The rollout comes as generative AI drives surging power draw and heat from high-performance graphics processing units, straining the conventional air-cooling systems that have long kept server rooms in check.
To answer that, LG CNS will deploy direct-to-chip (DTC) cooling — a method that fastens coolant-fed plates onto GPU chips to draw heat away far more efficiently than air.
The infrastructure is designed to run Nvidia's forthcoming Vera Rubin GPUs, owned by the state-backed National IT Industry Promotion Agency, which Naver Cloud will operate.
Vera Rubin racks are expected to deliver about 3.3 times the inference performance of Nvidia's current Blackwell Ultra generation. With a total power capacity of 80 megawatts, the Samsong facility is poised to become one of the country's largest AI factories.
The technology addresses a hard physical limit as air cooling typically tops out at 20 to 30 kilowatts per rack, yet high-density GPU server racks now exceed 200 kilowatts, making liquid cooling a prerequisite rather than a luxury.
"Data center cooling technology is becoming a core competitive edge in AI service quality and operational efficiency," said Cho Heon-hyeock, an executive at LG CNS' data center business division, adding the company would keep building infrastructure to underpin the growth of Korea's AI ecosystem.
AJP Takeaways
•The system is designed to run Nvidia's next-generation Vera Rubin GPUs, owned by the state-backed National IT Industry Promotion Agency and operated by Naver Cloud, with construction and testing targeted for completion in 2027.
•Liquid cooling is becoming essential as high-density GPU racks now exceed 200 kilowatts, far beyond the 20-to-30-kilowatt ceiling of air cooling.
Copyright ⓒ Aju Press All rights reserved.