I've thought about this over the years for any number of reasons. The "best" idea I came up with was to use "printed digital information". Some years back I read a (very small) article in Scientific American where Xerox had come up with this idea to implant the digital information of a graph in the graph itself, based on a printed pattern. Essentially something like / = 0 and \ = 1.
If you take that to a DPI level high enough you start to look at holding very large amounts of data; no where near digital storage (I think), but still respectable. If you keep the resolution "low enough" so that a low tech approach could be used to read it (magnifying glass or rough microscope) you add "backwards compatibility" if needed. Going in the direction of "higher tech", a scanner / ETL program could interpret the "data sheet" and produce the output automatically.
With the quality of printers / ink increasing as they have (I have an Epson Stylus 4500 with "archival" ink rated for 90+ years) you could in theory print the data archive on (photo?) paper, then store the paper. The volume of raw data would increase compared to written text. The potential for extracting the data again if desired is present (scanner, etc), and it has the possibility to be "backwards compatible" to lower tech retrieval (if needed).
I kept wanting to add to the density of storage by a couple of ways. The idea of a "denser" arrangement of pixels was one (the / and \ in a 2 by 2 pixel array uses half the available pixels). I kept thinking that using all 4 would be better (a nibble for all us old timers out there). If you add to that the idea of using color in the encoding, you should be able to go with some multiple of 4 bits quite nicely.
There are other aspects to this idea as well; adding instruction in the beginning to help decode (if needed), noting the standards used to encode data; define MP3, mpeg4, etc.
Beer's Law: Absolutum obsoletum
"if it works it's out-of-date"