"Many people say there is overinvestment in artificial intelligence (AI) and ask whether it is a bubble, but we have only just stepped into the entrance of a world made by AI."
Jeong Seok-geun, head of SK Hyper (head of the SK Telecom AI DC Integration Task Force), said this at SMARTCLOUD SHOW 2026, the country's largest tech conference, held on the morning of the 26th at the Westin Josun Hotel in Jung-gu, Seoul.
SK Hyper is a specialized corporate entity for AI data center business development that SK Telecom established last month. SK Hyper has drawn up a plan to build a 5 GW (gigawatt) AI data center by 2029 and expand it in stages to a total of 15 GW by 2035. Jeong is SK Hyper's inaugural chief.
Jeong said, "The mere emergence of coding agents has changed how software is developed, and that alone has significantly altered our daily lives," adding, "We will have very similar experiences across a great many fields going forward." He continued, "I have met several times with big tech CEOs, including Nvidia CEO Jensen Huang, to hear their views, and they think the same," and said, "AI is demanding an overwhelming level of investment, but compared to what lies ahead, we are only at the very beginning."
Regarding SK Hyper's ongoing business, Jeong said, "Our current goal is to operate 5 GW of AI data centers in Korea in 2029," adding, "By simple calculation, it is a 500 trillion won project, with 100 trillion won to construct buildings and 400 trillion won to install servers, and how to finance it and how to fill demand are major challenges."
Jeong defined data centers in the AI era as factories that produce intelligence. Buildings, graphics processing units (GPUs), large language models (LLMs), power, and user inputs are combined to ultimately generate "tokens." Jeong said, "In the past, having a good model and then securing GPUs were the sources of competitiveness, but now AI infrastructure is the biggest bottleneck." As data and algorithm capabilities are being standardized upward mainly in the United States and China, infrastructure has become the key.
Jeong said, "More than 90% of the world's AI infrastructure is currently concentrated in the United States, but it takes five to seven years to apply for grid connection there, the grid reinforcement expense is passed on to nearby residents' electricity bills, and resident backlash and regulations are growing," adding, "There is an opportunity for Korea." As Korea's strengths, he cited ▲ being a producer of memory semiconductors, which are core to AI ▲ stability, power, and cooling technologies accumulated from operating semiconductor fabs (factories) ▲ the potential to serve as a test bed for applying AI to new industries.
Jeong explained SK Hyper's business model in two ways. The first is co-location, which provides only buildings and power and network facilities, while big tech brings in servers directly. Stable revenue can be secured through 15- to 20-year long-term contracts. The second is an "AI factory," which includes servers and is rented out. The investment burden is heavy, but profitability is high, and it allows SK Hyper to lead optimization of servers, memory, and software.
Jeong said, "SK Group has all the elements needed for an AI factory within the group, including buildings, power, cooling, semiconductors, networks, and computing," highlighting synergies among affiliates. He also emphasized SK hynix's memory production capabilities as a differentiator. Jeong said, "Token output depends on the amount and speed of memory," adding, "Even without increasing GPUs, simply increasing memory produces more tokens."