Probably everyone has seen it. Main panel is full so a sub panel is set next to it, however it is not supplied with a neutral feeder, only hots and a bond. The circuits all enter the main panel first and only the hots are extended into the sub panel, the neutrals and grounding wires are landed in the first panel. Does a violation exist?
Neutral-less panels are as common as dust in industrial settings. These are designed to fit within the 'web' of a vertical I beam/ column. Since such panels are also very, very, likely to feed 3-phase loads, the neutral doesn't get a lot of action, anyway.
You'll find them listed in Square D catalogs.
They are frowned upon elsewhere. The operative assumption being that subsequent players are going to be so stupid as to not correctly land neutral conductors. The 'savings' from omission of a neutral path into the adjacent panel are insignificant.
I must be a rare electrician, I've never seen the set-up described. I don't work the heavy industrial side of the street.
I have seen them before but can't swear they were to code. I'm gonna keep digging because it has my cariousity. So far, in article 215 Feeders, there is a minimum grounded conductor size shall not be smaller the required by 250.122. However I'm not sure what they mean by that
The neutral does not need to extend beyond the service disconnect enclosure. Back when there were real computer rooms, we never brought the neutral to the computer panels. They had a "building" panel supplying the lights and the 120v loads but none of them were computer related, everything was 208 or 240 L/L loads. The 120v monitors were plugged into a service outlet in the CPU that ran off an isolation transformer.
There was still a neutral style bar in the panels but it had a green wire feeding it usually with a yellow tracer and was the isolated grounding conductor that went to the machines. The can and the raceways were bonded to the "bare wire" ground (the one with green only if it was insulated).
This fell out of vogue about the time they stopped water cooling the machines (and coincidentally the CMOS got a lot less noise sensitive)