LG CNS is expanding its artificial intelligence (AI) factory business through its data center design, build and operate (DBO) capabilities and next-generation liquid cooling technology. It is building infrastructure to operate high-performance graphics processing units (GPUs) reliably and expanding its business from data center operations to AI service delivery.
LG CNS built South Korea's first in-house data center in 1992 and introduced a DBO business model in 2019 that handles the entire process, from design and construction to operation. It has since developed its capabilities through hyperscale data center projects in Hanam, Jukjeon and Samsong. The Hanam data center, completed in 2023, was the first application of the DBO model. According to industry sources, LG CNS has an approximately 60% share of the design and consulting segment of the data center DBO business.
More recently, it agreed to work with Naver Cloud to deploy direct-to-chip (DTC) cooling technology at the Samsong data center. DTC absorbs heat through cooling plates attached to GPU chips, with coolant flowing through the plates. Its higher heat transfer efficiency compared with existing cooling methods makes it suitable for controlling heat from high-density GPU servers. As AI computing performance increases, power density per server rack is rising, making liquid cooling increasingly important. The companies plan to complete the related infrastructure and demonstration by 2027.
The facility is being built to enable Naver Cloud to operate Nvidia's next-generation Vera Rubin GPUs owned by the National IT Industry Promotion Agency. Existing air cooling generally has a capacity of around 20 to 30 kW per rack, while the power density of the latest GPU server racks can reach 200 kW or more. The Samsong data center has a total incoming power capacity of 80 MW.
At Data Centre World Asia 2026, held in Singapore in September 2026, LG CNS unveiled prefabricated modular data centers and full-stack AI factory services based on Nvidia's next-generation Vera Rubin GPUs. Prefabricated modular data centers involve manufacturing GPU servers and power and cooling equipment as standardized modules in advance, then installing them on site. This can reduce on-site work compared with the conventional approach of constructing a building before installing equipment. Capacity can be expanded by adding modules in line with customers' AI computing needs.
LG CNS is also pursuing a business building container-based AI data centers without requiring large sites or separate buildings. It has also launched a Vera Rubin-based module with an IT power load of 4 MW and a GB300-based 1.2 MW module.
The full-stack AI factory service combines GPU as a service (GPUaaS), which provides GPU resources through the cloud, services supporting AI model training and inference, and power and cooling infrastructure. It enables customers to use the AI computing resources they need without building data center facilities themselves.
The company is also accelerating its efforts to enter global markets. LG CNS is pursuing a One LG strategy that combines technologies including LG Electronics' cooling technology and LG Energy Solution's battery solutions. In 2025, it became the first South Korean company to secure an AI data center construction project commissioned by a local company overseas. It is currently building a hyperscale AI data center worth 100 billion won in Jakarta, Indonesia, targeting completion in the fourth quarter of 2026. It plans to expand into other Southeast Asian markets, including Vietnam.
Original Korean article: Chosun Ilbo
English brief by Korea Wind Intelligence, koreawindintel.com
