NETONX EN|DE Request a quote
← Liquid Connectors (UQD)
HomeData Center › NVIDIA GB200 / GB300 Liquid Cooling

NVIDIA GB200 & GB300 Liquid Cooling: UQD Quick Disconnects per NVL72 Rack

AI racks are moving fully to direct liquid cooling (direct-to-chip). With every GPU generation the number of required UQD quick disconnects rises sharply – a fast-growing procurement item.

Netonx Europe · 06/2026 · reading time approx. 4 min.

In short: An NVIDIA GB200 NVL72 rack needs roughly 126 UQD coupling pairs; a GB300 NVL72 rack roughly 250 pairs (industry estimates). Every GPU generation raises the quick-disconnect count, making UQD couplings a fast-growing, high-volume procurement item — and a strong case for a qualified second source on the OCP UQD interface alongside Stäubli, Parker or CPC.

Demand per NVL72 rack: GB200 vs. GB300

In the GB200 class, a compute tray in practice needs about six coupling pairs (cold-plate supply/return plus manifold connection). Scaled to an NVL72 rack (18 compute trays + 9 switch trays), this comes to roughly ~126 pairs per rack.

The GB300 generation uses individual cold plates per chip instead of large-area cooling. This raises the count per compute tray (six chips ≈ 12 pairs, with manifold ≈ 14). Market reports assume around 250 pairs per NVL72 rack – almost a doubling versus GB200.

FeatureGB200 (NVL72)GB300 (NVL72)
Cooling conceptlarge-area cold plateindividual cold plate per chip
Coupling pairs per compute tray~6~12–14
Coupling pairs per rack (≈)~126~250
UQD form factorUQD / UQDB (OCP)more compact UQD generation
Manifold / CDU / cartridgelargely as GB200

Figures are industry estimates from market reports and serve as orientation; the exact quantity depends on the specific tray/manifold design.

Trend: smaller couplings, higher requirements

As slot density rises, the quick disconnects get smaller. A more compact UQD generation saves installation space but places higher demands on manufacturing tolerance, sealing (non-spill) and blind-mate tolerance – at the same time as more mating cycles and more couplings per rack. Reference designs typically use OCP-compliant UQD from established manufacturers (CPC, Stäubli, Parker).

What this means for procurement & distribution: UQD demand per rack rises significantly from GB200 to GB300. Anyone supplying liquid-cooling projects needs a qualified, deliverable second source – compatible with the OCP standard, with proven sealing and sufficient volume.
UQD quick disconnects from NVIDIA-qualified production

Netonx (via Green MeOH): OCP/UQD-compliant, leak-free, as an alternative or second source – European point of contact.

To Liquid Connectors & inquiry →

Frequently asked questions

How many UQD quick disconnects does an NVL72 rack need?

GB200: ~126 pairs per rack (18 compute + 9 switch trays). GB300: per market reports ~250 pairs per rack due to individual cold plates per chip.

What changes with GB300 versus GB200?

Individual cold plate per chip → more couplings per tray and a more compact UQD generation with higher tolerance/sealing requirements. Manifold, CDU and cartridge remain largely the same.

Is there a second source for UQD?

Yes – alongside CPC, Stäubli and Parker, Netonx (via Green MeOH) supplies OCP/UQD-compliant quick disconnects from NVIDIA-qualified production as an alternative or second source.

Note: brands such as NVIDIA, CPC, Stäubli and Parker belong to their respective owners; references are for objective comparison. Quantities are industry estimates.