I'm imaging a way you could procedurally generate on-the-fly explorable planets without taking up a titanic amount of data.
The way is, basically, inception.
Essentially- You enter a solar system. The game generates a "planet" entity, which is a single, unsubdivided object with a procedurally generated size and combination of texture layers to create the planet's superficial appearance (i.e. layers of multihued striped for the gas bands of a gas giant, blue base plus continental splotches and a cloud brush to generate blue marbles, etc.)
If you were to leave the system now, it'd stay like this- a single entity, single poxel really I suppose. Minimal data needed to store it.
Then, as you approach, the planet-poxel divides into subvoxels. Hemispheres, perhaps, or quarters of hemispheres. These subdivisions are then procedurally generated the macrodetails of a planet- continents, polar caps, oceans, that kind of Huge Stuff. More detail, still working on the fly, only a relatively small bit of additional data to store.
This process continues- if you fly closer, then the global-subpoxel segment you're closest to subdivides further and generates somewhat greater detail in those subdivisions, but only those subdivisions. This process would continue until you touched down and landed into an MC-style chunk with individual block level terrain generation, but the beauty is that only that chunk you touched down on would have that level of detail, minimizing the data storage needed. Exploring about would generate other chunks, MC style.
Then, if you were to lift off again so you could travel a significant difference without taking months to do so, achieving flight altitude in whatever craft you were in would tell the game to put a proportional cutoff on the lowest subpoxel level it will render, so it would go back to increasingly larger and less detailed planet-subdivisions in your high altitude flyby, so as to not waste data storage and processing time generating the minute terrestrial details of everything you pass over in your flyby. If you were to do this and then land again, you would once again go through the increasing-detail-subvoxel-generation as you landed, or you could fly out to space again. In which case the planet is still there, still planet sized, but still not exorbitantly data or rendering expensive, because it only needs to store the relevant subdivisions, and only needs to display the highest-level ones again.
You could also use this method to implement scanning of planets from space- scanner tools would essentially be telling the game to apply a "brush" of sorts which causes it to procedurally generate a specified area/level of detail on the planet where your ship is scanning it. You would be able to tweak the parameters of your scan, to either sweep a broad swathe in low detail or focus in on a specific area for greater detail. This would then be as code-simple as correspondingly telling the game that for the moment it's only being allowed to procedurally generate up to a certain subvoxel level, and that as that level goes deeper, the size of the "brush" gets smaller. This both fits the idea of "focusing your scanner" and helps ensure that the game isn't overwhelmed having to generate too much high level detail at once.
Lastly, if a planet is intended to be inhabited/fortified by an NPC faction, then something like MC could be done where there's a chance one of the deep-level subdivisions spawns a town or a bunker/base or somesuch, for more organic/procedural placement. Or, if such things were part of a quest or other plot significant thing where a specific feature being at a specific place on a specific planet was important, the devs could just assign code to the designated planet along the lines of "whatever else, When you generate THESE LEVEL CHUNKS in THIS AREA, you must include THESE FEATURES."