Summary
- LG has joined the Nvidia Partner Network as a Power and Cooling Solution Preferred Partner.
- Its qualified CDU portfolio currently spans 600kW, 1MW, and 2.6MW systems for direct-to-chip infrastructure.
- LG plans to seek qualification for a 4MW CDU by the end of 2026 as centralised coolant-distribution capacity continues to increase.
LG Electronics has joined the Nvidia Partner Network as a Power and Cooling Solution Preferred Partner, strengthening its position in the infrastructure supply chain around high-density AI data centres.
The 28 September announcement gives LG a current commercial framework for a cooling portfolio that already includes Nvidia-qualified coolant distribution units at 600kW, 1MW, and 2.6MW. The company has also said it plans to pursue qualification for a 4MW CDU by the end of 2026.
The scale progression is important because coolant distribution is moving from rack-adjacent equipment towards larger centralised systems capable of serving multiple high-density rows or clusters. As individual racks move further into six-figure wattage, the mechanical systems linking the technology cooling loop to the facility water loop become a material part of the site’s capacity and resilience design.
A CDU maintains the flow, temperature, pressure, and separation required between direct-to-chip cooling circuits and the wider facility plant. Larger units can reduce the number of individual systems required across a data hall, but they also increase the amount of compute exposed to each unit’s pumps, controls, heat exchangers, valves, and maintenance regime.
LG’s preferred-partner status places those products within Nvidia’s broader ecosystem around DSX AI infrastructure, where compute, networking, power, cooling, facilities, and software are increasingly treated as one coordinated architecture rather than separate procurement packages.
The company’s 2.6MW CDU was qualified earlier in September, and the new partner announcement provides the fresh development rather than repackaging that product launch. LG is presenting the CDU range alongside chillers, air systems, controls, and other components as part of a wider chip-to-chiller offer.
That breadth matters for operators because direct-to-chip cooling rarely removes air cooling entirely. Memory, power supplies, networking equipment, storage, and other components can continue to reject heat into the room, leaving facilities with hybrid mechanical systems that have to manage liquid and air loads simultaneously.
Multi-megawatt CDUs also increase the importance of fault-domain design. Operators have to decide how many racks or clusters should depend on one distribution unit, how redundancy is arranged, how isolation can be maintained during servicing, and how leaks or coolant-quality problems are detected before they affect expensive accelerator hardware.
Retrofitting such equipment into existing buildings can be more difficult than the nominal CDU capacity suggests. Pipework sizes, floor loading, plant-room space, pumping capacity, heat rejection, electrical supply, and access for replacement components can all limit how quickly an air-cooled hall can be converted.
LG’s move into Nvidia’s preferred supplier network therefore reflects a wider shift in AI infrastructure: coolant distribution is becoming a core data centre system in its own right. The planned step from 2.6MW towards 4MW units shows how quickly thermal infrastructure is scaling alongside accelerator density.
