The majority of these maps were originally intended to be printed on 23x27 inch sheets of paper. Assuming they were printed at around 300 dpi, somewhere between 1-8 bits per pixel (after compression, this is the shakiest part of my estimate!), that's around 50 million pixels per sheet, somewhere around 6-50MB per sheet. So somewhere between 1/2TB to 4TB for the whole data set (and I could easily be too low or too high by a factor of 2).
"Always do the math" is a very nice piece of advice that I got from who knows where. More often than not you nail it.
However, it doesn't stop there. As part of the 3DEP (3D Elevation Program)[1], Lidar/IfSAR data is being collected across CONUS. One of the products that will be available (hopefully next year) is a 1-meter DEM, which should roughly be around 35TB (100 times as much data compared to 10-meter).
A while back I made a script to download NED elevation data from the USGS site. The files were elevation data for 1/3 arcsecond or 1 arcsecond squares and were available in a float format and/or ArcGIS. The files for a 1/3 arcsecond square in ArcGIS format were zipped up and came to ~320 Mb. The float format files are even larger. I don't remember exactly, but I think there are a few thousand of them for the entire United States.
And that was just NED elevation data, not feature data like roads and trails or lakes, or aerial imagery or topo quadrangles.
FWIW, the script is on GitHub, but doesn't work any more due to changes on the USGS website. It actually uses PyQt's webview to do the downloads because the download process used a ton of Javascript with callbacks and notifications because the data was in a tape library and had to be fetched and loaded before the download could begin.
https://github.com/jl2/GIS-Stuff/blob/master/map_download/ne...