2.5MW High-Capacity CDU Approved by Nvidia
Key Facility for Managing Heat in AI Data Centers
Aiming Squarely at the AI Data Center Market with Total Solutions

LG Electronics is ramping up its efforts to penetrate the next-generation artificial intelligence (AI) data center cooling market by receiving specification approval from Nvidia for its 2.5 megawatt (MW) class Cooling Distribution Unit (CDU). With approval for both its existing 600 kilowatt (kW) and 1MW products, as well as now its high-capacity model, the company has expanded its lineup of liquid cooling solutions.


On September 22 (KST), LG Electronics announced that its 2.5MW CDU has been approved by Nvidia as a ‘DSX Ready CDU’ product. The company also satisfied all additional requirements set forth in Nvidia's AI Factory certification process.


Nvidia has developed the ‘Nvidia DSX AI Factory Platform’, which integrates computing, networking, power, cooling, facilities, and software. Products meeting the platform’s requirements receive DSX Ready certification, and for CDUs, only models with a capacity of 2.3MW or higher required for AI factory deployments are eligible for approval.


LG Electronics secured Nvidia specification approval for its 600kW CDU in July, 1MW CDU in September, and now, the 2.5MW model. On the same day, LG Energy Solution's Battery Energy Storage System (BESS) was also approved as a DSX Ready product.


LG Electronics' liquid cooling solution CDU (Cooling Water Distribution Unit) lineup. (From left) 600 kW, 1 MW, 2.5 MW products. LG Electronics.

LG Electronics' liquid cooling solution CDU (Cooling Water Distribution Unit) lineup. (From left) 600 kW, 1 MW, 2.5 MW products. LG Electronics.

View original image

The CDU is a core facility for managing the heat generated by AI data centers. With the adoption of next-generation AI accelerators based on the Nvidia Blackwell architecture, server power density is increasing, making advancements in data center cooling technology essential. Server racks now typically exceed several hundred kW, reaching MW-class power densities.


As a result, the market is shifting from the ‘In-Rack’ method—installing low-capacity CDUs in each rack—to the ‘In-Row’ approach, which cools multiple racks together with high-capacity CDUs installed by row.


The 2.5MW high-capacity CDU is designed to address such highly dense and high-heat data centers. Considering that the latest AI server racks have a capacity of approximately 120kW, cooling a group of 20 racks requires a cooling capacity of about 2.4–2.5MW. Accordingly, the 2.5MW class has become the de facto standard for high-capacity CDUs.


According to market research firm TrendForce, the adoption rate of liquid cooling solutions in AI data centers is expected to rise from 14% in 2024 to approximately 60% in 2027. Dell’Oro Group also analyzed that with next-generation chipsets consuming more power, liquid cooling is becoming an essential infrastructure technology for AI factories.


With this latest approval, LG Electronics has broadened its cooling solutions lineup to manage heat in AI data centers, now offering chillers, Computer Room Air Handlers (CRAH), and CDUs in 600kW, 1MW, and 2.5MW classes in addition to its traditional air conditioning equipment.


The company also provides ‘Chip to Chiller’ total solutions, connecting cold plates—which directly cool AI chips—to the broader cooling infrastructure, enabling end-to-end coverage from chip to cooling plant.


Based on these capabilities, LG Electronics plans to provide cooling solutions tailored to the needs of data center operators and to bolster its presence in global markets.



Jaeseong Lee, President and Head of the ES Business Division at LG Electronics, stated, "With the recent Nvidia DSX specification approval, we will expand the supply of even more reliable and efficient cooling solutions to global big tech companies, securing our leadership in the next-generation liquid cooling market."


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing