[Edaily Reporter Yun Junghoon ] As even global Big Tech companies face a shortage of artificial intelligence (AI) computing resources, South Korea has fully entered the race to build ultra-large-scale AI data centers (AIDCs). With the government presenting a blueprint to build 18.4 GW of AI infrastructure by 2035, SK Group has agreed to handle 15 GW—81% of the total—leading to assessments that the success of South Korea’s AI infrastructure competitiveness hinges on SK’s ability to execute.

According to foreign media reports on the 30th (local time), AI infrastructure bottlenecks are becoming a reality, with even Google—a leader in the global AI race—limiting its “Gemini” service supply to Meta due to a shortage of computing resources. Meta, too, continues to invest heavily in the race to secure AI data centers.

As the AI industry shifts from model competition to a race for computing power, AI data centers are emerging as critical infrastructure that determines a nation’s competitiveness.

The industry anticipates that as we move beyond generative AI into the era of Agent AI, the usage of “tokens”—the smallest unit processed by AI—will skyrocket. As the number of tokens increases, demand for GPUs, power, and data centers will also grow exponentially.

NVIDIA CEO Jensen Huang forecasts that global investment in AI infrastructure will reach $5.5 trillion (approximately 9,400 trillion won) over the next five years, and the Ministry of Science and ICT also views this as a “historic infrastructure transition” and is pushing forward with the construction of ultra-large-scale AI data centers.

Naver 1 GW, GS Holdings 2.4 GW… SK Leads with an “Overwhelming” 15 GW

Among domestic companies, #SK Group has unveiled the most aggressive investment plan.

SKTelecom(017670)Centered around SK, the group plans to build AIDCs with a capacity of 5 GW by 2029 and 15 GW by 2035. This accounts for the majority of the government’s 18.4 GW target.

Compared to Naver, which is partnering with NVIDIA to build an AI factory and aims for 1GW in the long term, and GS Holdings, which is pursuing 2.4GW, SK’s scale is overwhelming.

SK’s strategy goes beyond simply building data centers. By jointly developing the Ulsan AI data center with AWS and raising funds through equity sales to investors such as the KKR and IMM-StoneBridge consortium, the company is establishing a model that secures both foreign capital and global AI demand. The plan is to expand its AI infrastructure business with a focus on the Southeast Asian market.

The industry believes that rather than self-financing the 1,000 trillion won investment, SK is preparing a financial structure to actively attract investments from overseas institutional investors and global Big Tech companies such as Anthropic and NVIDIA.

At a public briefing on the three major mega-projects, SK Group Chairman Choi Tae-won stated, “An AI data center is not a facility for storing data, but a factory for producing intelligence,” adding, “We will gradually build 15 GW by 2035 to lay the foundation for South Korea’s national AI infrastructure and ‘token factories.’”

AI Data Centers Are ‘Token Factories’… Interlinked with Regional Power Grids

The mega-projects proposed by the government are not merely data center construction projects. The strategy is to distribute AI data centers across regions with abundant power and land—such as Chungcheong, Yeongnam, Honam, and Gangwon—to simultaneously grow the power grid and the AI industry.

SK is also mobilizing its entire group of affiliates to build a vertically integrated model that combines SK Eco Plant’s construction capabilities, SK hynix’s HBM, and the power and cooling technologies of its energy affiliates.

The industry forecasts that AI data centers will evolve beyond mere storage facilities to serve as “token factories” that produce tokens—the output of AI computations. To this end, SKTelecom plans to build a gigawatt (GW)-scale AI cloud in South Korea based on NVIDIA’s AI Factory platform (DSX) and begin operations at its first AI factory in 2027. Additionally, the company plans to join the “NVIDIA Cloud Partner (NCP)” program to reduce AI service operating costs and secure a competitive edge.

“Domestic Demand Is Still Sufficient… The Real Battle Is in Overseas Markets”

The current supply of domestic data centers for data storage is considered sufficient.

Kang Joong-hyup, Chairman of the Korea Data Center Association, stated, “There are currently almost no cases where services cannot be provided due to a lack of data center capacity in Korea.” This mega-project represents a strategic investment aimed at preempting future domestic AI token demand and global AI demand, rather than merely meeting current demand.

Private companies, including SK, report that demand for AI infrastructure is growing much faster than expected. The reason the government is building an 18.4 GW supply chain and companies like SK are investing over 100 trillion won annually for the next 10 years is that demand is rising at a significant pace due to the proliferation of AI agents, physical AI, and robots.

Kim Young-jun, Executive Director of Data Center Operations at Dow Data, predicted, “The data center market has now become a competition among nations to attract AI demand,” adding, “If South Korea can ensure efficiency in power supply and services as well as price competitiveness, it can take the lead as a hub for global Big Tech companies.”