through that, the responsiveness is fantastic.
If you want responsive AI that uses GPUs this isn't anywhere near enough, but for any "conventional" app, $125 or $250 buys you a lot nowadays.
I ran an n100 until last week. Worked fine.
I have plex setup to transcode and it serves about 10 users just fine. My plexamp sonic analysis took like 4 days though, lol, but everyone says it takes forever.
My immich import took about 20 hours? So not bad.
I run all my home automation off it. 100+ devices, logging, etc. no issues.
I also sometimes run an OBS stream on it to transcode for YT. The n150 does fine.
Total cost for me is about $550. I saved a lot on HDD by going used server drives. $140 for the n150, $300 for drives, then a cheap UPS and router running openwrt.
As for difficulty, most of this is deployed in a few minutes using docker or install scripts. The hardest part is the choice between various solutions.
It handles all but the AI/LLM. I have a throwaway box with 32GB ram and cores with a nvidia 2080 that does the LLM side of things.