At the LG Uplus Paju AI data center (DC) construction site in Paju, Gyeonggi, on the 5th. On a 150,000-square-meter site—about 21 soccer fields—five large cranes were moving materials. The site, with a construction progress rate of about 20%, was running nonstop. This is the only 200MW hyperscale AI data center under construction in the greater Seoul area. It can accommodate about 70,000 graphics processing units (GPUs). That is enough capacity for the entire greater Seoul population to run Generative AI services simultaneously.
The LG Uplus Paju AI data center stands out for its design to control heat generation. The company said this enables simultaneous operation of air and liquid cooling, allowing flexible response to the rapidly changing AI technology environment and diverse market demand.
An Hyeong-gyun, head of the Enterprise AI Business Group at LG Uplus (executive director), said, "All contracts for Building 1, a 50MW facility slated for completion in June next year, have been finalized, putting it in a 'sold out' state," adding, "By 2030, we aim to achieve cumulative orders of 5 trillion won with a 600MW AI DC and expect AI DC revenue to log average annual growth of about 15–20% starting this year."
LG Uplus currently operates 15 AI DCs nationwide, with AI DC revenue at 422 billion won last year, up 18.4% from 2023. During the same period, business-to-consumer (B2C) revenue centered on telecommunications grew only 3.3%.
◇ Shorter build time with modular method… simultaneous support for air and liquid cooling
The reason the LG Uplus Paju AI DC reached 20% progress in about a year is because it adopted the standardized prefabricated modular data center (PMDC) method. PMDC standardizes major equipment, prefabricates key structures, and assembles them on-site. Through this, the company expects to shorten the build period by several months compared with the traditional approach.
The Paju AI DC also focused on cooling efficiency. Heat management is one of the core competitive factors for data centers where AI computation is concentrated. As GPU power density rises, cooling technology determines efficiency and stability. This site is being built with a hybrid structure that supports both air and liquid cooling, the first in Korea. The liquid cooling system addresses shortcomings of conventional air-cooling systems, which suffer from lower energy efficiency and loud fan noise. From the construction stage, the Paju AI DC designed building load, waterproofing, and piping for liquid cooling to handle GPU server heat. In particular, the liquid cooling facilities built in collaboration with LG Electronics use a direct-to-chip (D2C) method that attaches a dedicated metal cold plate to the GPU chip and circulates coolant through a coolant distribution unit (CDU) to remove heat directly.
The company said, "Internal validation confirmed about a 24% improvement in energy efficiency compared with conventional air cooling." On not adopting an immersion liquid cooling (ILC) method that submerges servers directly in liquid at the Paju AI DC, Jeong Suk-kyung, executive director in charge of the AI DC business, said, "It is true that immersion cooling is highly energy-efficient, but Nvidia is currently verifying immersion technology, and LG Uplus also needs to validate the technology, so we plan to adopt it after verification is completed."
After completion, the Paju AI DC plans to use robots to monitor building temperature and humidity, leaks, and dust around the clock, 365 days a year, and to monitor the outer grounds to strengthen stability. Through technological collaboration among LG Group affiliates, key equipment such as cooling facilities, batteries, power equipment, and robots will be localized. In cooling, LG Electronics will produce the coolant distribution unit and the D2C liquid cooling solution, as well as an air-cooled "free cooling chiller" that produces coolant. LG Energy Solution's UPS batteries will immediately stabilize power during outages or voltage fluctuations, and their multi-layered safety design from cell to pack will minimize fire and thermal runaway risks. To address high power usage, the DC 800V power distribution system is being co-developed by LG Uplus and LS Electric. LG Uplus is also developing an AI-based DCIM (data center infrastructure management) system to operate such equipment in an integrated manner.
An, head of the Enterprise AI Business Group, said, "LG Uplus plans to leap forward as an 'AI factory operator,'" adding, "We will go beyond simply building data centers or leasing server space to become an infrastructure operator that integrates management of GPU resources, power, cooling, and all other elements like a 'factory' so AI can run in optimal conditions."
◇ SK Telecom and KT also focus on AI DCs
LG Uplus is not alone in ramping up AI DCs. All three telecom companies noted in recent earnings that the AI DC institutional sector posted double-digit growth, confirming it as a future growth engine. SK Telecom's AI DC revenue in the first quarter this year was 131.4 billion won, up 89.3% from a year earlier. SK Telecom currently operates nine AI DCs in Korea. KT Cloud, which handles data center operations within the KT Group, posted first-quarter revenue of 250.1 billion won. KT currently operates 16 AI DCs in Korea.
They are also accelerating AI DC builds. SK Telecom is working with Amazon Web Services (AWS) to build an AI DC in Ulsan. SK Telecom plans to secure more than 300MW of data center capacity through 11 AI DCs by 2030. KT also plans to expand its AI DC infrastructure to more than 500MW by 2030 by adding facilities under construction in the western region.
Revenue models have also diversified. In addition to its GPU-as-a-service (GPUaaS) business that rents computing infrastructure in the cloud, SK Telecom is also running a colocation business that builds DCs and leases them out, like the Ulsan AI DC. At the Gasan AI DC, KT Cloud introduced Colo.AI, a subscription-based personalized GPU infrastructure service.