Chey Tae-won, chairman of SK Group, has unveiled a blueprint to build a 15-gigawatt (GW) class artificial intelligence (AI) data center in Korea, but market attention is focused more on financing than on technology. SK Telecom recently joined hands with Nvidia to launch "AI Factory," a next-generation data center specialized for AI computation, in Korea in 2027, and said it would gradually expand it into GW-class AI infrastructure. Collaboration with Nvidia can bolster graphics processing unit (GPU) and system design capabilities, but how to secure investment approaching 1,000 trillion won remains an open question.
On the 29th, Chey said at the "National Briefing on the Three Mega Projects for Korea's Great Leap" held at the state guesthouse of the presidential office that "we are planning about 1,000 trillion won for the AI data center project." Chey said that over the next 10 years, SK would continue to invest an average of more than 100 trillion won annually in Korea, and that by building AI data centers on a large scale, it would create a foundation to "export intelligence, not products."
The plan Chey presented is to build a total of 15 GW of AI data centers with SK Telecom at the core. First, 5 GW would be divided nationwide in 0.5–1 GW units, then an additional 10 GW would follow depending on power, sites, and water conditions.
◇ Seven to eight times all data centers in Korea… the weight of the 15 GW figure
In SK Telecom's collaboration with Nvidia, 2027 is not when a GW-class AI Factory immediately goes live. 2027 is when the first AI Factory begins operating in Korea, and the GW-class infrastructure is a target for phased expansion after that. One GW is 1,000 MW. Facilities in the tens of MW and those at 1 GW are completely different in the scale of required capital, power, cooling, and customer contracts.
How large 15 GW is becomes clearer when compared with the current total capacity of data centers in Korea. According to data from the Energy Information Statistics Center of the Korea Energy Economics Institute (KEEI), the Korea Data Center Council estimated the total receiving capacity of domestic data centers at about 1,913 MW as of 2023. Based on KEPCO's status of power supply to data centers, the contracted power of 150 data centers in Korea at the end of 2023 was 1,986 MW. That means the current total capacity of data centers in Korea can be viewed as about 1.9–2.0 GW.
Fifteen GW is 15,000 MW. That is equivalent to seven to eight times the total receiving capacity of data centers currently operating in Korea.
◇ Tens of trillions of won needed to build 1 GW… financing is the first hurdle
The biggest variable for AI data centers is money. The industry believes that building a single 1 GW-class AI data center could require as much as around 70 trillion won. A simple conversion to 15 GW puts the scale at 900 trillion–1,050 trillion won.
One direct basis for this estimate is a remark by Foxconn Chairman Liu Young-way. In a lecture hosted by the National Association of Industry and Commerce of Taiwan on the 18th, Liu said that building a 1 GW-class AI data center could require up to $47 billion. This includes not only land and buildings, but also graphics processing units (GPUs), AI server racks, network equipment, power facilities, and cooling infrastructure. AI data centers have higher power density and more expensive equipment than conventional data centers.
The Ulsan AI data center already being pursued by SK Group is also the first test bed for this vision. SK Group and Amazon Web Services (AWS) have presented plans to build a 100 MW-class AI data center in the Ulsan Mipo National Industrial Complex and expand it to a 1 GW-class facility later. The facility is first slated to go live at 41 MW in Nov. 2027 and expand to 103 MW in Feb. 2029.
Even so, the financing structure is not simple. SK Telecom has been seeking to sell up to a 49% equity stake in the Ulsan AI data center, and a consortium of KKR and IMM Investment–Stonebridge Capital has reportedly been selected as the preferred bidder. The transaction reportedly involves selling up to 49% for the mid-2 trillion won range, with KKR acquiring 29% and the IMM Investment–Stonebridge Capital consortium acquiring the remainder.
◇ Power and securing long-term customers are also key
An AI Factory is a next-generation data center that handles AI training and inference based on large-scale GPUs. Securing GPUs alone does not make the business work. It is closer to a "factory" that feeds in power and data to produce AI computation outputs. For a factory to run, you need not only equipment but also electricity, sites, cooling, and customers.
Power is an even bigger bottleneck. If 15 GW is based on server load and we simply assume 100% operation all year, the power needed just to run servers would reach 131.4 terawatt-hours (TWh) annually. On a 1 GW basis, that is 8.76 TWh a year. That corresponds to about 24% of Korea Electric Power Corporation's total 2025 power sales of 549.4 TWh.
Given that nuclear power generation in Korea in 2024 was 188.8 TWh, or 31.7% of total generation, the server power demand alone for a 15 GW AI data center would approach 70% of the annual nuclear generation in the country. Adding cooling facilities, power conversion losses, reserve power, and network equipment usage would only increase the actual power burden.
Securing long-term customers is also important. AI data centers in the tens of trillions of won cannot be pursued by simply looking for customers after the fact. Long-term lease contracts with global big tech, large enterprises, and the public sector are needed to structure project financing. While AI compute demand is rising quickly, securing a customer base that can reliably absorb 15 GW of compute resources over a long period is a separate challenge.
Chae Hyo-geun, vice chairman of the Korea Data Center Council, said, "Cooperation with Nvidia is meaningful in terms of securing GPUs and system design," but added, "For the 15 GW AI Factory vision to become reality, execution is needed that ties together money, electricity, and demand beyond a technology alliance."
Kim Kyung-won, a distinguished professor in the Department of Business Administration at Sejong University, said, "Building a 15 GW AI Factory seems like a plan with limited feasibility. Above all, a solution for power supply should have preceded the plan's announcement," and added, "There is a risk that power needed by other industries could fall short, and it could also lead to a sharp increase in electricity rates in the future."