The focus of the artificial intelligence (AI) industry is shifting from data centers to 'AI factories.' Unlike traditional data centers that merely store data, AI data centers (AIDC) serve as critical infrastructure that performs large-scale computations for AI model training and inference, producing new 'intelligence' and tokens.
Lee Yong-tak, Vice President of AI Policy and head of the AI Data Center Integration Promotion Team at SK Telecom, explained during a keynote speech at the 18th Good Growth, Good Jobs Global Forum (2026 GGGF) held on September 3 in Seoul, "While traditional Internet data centers (IDC) act as data storage, AI data centers function as AI factories, performing learning and inference through large-scale AI computations and providing the resulting intelligence and tokens externally."
Lee emphasized that the importance of AI data centers is growing beyond mere industrial infrastructure to become assets for national economic security. By securing infrastructure capable of developing and operating AI models and services domestically, the outflow of national data and AI computations overseas can be reduced, contributing to the establishment of 'AI sovereignty.'
The competition to attract global tech giants' AI data centers is intensifying, particularly in the Asia-Pacific (APAC) region, where countries are leveraging their unique strengths to attract investment.
"Japan is actively supporting the attraction of big tech data centers, while Malaysia highlights its relatively low labor and construction costs as advantages," Lee noted. "Australia boasts competitive power supply and ample land, leading to a competitive landscape among countries based on their respective strengths in attracting AI data centers."
South Korea's strengths include stable power supply conditions and semiconductor competitiveness. The country has established a supply chain for key semiconductors such as GPUs and memory, enhancing its competitiveness as a destination for global tech investments in AI infrastructure.
However, building large-scale AI data centers requires substantial power and funding. According to Lee, as of last year, there were 161 data centers and AI data centers operating in South Korea, with a total power capacity of approximately 2.3GW. Currently, 150 entities have applied to Korea Electric Power Corporation for power supply for new data centers, with a total demand of about 9.4GW. This indicates a significant potential increase in power demand as AI data centers grow larger.
Investment costs are also considerable. Lee estimated that the investment for a 100MW AI data center being constructed by SK in Ulsan was around 7 trillion won last year, and considering recent price increases for GPUs and related equipment such as power, cooling, and air conditioning, the current estimate is about 9 trillion won. This suggests that establishing a 1GW AI data center could require tens of trillions of won in funding.
As a result, large-scale AI data centers are likely to evolve beyond facilities used solely by specific companies to provide AI computation services externally, according to Lee. Increasing the scale of data centers enhances the efficiency of infrastructure like GPUs and reduces computation costs, making investment and collaboration with global tech giants crucial.
In June, SK Group announced plans to gradually open a 5GW AI data center starting in 2029. Lee emphasized that the 5GW will not be built all at once but will be expanded in phases.
Lee highlighted that now is the golden time for investment in AI data centers. Even if GPUs are secured, the lack of physical data centers for installation and operation means that proactive investment is necessary, as building a data center takes about 30 months. He explained that since AI computations are processed by connecting multiple GPUs in parallel, clustering data centers in specific regions can enhance infrastructure efficiency.
SK is pursuing the construction of AI data centers centered around five regions, considering power supply, land acquisition, and regional balanced development. Lee stated, "The reason for announcing five regions includes regional balanced development, as we considered both power supply and regional balance." While the metropolitan area faces constraints in securing large-scale power, non-metropolitan areas have relatively more opportunities to secure power systems.
To secure the massive investment needed, collaboration with global tech giants will be essential. SK Group will not bear the entire cost of the 5GW construction but will share investment and costs with external tech companies. Lee remarked, "SK cannot do everything alone; we will collaborate with external tech giants to share costs, while SK will focus on bringing in demand and planning investments and data center construction." He also indicated that SK Group itself could invest around 20 trillion won.
To expand the AI data center industry, tax incentives should also be increased. Currently, AI data centers are recognized as facilities for commercializing national strategic technologies and can benefit from investment tax credits, but the system is primarily designed around self-consumption facilities.
He stated, "AI data centers incur substantial costs and often provide the computational power produced externally. Therefore, a certain level of tax credits is necessary not only for self-consumption facilities but also for externally provided data centers."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.