You print the code out as OCR text using an OCR font called ISO 1073-1:1976. Then it can be scanned back in as fast as the scanner can read it. Paper isn't just for preserving things for centuries, it actually saved IBM a ton of money when they learned that they needed the source code for a program created just 4 years earlier. The backups of the code had become unreadable from the only hard drive that contained the code. If it wasn't for the paper copies the code would have been lost.
Pardon my saying so, but storing such critical data in only one place is a hell of a stupid thing to do. IBM or not. In fact, for comparable cost/space, I would count digital storage much more efficient and safe. For example, a ream of paper costs what, $5? That seems reasonable. Imagine all the code you could fit into that!
Until, of course, you realize that the same data could easily fit (with room for another 50 pages or so) into a 2MB text file.
So let's scale that up. For a 1TB hard drive's worth of code (a very cumbersome beast to deal with regardless of format, I'm sure) you would need (using base 10 instead of 2) about 275,000,000 sheets of paper - that's 550,000 reams. Obviously, double-sided makes 275,000 reams or 137,500,000 sheets. For that cost, over a million dollars, you could have hundreds of copies distributed around the world - in the cloud, in bunkers in the middle of the sahara, in the CEO's office -
and you would still have the money left over to maintain these copies by periodically cloning the data to new drives ad infinitum. Never mind the space needed to store 275,000,000 sheets of paper. I mean, one ream is reasonably heavy already, for just paper. Imagine half a million of them.