| Title |
A Cost-Optimal Electricity Procurement Portfolio for AI Data Centers in South Korea |
| Authors |
허강혁(Kang-Hyeok Heo) ; 최어진(Eo-Jin Choi) ; 김승완(Seung-Wan Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.4.765 |
| Keywords |
AI data center; Optimal power demand portfolio; Electricity market |
| Abstract |
The rapid expansion of artificial intelligence (AI) technology has led to a significant increase in data center power demand. Since data centers consume large amounts of power, electricity procurement costs account for a significant portion of total operating costs. Accordingly, interest in minimizing total electricity costs is expected to grow through the joint consideration of procurement options and load management strategies. Nevertheless, research on such strategies remains limited. To fill this gap, this study develops an optimization model for minimum-cost annual electricity procurement for large-scale AI data centers in Korea, reflecting domestic tariffs and settlement rules. The candidate procurement options include Korea electric power corporation (KEPCO) selective rate plans, the power market direct purchase system, and renewable energy power purchase agreements (PPAs). The model co-optimizes battery energy storage operation and time-shifting of AI training loads. Hourly demand scenarios are generated from capacity, utilization, inference/training ratio, and state-transition characteristics, while renewable energy supply is represented by regional hourly solar generation scenarios. Case studies show that average procurement cost declines as contracted PPA capacity increases, and that combining KEPCO Choice 3 with a PPA yields the lowest cost under the tested conditions. In this case, time-shifting of the training load reallocates consumption from high-rate peak hours to off-peak periods, enabling effective peak-cost avoidance. The proposed framework provides quantitative decision support for AI data centers within Korea's institutional context. |