[K-Industry Great Leap] ④AI Data Centers: Securing Power Grid, Domestic NPU, and Revenue Models Is Key

18.4GW Target by 2035

Expanding Transmission Networks and Power Sources

"Securing Customers to Maximize GPU Utilization is Essential"

To achieve the government’s goal of building AI data centers (DC) with a total capacity of 18.4 GW by 2035, resolving power supply issues, promoting localization of components, and securing viable business models have been identified as the top priorities.

On June 30, experts emphasized that securing electricity is the fundamental prerequisite for advancing the AI DC initiative. They advised that, in addition to expanding power transmission networks, policy support should be provided to enable the use of various energy sources—such as renewable energy, liquefied natural gas (LNG), and small modular reactors (SMR)—to broaden the foundation for electricity supply in the mid- to long-term.

Professor Han Sekyung of the Department of Electrical Engineering at Kyungpook National University pointed out, "In the Honam region, there is a lack of transmission networks and substations, making it difficult to link with power or renewable energy sources. Data centers require substantial electricity to operate, but as most power generation is concentrated on the East Coast, it’s not easy to transmit electricity through the grid." Professor Han added, "Because data centers run 24/7 and their output fluctuates according to AI usage, the timing of power supply from renewables may not align. As a result, energy storage systems (ESS) are a structural necessity, but they come with cost issues."

A data center industry insider also noted, "AI DCs require much more electricity than conventional data centers—10 GW is equivalent to the power consumption of around 10 nuclear power plants. Currently, even domestic data centers are experiencing delays in obtaining permits due to power shortages, so ensuring stable electricity supply must be the top priority."

Industry stakeholders especially agree that securing a viable business model is crucial for the successful operation of AI DCs. Shin Jungkyu, CEO of Rablup, stressed during a discussion at the 'National Report Meeting on the Three Major Mega Projects for Korea’s Big Leap' held the previous day that optimizing AI DC investment is necessary. He explained, "For example, if a data center built at a cost of 300 billion won remains idle for just one day, it incurs 1 billion won in depreciation. Therefore, we must coordinate the entire pipeline—including the speed of energy supply, the distribution of domestically produced neural processing units (NPUs), and service business models—to maximize capital efficiency in line with market demand." He further noted that, to ensure the sustainability of AI DCs, securing clients who can maximize the utilization rate of the substantial investment in graphics processing units (GPUs) is paramount.

Another industry representative commented, "In the United States, big tech companies generate enormous demand for AI, while in the Middle East, sovereign wealth funds and global AI companies are attracted simultaneously. Korea also needs to continuously create demand by accelerating the adoption of AI services, AI startups, and public AI initiatives, so that AI DC investments can be recovered."

It is also important to resolve regional conflicts and establish a foundation for mutual growth with local communities. Professor Han observed, "Since data centers are not manufacturing businesses that can hire local talent, the expected employment effect is lower than anticipated. Rising electricity rates and water shortages can also trigger regional conflicts, so resolving these issues with local residents is a prerequisite."

Professor Choi Jaesik of the KAIST Graduate School of AI emphasized, "While data centers are built in each country, there is a significant gap with U.S. big tech companies in terms of technology for installing, operating, and recovering equipment. Software and service offerings at data centers also need to be globally competitive."

Localization of components is another major challenge. Deputy Prime Minister and Minister of Science and ICT Paik Kyunghun stated at the report meeting the previous day, "Domestic NPU companies have succeeded in commercializing AI chips after seven to eight years of research and development, so I ask that many companies show interest and adopt these technologies." An industry official added, "Operating an AI data center requires the comprehensive integration of advanced solutions—including AI chips and memory, energy, next-generation cooling systems, and network operations. Therefore, systematic government policy support is needed to promote the joint growth of all companies within the domestic value chain."

Step-by-step Construction of SKT and Naver AIDC... Core Hubs for AI Operations

Deputy Prime Minister and Minister of Science and ICT Paik Kyunghun and SK Group Chairman Chey Tae-won are having a conversation at the National Report Meeting on the Three Major Mega Projects presided over by President Lee Jae-myung at the Blue House on June 29, 2026. Photo by Blue House Press Corps, Yonhap News Agency

With the government presenting AIDC as one of the three major mega projects for Korea’s leap forward, private companies such as SK, GS, and Naver are planning to invest over 1,000 trillion won by 2035 to build AI DCs with a total capacity of 18.4 GW.

Deputy Prime Minister Paik stated, "We plan to invest 550 trillion won by 2029 to build data centers, which serve as the heart of AI, with a capacity of 8.4 GW, and then invest an additional 10 GW by 2035, bringing the total to 18.4 GW." One gigawatt is enough electricity to supply about 1 million households. The total investment required to build AI DCs with a capacity of 18.4 GW by 2035 will exceed 1,000 trillion won. Led by private companies, the plan is to move away from concentration in the Seoul metropolitan area, distribute construction across different regions, and contribute to regional revitalization.

The first phase of AI DC construction will be carried out by SK (5 GW), GS (2.4 GW), and Naver (1 GW), while the second phase will be led by SK (10 GW). SK Telecom will select construction sites based on various factors, including linking the project with national and local government efforts for balanced regional development, securing land, power supply, and anchor tenants. Funding will come from a mix of SK’s own investment, strategic partner investments, long-term contracts with global clients, and project financing. Diverse financial and investment structures are possible for each project, and the specific investment amounts will be determined as partnerships, equity structures, and contract terms are finalized for each project.

The first-phase 5 GW project being built by SKT includes the Ulsan AI DC, which is being developed in partnership with Amazon Web Services (AWS) with the goal of beginning operation in the second half of 2027, and a collaboration project with Nvidia. An SKT representative said, "The ongoing global partnerships under discussion will become a critical pillar of our overall AI infrastructure roadmap. Moving forward, we will establish strategic partnerships to co-plan and implement AI infrastructure optimized for the specialized needs of client companies, leading the market as a pioneering operator."

Naver plans to build a 1 GW capacity DC on additional sites including Sejong by 2029, utilizing its own large-scale AI models and cloud operation experience to serve as a key hub for AI research and service operations. Through these DCs, Naver will advance its generative AI and agentic AI services, and further expand AI search, cloud, and enterprise AI solutions.

GS will construct a 2.4 GW AI DC campus in Bukpyeong 2nd Industrial Complex in Donghae, Gangwon Province. It will be the largest in Asia, with a total investment of 30 trillion won. The Ministry of Science and ICT expects that once the first phase of the AI DC project is completed, Korea will become the largest AI infrastructure hub in the Asia-Pacific region by 2030.

Meanwhile, Deputy Prime Minister Paik also set a goal for Korea to become a global leader in physical AI. He stated, "The next three years are the golden time," and pledged to designate physical AI as a national strategic industry. The plan is to focus on securing data and, within three years, build a universal physical AI foundation model based on a 'world model' capable of understanding the real world and predicting actions. Based on this, Korea will develop specialized models for each field and apply them in practice, while also promoting localization throughout the entire physical AI stack, including robots, universal physical AI models, world models, and network security.

This content was produced with the assistance of AI translation services.