South Korea's Telecoms Race to Build AI Data Centers as GPU Demand Soars
LG Uplus is constructing a 200-megawatt AI data center in Paju, Gyeonggi province that will house roughly 70,000 graphics processing units-enough to run generative AI services for the entire greater Seoul population simultaneously. The facility, which is 20% complete after one year of construction, sold out its first 50-megawatt building before completion.
All three of South Korea's major telecom carriers have identified AI data centers as their primary growth engine. LG Uplus generated 422 billion won in AI data center revenue last year, up 18.4% from 2023, while its traditional telecommunications business grew just 3.3%.
Modular Construction and Hybrid Cooling Cut Build Time
LG Uplus adopted a prefabricated modular data center method that standardizes equipment and assembles structures on-site rather than building from scratch. The approach shortens construction by several months compared to traditional methods.
The Paju facility is the first data center in Korea to support both air and liquid cooling simultaneously. As GPU density increases, cooling efficiency directly affects operational costs and system stability. LG Uplus designed the building's load, waterproofing, and piping from the start to accommodate liquid cooling.
The direct-to-chip liquid cooling system, developed with LG Electronics, attaches a metal cold plate to each GPU and circulates coolant through a distribution unit to remove heat. Internal testing showed 24% better energy efficiency than conventional air cooling. The company opted against immersion cooling-which submerges servers directly in liquid-citing the need to wait for Nvidia's verification of that technology.
LG Uplus plans to deploy robots to monitor temperature, humidity, leaks, and dust around the clock after the center opens. Key components including cooling systems, batteries, and power equipment will be manufactured by other LG Group companies to keep production domestic.
An Hyeong-gyun, head of LG Uplus's Enterprise AI Business Group, said the company aims to become an "AI factory operator" that manages GPU resources, power, cooling, and infrastructure as an integrated system rather than simply leasing server space.
SK Telecom and KT Expand Capacity Targets
SK Telecom's AI data center revenue reached 131.4 billion won in the first quarter of 2026, up 89.3% year-over-year. The company operates nine AI data centers and plans to secure more than 300 megawatts of capacity through 11 facilities by 2030, including a center being built with Amazon Web Services in Ulsan.
KT Cloud generated 250.1 billion won in first-quarter revenue and currently operates 16 AI data centers nationwide. KT plans to expand its AI data center infrastructure to more than 500 megawatts by 2030 through new facilities under construction in the western region.
Revenue models have diversified beyond simple server leasing. SK Telecom offers both GPU-as-a-service for cloud computing and colocation services where customers lease entire data centers. KT Cloud introduced Colo.AI, a subscription service that lets customers customize their GPU infrastructure.
LG Uplus expects AI data center revenue to grow 15-20% annually and targets cumulative orders of 5 trillion won with 600 megawatts of capacity by 2030.
For management professionals, these expansions signal a shift in telecom business models. Data center infrastructure now drives growth faster than traditional services, requiring companies to master unfamiliar operational challenges-power distribution, thermal management, and equipment integration-while competing on capital deployment and technical expertise. See AI for Management and AI for Operations for resources on infrastructure strategy and operational planning.
Your membership also unlocks:
