[Edaily Reporter Kim Hyun-ah ] #SK Group has reaffirmed its goal of building a total of 15 gigawatts (GW) of artificial intelligence (AI) data centers in South Korea by 2035. The cost of building the first 5 GW is estimated at $350 billion (approximately 464 trillion won). The group is discussing funding with global Big Tech companies and investors, and the key challenge is securing actual customer demand to turn this large-scale investment into a viable business.

Jeong Seok-geun SKTelecom(017670), Head of the AI CIC and CEO of SK Hyper, recently stated in an interview with Bloomberg TV’s “Bloomberg Tech Asia” that the benchmark for competition in AI infrastructure is shifting from the number of graphics processing units (GPUs) to token costs and generation efficiency.

CEO Jeong said, “Until last year, it was important who secured the most GPUs, but now it is becoming more important who can achieve the lowest token cost and the highest token generation efficiency.”

A token is the basic unit through which an AI model processes and generates information. The more tokens generated for the same cost, the more efficiently AI services can be provided. He explained that competitiveness can only be enhanced by optimizing not only GPU capacity but also data centers, power supply, memory, semiconductors, and communication networks.

SK Group’s strategy is to pool the capabilities of its affiliates to respond to the competition on token costs. The plan is to enhance AI computing efficiency by integrating SK hynix’s High Bandwidth Memory (HBM), the energy businesses of SK Innovation and SK E&S, and SKTelecom’s telecommunications and cloud infrastructure.
Estimated Cost

of First 5 GW: 464 Trillion Won… Securing Investment Is Key

SK Hyper plans to build AI data centers totaling 15 GW by 2035. It estimates that the construction

of

the first

5 GW

, including servers and equipment, will require $350 billion.

CEO Jeong stated that the company is discussing funding options with potential partners, including global Big Tech firms, hyperscalers, and financial investors. The $350 billion figure represents an estimated cost for the first 5GW phase and does not reflect funds that have already been secured or investments that have been finalized.

SK Hyper’s strategy is to proceed with the project in phases while assessing customer demand, rather than finalizing the investment scale upfront. CEO Jeong explained that demand from AI model developers and service providers is growing, and the company will adjust its plans according to market conditions and manage investment risks.

From Korea’s 1GW Market to 5GW… Targeting Demand in Japan and Taiwan

For large-scale infrastructure development, securing power and land is crucial, but so is attracting actual users. CEO Jeong noted that South Korea’s existing data center capacity stands at approximately 1GW, adding that to build 5GW of infrastructure, the company must secure overseas customers in addition to meeting domestic AI demand.

SK Hyper is exploring ways to attract global companies’ AI training and inference infrastructure to Korea, targeting Asian markets such as Japan and Taiwan. The vision is to leverage KoreaElectricPower’s power supply and location advantages to grow the country into an Asian hub for AI computing. CEO Jeong identified securing global demand and raising investment capital as the key challenges for the first 5GW project.

He also views the trend of countries seeking to secure their own AI models and infrastructure as a business opportunity. He explained that to reduce dependence on specific overseas AI models, countries need not only their own models but also the data centers and computing infrastructure to operate them.

Following the broadcast, CEO Jeong also emphasized on Facebook the trend where token costs are becoming more important than GPU holdings, as well as the potential for South Korea to become an Asian AI computing hub, starting with the first 5GW project. The plan is to integrate SK Group’s capabilities in semiconductors, energy, and telecommunications into the AI infrastructure business.

To achieve the 15GW target by 2035, the company must secure large-scale investment funds and acquire global customers. The success or failure of the business is expected to hinge on whether it can translate growing AI demand into actual data center operations and stable revenue.