It is being done at the IF; it just happens that the final IF is around
1.5kHz (assuming SSB filters). There is nothing intrinsically audio
in the signals, so the distinction between IF and audio is arbitrary
and based on the original intended use of the receiver. If you
custom designed the receiver, you could do the A/D conversion at any
centre frequency you chose.
> available, I've got to believe there are some super fast but
> probably (still) expensive DSP chips available.