Cooling — Direct-to-Chip Liquid, CDU & Immersion

AI racks have moved from ~20 kW to 120 kW and beyond, past the point where air cooling is physically viable. Moyan Technology supplies the liquid-cooling hardware that makes dense AI racks possible — direct-to-chip (DTC) cold plates, coolant distribution units (CDUs), manifolds, rear-door heat exchangers and immersion systems — sourced direct from the Taiwan and China factories that build most of the world’s AI-rack thermal hardware.

Cooling is capacity-constrained, with 12–18 month lead times on cold plates, quick-disconnects and rack CDUs — a real reason to secure second-source Taiwan/China supply. Most of the category is quoted per project after a thermal assessment.

What we supply

Cold plates — copper microchannel DTC plates for GB200/GB300 GPUs and CPUs.
CDUs — in-rack (to 250 kW), row-based (to 2 MW), and liquid-to-air.
Manifolds & quick-disconnects — OCP UQD-standard, blind-mate.
Rear-door heat exchangers — 50–75 kW brownfield bridge.
Immersion — single- and two-phase tanks and CDUs.

Products, capacities and indicative anchors are in the catalog table on this page (stamped “as of” a date; most systems are project/RFQ).

Why source cooling through Moyan Technology

Second-source Taiwan/China supply for lead-time-constrained parts.
Component to turnkey — cold plates through full CDU loops and immersion pods.
Standards-aware — OCP UQD quick-disconnects for drop-in multi-sourcing.
Density-matched — sized to your rack kW and cooling architecture.

Cool your GPU racks and design alongside racks & chassis and power.

Request a cooling quote →

FAQ

Why is liquid cooling mandatory for AI racks? Air tops out around 30–41 kW per rack; GB200 NVL72 racks reach ~120–132 kW and next-gen platforms push toward 200–250 kW. Direct-to-chip liquid removes heat far more effectively.

What is a CDU? A Coolant Distribution Unit circulates and conditions coolant between the facility and the rack loop, in in-rack, row-based or liquid-to-air forms, from ~250 kW to 2 MW.

Cold plate or immersion? Direct-to-chip cold plates are the NVIDIA-specified default for rack-scale AI; immersion is a chip-agnostic alternative. We supply both and can advise per site.

How are cooling systems priced? Turnkey CDUs, loops and immersion pods are quoted per kW / per project after a thermal assessment; component anchors are indicative.