(there are actualy ways around those problems to, but they are rather technical. One example is that you could describe 1 group of 1000 stars, then you just have several copies of that, with randomly generated diferences only when the player actualy examines them so it dosn't seem like it's the same. Thats just one example, and that could be combined with loads of other techicues. or if the drawbacs from that spesific techniue are to limiting for the curent aplication, using some of the other would still be enought.)
(((((Noctis turns a single number into an entire galaxy. What you would then do is continue the data seeding to produce civilizations and other components, with the civilizations being affected by nearby systems. You have to generate them each time you use them, but that way you have a deeply described civilization, with the only issue being that they can only "interact" with nearby civilizations, and more importantly, cannot change those interactions or evolve in any way.
It works by using the 3d coodrdinates to produce an input to a function that uses the input to decide if a planet exists through pseudorandom methods only relying on the input to determine the output, and weighted to be more likely in the denser regions of the centre of the "galaxy".
Take this one step further, and fully save just the last thousand or two systems the player has meaningfully interacted with in full detail, then another hundred thousand vaguely. (The hundred thousand basically saying how well known the player is or something simple). The player will take a long time to fill up this cache, so they will never notice a civilization "revert" (10 minutes to properly interact with a system * 1000 systems = too much time).
Add to this the probability of a given civilization existing somewhere, and you get maybe one in ten systems with inhabitants, and you can give the player the feeling of an entire galaxy to explore while only saving the details for a few thousand.)))))