The thing is, you get down to the noise floor on every cycle. When you swing from negative to positive and back, you go through zero. The shorter the word length, the more dynamic range between the noise floor and zero, or the less low level resolution.
I believe that 1 bit represents 6db of dynamic range.
I know it's a long used convention to speak in this way but I suggest that speaking of that line half way between positive and negative swings as "zero" or "the zero crossing point" only adds to the confusion here, especially when we already speak of the digital clip point, the maximum useable level, as a zero. If we do speak of the clip point as zero, then there is no meaningful "zero" down there at all, just points a certain number of db below clip point.
Ethan spoke of a top converter having Johnson noise at -130dbFS. Think of that as the outer borders, both positive and negative, in a band which extends equally above and below the middle line. Anything quiet enough to be within that band is effectively overwhelmed by that noise, so we have to speak of that entire band as "indeterminate". The system cannot resolve it.
The 24 bit point, at -144dbFS, can similarly be thought of as a band but just a narrower one.
Better, I suggest, not to speak of "going through zero" on each cycle but crossing into the "indeterminate zone" - and every audio system has such a zone - because any point within that zone is beyond the resolving power of the system.
Once you get into this unresolveable zone, it's not possible to know at any given moment whether the audio wave is either positive or negative, let alone give it a value.
The midway line can only be spoken of in actual level terms as "minus infinity dbFS." or in digital terms "infinite bits". It's a theoretical construct, not a reality we can measure.
Speaking practically, the system, like a chain, can only ever be as good as its weakest link. In noise terms, if you are recording at 24 bit resolution and with converters equivalent to 20 bits but any other part of the chain,
including the aircon noise in the building, is at say 12 bits, then 12 bits is the maximum dynamic range you can resolve. In this case, recording at 16 bits would sound no worse. You would lose nothing.
Hope this helps.
Tim