SK Telecom laid out a blueprint to open, in phases starting in 2029, an artificial intelligence (AI) data center in Korea with a capacity of 5 gigawatts (GW), and to build up to 15 GW of AI data centers by 2035. The plan aims to make Korea an "Asia AI hub," but calculations show that if capacity is secured as planned and a 15 GW data center runs at full tilt year-round, the annual electricity bill alone could reach the 24 trillion won range.
Of course, the 24 trillion won range does not mean a confirmed expense for SK Telecom. The actual electricity bill can vary depending on utilization rate, power usage effectiveness (PUE), contracted demand, tariff category, seasonal and time-of-use rates, power purchase contract structure, and how expenses are shared with customers. Still, the figure serves as an indicator of how sensitive power costs could be at the 15 GW level.
◇ If 15 GW runs full-time, 131.4 TWh a year… converted electricity bill in the 24 trillion won range
On the 8th, based on the current average industrial power tariff, ChosunBiz made a simple calculation of the power needed and the electricity bill if SK Telecom's 15 GW AI data center plan materializes, finding that 131.4 terawatt-hours (TWh) of power would be required annually. The figure assumes a 15 GW facility running 24 hours a day, 365 days a year.
Applying 185.5 won per kilowatt-hour (kWh), cited as the annual average unit price for industrial power, the annual electricity bill is estimated at about 24.4 trillion won. That is about 1.4 times SK Telecom's 2025 consolidation-based revenue of 17.0992 trillion won.
Even looking only at the 5 GW capacity targeted to start running in 2029, the burden is not small. If we simply assume a 5 GW facility operates all year, annual power consumption is 43.8 TWh. Applying the same tariff, the annual electricity bill comes to about 8.1 trillion won. That is close to half of SK Telecom's 2025 consolidation revenue.
◇ Power supply and tariff structure still under review
The problem is that the plan to secure power has not yet been fleshed out. SK Telecom said it "will comprehensively review" site selection and securing power supply. While the company has disclosed a goal for a supersized AI data center, the power grid interconnection, long-term power procurement, and expense-sharing structure to support it remain under review.
An AI data center does not stand on site and customer acquisition alone. Massive power is needed to run high-performance semiconductors such as graphics processing units (GPUs) at scale. Cooling power is also required to lower the heat generated by servers. If power supply is unstable or electricity expenses grow, both the utilization rate and profitability of the data center can be affected.
SK Telecom recognizes these risks. In its Climate Information Disclosure Report 2025-2026, published on the 30th of last month, "rising power demand" and "higher electricity rates" stemming from the expansion of AI data centers are listed as key business risks. The company effectively acknowledged that while building AI data centers is a growth opportunity, it also carries the risk of increased power use and higher expenses.
AI data centers are quintessential power-hungry facilities that must run high-performance semiconductors such as GPUs and cooling systems at scale. Even if sites and customers are secured, the business case can wobble without a stable power supply chain and an electricity expense structure to back it up. An industry official said, "Electricity bills account for 40% to 60% of operating expenses at AI data centers," and added, "If the government prepares various support measures for power supply, its megaprojects could proceed with more speed."
◇ The key variable for profitability is "power"
That said, the 24 trillion won range does not represent the actual electricity bill SK Telecom will pay. Actual electricity costs can vary widely depending on contracted demand, tariff category, seasonal and time-of-use rates, basic charges, maximum demand, utilization rate, cooling efficiency, PUE, and the power purchase contract structure.
Who bears the expense could also change depending on whether SK Telecom operates the data center directly, signs long-term lease deals with global big tech, or uses a separate joint venture or project financing (PF) structure. If the contract passes part of the electricity bill on to customers, SK Telecom may not shoulder all power expenses. Therefore, the 24 trillion won range should be seen not as a confirmed expense, but as a simple conversion indicating the power cost burden and profitability sensitivity inherent in a 15 GW scale.
Even so, power risk cannot be taken lightly. The competitiveness of AI data centers is no longer decided by securing GPUs alone. That is because the business case hinges on how quickly they can connect to the grid, whether they can secure stable power over the long term, how much they can improve cooling efficiency, and how they can share rising electricity costs with customers.
Because SK Telecom presented a 15 GW blueprint before detailing its power procurement plan, the market could raise concerns that "the announcement jumped the gun." Once built, AI data centers are infrastructure that consumes large amounts of power over long periods. Beyond initial capital spending, operators must weigh operating expenses, grid interconnection costs, cooling costs, and electricity price volatility.
In the end, SK Telecom's 15 GW AI data center plan is more likely to be tested not by "how big it is," but by "what kind of power it secures, at what price, and how reliably."
Chae Hyo-geun, vice chair of the Korea Data Center Council, said, "The AI hub vision is a growth strategy, but an AI data center without a power plan can become a cost risk," and added, "If SK Telecom is to make its 15 GW blueprint a reality, it must first present a power procurement plan that can justify the simple conversion figure of an annual electricity bill in the 24 trillion won range, in addition to sites and customers." He added, "Small modular reactors (SMRs) could be an alternative."