A majority of artificial intelligence data centers in South Korea fall short of core operational requirements including power infrastructure capacity and site conditions, a new study has found.
The Korea Artificial Intelligence Industry Association has established a new set of evaluation criteria for AI data centers (AIDC) tailored to Korea's policy environment and published a research report titled "Establishing Evaluation Standards for Domestic AI Data Centers," applying the criteria to 42 major data centers nationwide.
The study carries significant meaning in that it goes beyond simply assessing the technical capabilities of individual data centers. It is the first to design an evaluation framework reflecting Korea's policy characteristics — including power grid structure, site conditions and permitting environment — and apply it to actual cases.
International standards currently used for data center evaluation are effective in measuring technical completeness and operational stability of facilities. However, they have limitations in adequately reflecting national policy environment factors such as grid capacity, site suitability and business feasibility. Starting from this awareness, the research built a multi-criteria evaluation framework that supplements existing technology-focused assessment systems by comprehensively considering power infrastructure capacity, site and risk factors, corporate reliability and operational capability, and AI infrastructure characteristics. The framework uses a single scoring system with a maximum of 205 points, designed to enable quantitative assessment of a data center's overall level by integrating various evaluation factors.
Evaluation results are classified into three grades — Green, Yellow and Red — based on the score ratio relative to the total. To enhance policy applicability, three tiers of criteria were presented according to strictness: Strict, Moderate and Flexible. When these criteria were applied to 42 major AI data center construction project plans in Korea, only six sites (14.3%) received a Green rating under the most stringent Strict standard. Under the Moderate standard, this expanded to 12 sites (28.6%), and under the Flexible standard, to 21 sites. Even when criteria were relaxed, a common finding was that a majority of projects require improvement in power infrastructure, site conditions and project execution.
As representative best-practice cases, Megadata's data center in Wonju, Gangwon Province, and LG Uplus's data center in Paju, Gyeonggi Province, received Green ratings across all three tiers. Megadata in Wonju scored consistently high marks for securing dual power supply routes, minimizing natural disaster risks and maintaining a stable construction schedule. LG Uplus in Paju demonstrated strengths in corporate reliability and technical certification, earning a well-balanced evaluation overall.
A particularly noteworthy point of the study is the structural gap between supply and demand surrounding AI data centers. According to consulting firm McKinsey, global AI data center demand is projected to reach 156 GW by 2030, with AI workloads accounting for 70% of total data center demand. However, current construction speeds are insufficient to meet this demand. In Korea as well, despite 290 applications for data center power usage being filed nationwide over approximately 11 months from August 2024 to June 2025, AI-dedicated infrastructure requiring high-density GPU integration and large-scale cooling remains critically insufficient.
The research team, led by Prof. Yoo Byung-jun of Seoul National University's AI Research Institute, stated, "AI data centers are not mere IT facilities but strategic assets that determine national digital sovereignty and industrial competitiveness." They added, "The core finding of this study is that even projects with high technical completeness cannot proceed to actual construction unless prerequisites such as grid capacity and site conditions are met." They emphasized, "To respond to rapidly increasing future demand, extraordinary policy support and regulatory easing at the government level, including preemptive expansion of power infrastructure, must be expedited."