I trades one technical limitation for another. Instead of recomputing the geometry each and every time, it goes into RAM and treats the whole world like a big giant look up table. Note there is currently no physics, no animation(have to attach those voxels to a frame work or skeleton), no real time lighting, the same "set" of objects, no ability to alter the environment, it's completely non-interactive.
They have traded processing power, something that so far effectively still doubles every 18/24 months for a godly amount of RAM needed to store all those objects, not to mention the loading times even with an SSD. No Next Gen console will use this whole, just from the lack of RAM. It might find limited use as far as hybrid systems where you might want one object with very high detail, say a non-destructible building, but it's resource footprint would be huge compared to it's traditional counterpart especially on things like consoles. Note they don't tell you what that demo is running on.
So yeah, you might see some limited applications next gen, but I wouldn't even put money on that.