Part of the day: Power shelves & PSUs

Assembled by the nightly build from this catalog’s own entry for Power shelves & PSUs (4.12, Layer 4 — Servers, racks & integration). The tables below are rebuilt every night; published 2026-09-28.

The rack-mounted power shelves that convert facility AC or DC input into the regulated rail distributed to trays across the rack, built to the Open Rack V3 33 kW-class spec used in rack-scale AI systems.

Power shelves sit between the facility’s electrical distribution (layer 7) and the busbar or whips (4.15) that carry power the rest of the way to each tray; they’re the rack’s own bulk power-conversion stage. The Open Rack V3 form factor standardizes a 33 kW-class shelf design shared across ODMs, which lets a rack-scale system swap in shelves from multiple suppliers without re-engineering the rack’s power architecture. AI racks need far more power-shelf capacity than a conventional server rack because GPU trays draw far more continuous power than general-purpose compute, and a growing share of that conversion work happens with wide-bandgap power semiconductors (4.12a) rather than silicon, since GaN and SiC devices handle these currents at higher efficiency and in less space. For investors, this tier tracks rack power density directly: as GPU generations draw more power per tray, shelf count and shelf capacity per rack both rise, and the suppliers who qualify early on a new shelf generation tend to hold that position through the generation’s full production run.

AI delta: GPU trays draw far more continuous power than general-purpose compute, which is why rack-scale AI systems need substantially more power-shelf capacity per rack than a conventional server rack, and why wide-bandgap semiconductors are displacing silicon inside the shelf.

Who makes it

The full entry — sub-categories, news, videos and sources — lives on the Power shelves & PSUs page.