Why LG CNS Is Betting on Liquid Cooling as AI Racks Push Beyond 200kW

South Korea's 80MW Samsong Data Center is a symbol of the evolving landscape in AI infrastructure. As artificial intelligence systems demand more power and become denser, cooling has emerged as a critical constraint. LG CNS, in partnership with Naver Cloud, is addressing this challenge by implementing direct-to-chip (DTC) liquid cooling at the Samsong Data Center.

This infrastructure is designed to support NVIDIA's next-generation Vera Rubin platform. The shift towards liquid cooling is significant because traditional air cooling systems can only handle around 20 to 30 kilowatts per rack. However, high-performance GPU server racks are now surpassing the 200kW mark, necessitating more efficient cooling methods.

Direct-to-chip cooling, which places coolant-carrying cold plates directly on heat-generating components such as GPUs, allows for greater computation within a rack while managing the heat produced by powerful processors. This approach is particularly crucial as NVIDIA progresses towards rack-scale AI systems. NVIDIA's Vera infrastructure is designed for high-density AI factories and features liquid-cooled rack systems to complement the NVIDIA Vera Rubin NVL72.

The Vera Rubin architecture represents a broader industry transition towards treating compute, networking, power, and cooling as an integrated system. For data center operators, this shift necessitates the design of facilities that can accommodate AI systems consuming significantly more power within the same physical space. Samsong Data Center is a test case for Korea's ambitions in AI factories.

The facility, with an 80MW total incoming power capacity, is being developed as a large-scale AI computing hub. LG CNS, already operating the data center on a lease agreement and having a colocation arrangement with Naver Cloud, will integrate DTC cooling with GPU server operations, power supply infrastructure, and AI platforms' operating foundation.

This move positions the company further into the physical infrastructure layer required to operate AI at scale. The AI infrastructure battle is evolving below the GPU, as the development at Samsong Data Center illustrates. AI infrastructure competition has shifted from securing advanced GPUs to ensuring these accelerators receive enormous amounts of electricity, are interconnected through high-speed networks, and are kept within safe operating temperatures continuously.

As rack density increases, these supporting systems increasingly determine the extent of AI compute a data center can deploy. NVIDIA's shift towards an integrated model, combining rack-scale computing, high-speed networking, and liquid-cooled infrastructure, highlights this transition. LG CNS, with its data center business expertise in design, build, and operate (DBO), can benefit from adding liquid cooling, expanding its role from data center construction and operation to engineering next-generation AI computing environments.

The significance of liquid cooling extends beyond the engineering aspect, as the global data center liquid-cooling market is projected to grow from $4.07 billion in 2026 to $27.65 billion by 2033. This growth is driven by the physical constraint of processors becoming more powerful while generating heat that must still be removed.

South Korea's position in the semiconductor supply chain, with key players like Samsung Electronics and SK hynix, further emphasizes the broader competition in AI infrastructure beyond semiconductor manufacturing and AI models. LG CNS's liquid-cooling project at Samsong Data Center exemplifies the expanding significance of cooling in the AI infrastructure market.