LG Electronics is scaling liquid cooling for the artificial intelligence data center buildout with a 2.6-megawatt Coolant Distribution Unit (CDU) that has qualified as NVIDIA DSX Ready. The milestone pushes cooling infrastructure further into megawatt territory as dense accelerator clusters increase the amount of heat that operators must remove from server rooms.

According to LG’s announcement, the new CDU meets applicable NVIDIA DSX reference-design requirements for AI factory infrastructure. It follows qualification of LG’s 600-kilowatt and 1-megawatt systems.

What a 2.6 MW CDU does

A CDU (Coolant Distribution Unit) is a central part of Direct-to-Chip (DTC) liquid cooling. It circulates coolant through cold plates attached to high-heat components such as CPUs (Central Processing Units) and GPUs (Graphics Processing Units), then transfers that heat into the facility cooling system.

That matters because modern AI racks can concentrate far more power and heat into a smaller footprint than conventional enterprise computing. Instead of treating cooling as a secondary mechanical system, operators increasingly have to design compute, power distribution and thermal infrastructure together.

From the chip to the cooling plant

LG describes its broader strategy as “Chip-to-Chiller.” Alongside CDUs, the portfolio includes cold plates, chillers and CRAHs (Computer Room Air Handlers). The objective is to manage heat from the semiconductor package through the data hall and ultimately into the facility heat-rejection system.

For data-center technicians, this shift also changes the operational checklist. Flow, pressure, coolant quality, leak detection, pump redundancy and heat-exchanger performance become increasingly important alongside familiar electrical, networking and server-health measurements. BitcoinVersus.tech has previously examined what happens inside an AI data center, where the physical infrastructure behind computation is just as important as the silicon.

Why megawatt-scale cooling matters

The 2.6 MW rating illustrates how quickly AI infrastructure is moving toward centralized, high-capacity liquid cooling. LG says the system is intended to manage thermal loads across multiple high-density server racks while making more efficient use of available data-center space.

The development also connects two markets BitcoinVersus.tech follows closely: semiconductor efficiency and physical data-center infrastructure. Faster accelerators only deliver useful compute when facilities can continuously supply power and remove the resulting heat. As rack densities rise, cooling equipment is becoming part of the performance envelope of the AI system itself.

LG’s qualification does not by itself guarantee deployments or customer orders. It does, however, establish the 2.6 MW CDU as meeting the applicable NVIDIA DSX requirements and adds another high-capacity option for builders designing next-generation AI facilities.

Video context

A recent report on LG’s earlier NVIDIA liquid-cooling qualification provides additional background on the company’s CDU technology and the growing thermal challenge in AI data centers.