Korean data centers adopt liquid cooling as AI GPUs drive heat surge

As the artificial intelligence (AI) data center market expands at a rapid clip, liquid cooling technology is emerging as a critical solution to manage the intense heat generated by high-performance graphics processing units (GPUs). Traditional data centers have long relied on air cooling, which pushes cold air through server rooms to regulate temperature. However, AI data centers operating state-of-the-art GPUs at scale are increasingly turning to liquid cooling, as airflow alone is no longer sufficient to dissipate the heat produced by these dense, power-hungry chips.

According to market research firm MarketsandMarkets, the global data center cooling market is projected to grow at a compound annual growth rate of 16.1%, from $13.23 billion this year to $37.62 billion by 2033. The surge is driven by escalating power consumption and compute loads from AI and cloud services, which have made efficient cooling a top priority for operators worldwide.

The shift is particularly pronounced with advanced GPUs, which carry high power density and generate substantial heat. If the thermal output of AI semiconductor chips is not managed in real time, equipment can overheat, leading to degraded performance or even system failures. Conventional air cooling, while simpler to install and maintain, falls short in AI environments where power density per rack is more than double that of standard data centers. Air cooling typically handles 20–40 kW per rack, but racks populated with Nvidia's next-generation "Vera Rubin" GPUs can reach up to 230 kW.

Liquid cooling has emerged as a compelling alternative. One widely adopted method is direct-to-chip (D2C) cooling, which attaches a cold plate—through which coolant flows—directly to AI chips to absorb heat. Because liquids transfer heat far more efficiently than circulating air, this approach can recover 65–75% of the heat generated by GPUs and neural processing units (NPUs). Still, the remaining 25–35% of heat from memory, networking gear, and other components must be handled by air, meaning most AI data centers ultimately operate in a hybrid configuration that combines both cooling methods.

Industry observers also expect immersion cooling, which submerges entire servers in a dielectric coolant, to gain traction in the coming years. While high coolant costs and maintenance complexity remain hurdles, demand for immersion cooling is likely to rise once rack-level power consumption in AI facilities exceeds 1 megawatt (MW).

Korean companies are already moving aggressively in this space. LG CNS is set to deploy direct-to-chip liquid cooling at the Samsong data center in partnership with Naver Cloud, and is conducting joint research on immersion cooling with the Korea Energy Technology Evaluation and Planning (KETEP). AI infrastructure firm Elice plans to introduce a hot-water cooling system—using water at 40 degrees Celsius or higher instead of chilled water—at a modular data center currently under construction, aiming to efficiently manage heat from high-power GPU equipment.

Global momentum is also building. In June, Nvidia established "100% water cooling" as the standard for its next-generation Vera Rubin–dedicated AI data centers, arguing that fully liquid-cooled infrastructure can dramatically cut power consumption. According to Nvidia, a large-scale 50 MW data center switching to liquid cooling could save more than $4 million annually in cooling-related energy and water expenses.

Samil PwC, in a recent report, noted that the competitive landscape of the AI industry is shifting from merely securing GPUs to mastering thermal management. "As the power density and heat generation of AI servers surge, how efficiently limited power and space are used has emerged as the factor that determines a data center's competitiveness," the report said. This evolving dynamic underscores that cooling technology is no longer a secondary consideration but a core strategic element in the build-out of next-generation AI infrastructure.