The challenge of supplying large-scale power to artificial intelligence (AI) data centers has emerged as a key issue in energy policy. While liquefied natural gas (LNG) power generation can respond quickly, it conflicts with carbon neutrality goals. Nuclear power and renewable energy each face limitations in construction time and intermittency, respectively.
Experts suggest that rather than relying on a specific power source, a combination of energy sources tailored to regional conditions should be established, alongside efforts to distribute data centers and improve power efficiency.
◆Projected 2040 Data Center Demand Equivalent to 5-6 Large Nuclear Reactors
According to relevant authorities, the Climate Energy Environment Ministry recently revised its 12th Basic Plan for Power Supply and Demand, forecasting maximum power demand to reach 165.0 GW by 2040. This represents an increase of 26.8 GW compared to previous estimates.
It is anticipated that an additional 20.6 GW will arise from advanced industries such as semiconductors, with 7.9 GW expected from data centers alone. This additional demand from data centers corresponds to the capacity of 5 to 6 large nuclear reactors, each with a capacity of 1.4 GW.
Particularly, the power demand from AI data centers is expected to surge. A report from the Korea Energy Economics Institute predicts that domestic data center electricity consumption will rise from 5 TWh in 2023 to 65.4 TWh by 2060, and could reach 134.8 TWh under an accelerated AI scenario.
Improving efficiency to reduce demand is also a major task. The institute assumes that the average Power Usage Effectiveness (PUE) of domestic data centers in 2023 is at the global average of 1.58. If this can be reduced to 1.2 by 2030 and below 1.1 in the mid-2030s, data center electricity consumption could decrease by approximately 16.5% to 17.8%.
However, a significant issue is that most domestic data centers are located far from power sources. As of 2024, there are 165 data centers in the country, with 60.4% concentrated in the metropolitan area. The share of private data centers in the metropolitan area reaches 75.3%, increasing the burden on the local power grid.
Data centers must operate continuously, and the power load can change rapidly during AI training and inference processes. This necessitates a power configuration that can ensure stable supply while meeting carbon neutrality goals.
◆LNG, Nuclear, and Renewable Energy Have Clear Advantages and Disadvantages; Regional Combinations Needed
The advantages and disadvantages of each power source are distinct. LNG power generation has a shorter construction time than nuclear power and can be installed near demand centers, making it capable of responding to rapid load changes in AI data centers, except for battery storage. It is also possible for data centers to build their own LNG combined heat and power plants to secure necessary electricity.
However, expanding LNG power generation may conflict with carbon neutrality policies. Fluctuations in international fuel prices can significantly affect generation costs, and as self-generation by data centers increases, private imports of natural gas may rise more quickly than expected. While it can serve as a short-term supply alternative, its limitations as a long-term primary power source are evident.
Nuclear power has the advantage of providing large-scale, carbon-free electricity reliably. Placing data centers near existing nuclear plants on the east coast can reduce the burden of transmitting power to the metropolitan area. However, new nuclear plants take a long time to build and face challenges related to community acceptance.
Small modular reactors (SMRs) can utilize existing nuclear or decommissioned coal power plant sites, but the timeline for commercialization, construction costs, and related regulations remain uncertain. It is expected to take considerable time for them to become a primary power source for data centers.
Renewable energy aligns with carbon neutrality goals but cannot supply power to data centers alone due to variability in generation based on weather conditions. A hybrid supply system combining solar, wind, and energy storage systems (ESS) based on the existing power grid needs to be established.
In the short term, combining renewable energy and energy storage systems (BESS) in the Honam and Yeongnam regions is seen as the most practical approach. However, it is suggested that existing power sources such as nuclear and LNG must also be utilized to ensure stable power supply to data centers.
Kim Cheol-hyun, a senior researcher at the Korea Energy Economics Institute, noted, "The increase in data centers is likely to lead to a significant rise in commercial natural gas imports, potentially increasing private LNG imports faster than the government anticipates. The government needs to expedite improvements to the direct import system and strengthen the stability of natural gas supply."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.