This would be a great side project to play around with on an Arduino or something.
An air-gapped computer is likely devoid of any special radio-frequency hardware (no wi-fi or BT). Acoustic signals (via speaker or mechanical moving parts) and visible light signals (via the screen or various LEDs) are easy to notice.
OTOH a GSM frequency signal is not readily visible and also does not look like coming from a computer unless you look pretty hard; a signal at this frequency can be coming from a mobile device of a passer-by across the street.
Interesting, though, to see they came up with same solution. Just like old keyboard and pin pad attacks. The user has to enter it (input) or see it (monitor). There are many ways to stop outsiders from getting into that information. So, my idea (and NSA's as well) was to make that raw data come to us in whatever way possible. Simpler methods were available such as cheap bugs at 10Ghz that amateur spectrum tools wouldn't see. However, I found natural emanations to be best risk because (a) they're always there, (b) they often increase for non-suspect reasons, and (c) nobody did detection or defense outside of TEMPEST customers.
It's why I keep advising a damn-near Manhattan project on EMSEC all the way from rooms to COTS components down to automating analysis in synthesis tools for ASIC's. The latter already do it quite a bit for non-malicious interference but the game changes going from designing Murphy's ASIC to designing Satan's. ;) Whatever is produced should bring technical sophistication and cost down dramatically with plenty reusable solutions & reference components.
It's also much more efficient given the communication rate in the article is on the order of just bits/s. Of course, if you're looking for a steady stream of data over time, then this is probably the optimal solution.
I'm not so sure. It's much easier to drop an infected thumb drive in a parking lot than to get that thumb drive back after it's been plugged into a facility computer.