At the National Assembly Members' Office Building on the 9th, Professor Park Jong-bae of Konkuk University's Department of Electrical and Electronic Engineering (far left), who delivers the keynote at the forum on practical power supply plans for AI data centers, gives a presentation. /Courtesy of Shim Min-kwan

"With power demand concentrated in the greater Seoul area and slow expansion of the transmission grid intersecting, approvals for AI data centers (AI DC) are being structurally blocked. On the premise of dispersing outside the capital region, regulations on power transaction should be eased, and the rate and grid usage systems should be redesigned on a regional basis."

At the "practical power supply plans for AI data centers" forum held at the National Assembly Members' Office Building in Yeouido, Seoul, on the 9th, Park Jong-bae, a professor in the Department of Electrical and Electronics Engineering at Konkuk University, stated accordingly.

According to Park, Korea depends on overseas imports for 93.8% of its energy, and its energy self-sufficiency remains at around 20%. While the capital region's share of power demand has grown to 40%, its share of power generation is only 24.3%, and the share of renewable energy remains around 8.0%, entrenching an imbalance between supply and demand.

The transmission grid was identified as the key bottleneck. Park said, "Even when the standard construction period is set at nine years, it often takes 13 to 15 years due to declining public acceptance and permitting delays." There are cases delayed more than 150 months, such as the Bukdangjin–Shintangjeong section. The shortage of transmission lines simultaneously increases generation and transmission constraints, undermining the stability and economic efficiency of the power industry.

The spread of AI was projected to worsen this bottleneck. Park said, "Global data center power consumption is expected to surge 2.3 times from 415 TWh in 2024 to 945 TWh in 2030," adding, "In Korea as well, under the 11th Basic Plan for Long-term Electricity Supply and Demand, the maximum power for data centers is expected to increase 5.2 times from 1.2 GW in 2025 to 6.2 GW in 2038."

The problem is that data center demand is excessively concentrated in the capital region. Park said, "About 73% of domestic data centers are concentrated in the capital region, and more than 80% of new applications also head there, but new approvals are effectively blocked in the region due to insufficient transmission capacity." He cited figures showing that only four of 195 recent new construction applications in the capital region passed the power system impact assessment. The power system impact assessment is a procedure that examines the impact of large power demand facilities on system stability to determine connection feasibility and mitigation measures.

As a solution, Park proposed "diverting cars onto less congested national highways." In the short term, he said, incentives to move outside the capital region should be strengthened, with a regional strategy to attract RE100-type AI data centers based on renewable energy to the Honam region, and large-scale, low-cost AI data centers leveraging the price competitiveness of low-carbon sources such as nuclear and LNG to the Yeongnam region.

Cho Dae-geun, a senior adviser at Lee & Ko who also gave a keynote presentation, said, "In the AI era, competitiveness is determined by the speed of securing power," adding, "The moment AI data centers become 'prisoners of the power grid,' industrial expansion stops." AI data centers consume more than 10 times the power of general data centers, and while demand is soaring, supply is structurally slow to catch up, he noted.

Cho explained, "A center can be built in about 10 years, but expanding the transmission grid and constructing power plants entail time lags of five to more than 10 years." He viewed that even after large-scale training ends, the energy demand that accumulates in the service phase of "inference" can raise power consumption over the long term, making the power bottleneck not a temporary variable but a structural risk.

He also pointed to the speed of institutional responses as a problem. Cho said, "In a KEPCO-centered regulatory transaction structure, the options for consumers to 'buy and use' power are limited, and the less capacity the grid has, the more likely new investment is to be delayed at the permitting threshold," adding, "If AI infrastructure is tied up in power system review procedures, the administrative period can become longer than the construction period."

As a Korean-style solution, Cho presented the keyword of "expanding consumers' autonomy in power procurement." He said, "In hybrid power purchase agreements (PPAs) that mix renewable energy and carbon-free sources, we should lift restrictions on eligible parties, scale, and duration so corporations can design their own power portfolios." He also proposed encouraging onsite direct transaction that consumes power right next to power plants and expanding the scope of dedicated line construction to allow private networks. The logic is to open supply routes that do not pass through the grid to minimize the impact of transmission bottlenecks.

※ This article has been translated by AI. Share your feedback here.