What if the terrain is "seeded", i.e. generated based on a predetermined hash, and then only thehash needs to go in the savegame data.
I don't know exactly what data format defines terrain in Skyrim or Fallout 4 as-is, but I doubt it would be worth it. The only benefit in procedurally generating terrain at runtime would be expanding the space for exploration at the expense of the level designer's opportunity to handcraft. You could get around that detriment with a mix and match approach, parts X and Y of a planet are developer handcrafted, and beyond their bounds, the game will generate random landmasses, but you'll soon run into a serious case of diminishing returns...
Think of it this way, if you have each planet offer an explorable space the size of, say, Far Harbor, you might as well have pregenerated the heightmap and dumped it to the 125GB of game files, tweaked or not - no save bloat, no CPU load. Proc-gen would only help if you really wanted planets to be
planet-sized and fully explorable, but at that point I don't think it'll matter how much you optimise the seed's format (and ulterior runtime decompression!), you're talking about a thousand planets, each with potentially more terrain than any single prior Bethesda game, that could be massive bloat for a save that has to stay within a couple of dozen MBs. Basically, the more you take advantage of the feature, the more you bloat the save. And you can't really take cell reset buffer approach because it's one thing for Lexington to clear dead raiders after a while, it's another for the same set of coordinates to land you on a mountain today and a lake next week. It's possible, but it'd be a departure from prior form, and I don't think it's very appealing.
And finally, I'm not even sure what the gameplay case would be to have such large planets, not for general audiences anyway. It'd be a cool novelty and you know there's
that one guy who'll want to explore
everything, but I imagine most people will have had their fill after their 160th Far Harbor. If this were a single star system, I could see the appeal of being able to "land anywhere" on five or six planets maybe justifying the technical effort, but with
a thousand planets?... That, to me, suggests design priorities were likely elsewhere.