The spread of artificial intelligence (AI) is changing data center design and site selection criteria. As GPUs are installed at higher densities, conventional air cooling is giving way to liquid cooling. With access to large amounts of electricity becoming more important, data center locations are also expanding beyond the Seoul metropolitan area.
Jo Jeong-min, who is responsible for business at SK Telecom's AIDC Integrated Promotion Group, outlined these changes in a presentation titled "The data center paradigm shift in the AI era and SK's AI DC strategy" at iForum 2026, held at Seoul Dragon City in Seoul's Yongsan district on October 22.
"While conventional data centers were facilities for reliably housing and operating servers, what matters for AI data centers is how efficiently they can use GPUs' computing performance," Jo explained. "As GPUs are installed at higher densities, all design criteria are changing, including those for buildings, power, cooling and piping."
From air cooling to liquid cooling: how AI is changing data center design
The biggest change is in cooling methods. Conventional data centers have mainly used air-cooling equipment to remove heat generated by servers. But as AI servers' power density rises, controlling heat with air alone is becoming difficult.
"Conventional air cooling reaches its limits at around 50 kW," Jo said. "As GPU performance improves and power consumption increases, AI data centers are moving to liquid cooling."
The change in cooling methods also affects data center building structures. Newly introduced equipment includes cold plates attached directly to AI chips to remove heat, along with coolant distribution units (CDUs). Structural changes are also being made to accommodate higher floor loads as servers become heavier.
Power supply methods are changing as well. High-voltage equipment and overhead wiring are being introduced to provide reliable power to high-density AI servers. Piping required for liquid cooling is also becoming increasingly important as a core part of the infrastructure.
"Piping is like blood vessels in the human body," Jo said. "As the transition to AI data centers progresses, piping design becomes even more important."
Jo also noted that AI is being actively introduced into data center operations. AI servers generate considerable heat, and the time it takes to reach dangerous conditions after a fault is short, limiting the effectiveness of conventional responses that rely on personnel.
SK Telecom is applying AI to its data center infrastructure management system (DCIM) to improve operating efficiency. The approach involves integrated monitoring of servers and equipment and advance analysis of potential faults.
"AI servers generate a lot of heat, and when a fault occurs, the time it takes to reach a dangerous level is very short," Jo explained. "There are limits to people making judgments and responding directly, so we are also introducing AI into data center management systems to enhance them."
Beyond the Seoul metropolitan area: power access and local workforce training
Site selection criteria for AI data centers are changing too. Conventional data centers were concentrated in the Seoul metropolitan area because of proximity to users, but AI data centers are being built primarily in regions where large amounts of electricity can be secured.
"It would not be an exaggeration to say that about 90% of existing data centers are concentrated in the Seoul metropolitan area," Jo said. "AI data centers require large sites and substantial amounts of power, so their locations are shifting to areas where these can be secured."
Jo explained that while spare power supply capacity in the Seoul metropolitan area is limited, southern and eastern regions are relatively better placed to secure large amounts of electricity. Industrial complexes outside the metropolitan area are also separated from residential areas, resulting in relatively fewer concerns about complaints over data center construction.
"SK Group is currently focusing its search for data center sites on areas outside the Seoul metropolitan area rather than within it," Jo explained. "Other regions are relatively advantageous in terms of securing power and developing large sites."
Jo also emphasized the benefits of AI data center construction for local economic activity and specialist workforce development. According to Jo, about 900 people currently work at the Ulsan AI data center construction site.
SK Group is working with Ulsan College to run the SCALA training program, which is training about 120 AI specialists. The aim is to meet the growing demand for specialist personnel as data centers expand in the future.
"It is not easy to secure specialist personnel locally who can directly operate a data center," Jo explained. "Data centers require a large workforce during construction but relatively few highly skilled workers during operation, making local specialist training important."
"When AI data centers expand, the benefits extend beyond the data centers themselves to the entire related industrial ecosystem," Jo said. "SK Group views AI data centers as a new export industry and a strategically important national security asset."
Original Korean article: inews24.com
English brief by Korea Wind Intelligence, koreawindintel.com
