> I was once interning for a chemist, and asked if he ever thought we would have desktop NMR machines. He said there was no way they would ever be small enough. I then pointed at his laptop and remarked that is what people thought about computers 50 years ago.

Good-resolution NMR requires high frequencies and high magnetic fields. A 900 MHz NMR would use around 20 T--which is strong enough to pretty much require superconducting cryogenic magnets. Admittedly, a 100 MHz NMR requires only about 2-3T, which is close to what can be produced with a neodymium magnet, but you do get much, much noisier spectra.

With the larger field strengths, the need for magnetic field shielding pretty much restricts the ability to shrink it down to bench-scale machines.