SK Telecom plans a network of AI data centers in South Korea totalling up to 15 GW—a capacity roughly 10% of the country’s entire installed power generation—backed by a government that has designated hyperscale data centers as national infrastructure and a conglomerate that controls semiconductor, energy, and construction arms under one roof.
The first concrete test arrives in 2027, when SK Telecom’s initial “AI Factory” is set to launch at multi‑hundred‑megawatt scale. Whether global AI firms commit long‑term hosting contracts by 2028 will determine if the full 15 GW roadmap survives contact with reality—or shrinks to a more incremental, regional buildout.
South Korea is betting that the next decade’s AI spoils go not to the company that writes the smartest model, but to the one that controls the cheapest, most stable electricity at scale. It is a bet on concrete and cooling loops, not code. And on July 6 SK Telecom put a number on it: 15 gigawatts of new AI data center capacity, a figure that rewrites the geography of global compute.
The national telecom operator, backed by the sprawling SK Group and a government that has legislated data centers into strategic assets, intends to build clusters in the southeast and southwest of the country, starting with a 2‑GW‑plus hub in Ulsan. The first 5 GW are slated to come online in stages beginning in 2029. The ambition is not simply to add server racks. It is to turn the Korean peninsula into a rentable floor for the AI workloads that US and Chinese firms cannot host inside their own borders at the latency Asian users demand.
A 15‑gigawatt land grab for Asian compute
The electricity alone tells the story. South Korea’s total installed generation capacity stood at roughly 141 GW in 2024. The SK Telecom plan would bolt on an extra 10% to the national grid—an extraordinary claim on a system that already runs some of the world’s most power‑dense semiconductor fabs.
Each gigawatt of AI‑grade data center costs an estimated KRW 70 trillion to build, excluding land. The math is unforgiving: 15 GW would need funding measured in the quadrillion‑won range, a multi‑decade capital commitment that SK Telecom says it will stitch together from its own balance sheet, strategic partners, long‑term customer contracts, and project financing. None of those customer contracts are signed yet.
Jung Jai‑hun, President and CEO of SK Telecom, laid out the roadmap at the SK AI Summit in November 2025. He described the company’s ambition to become Korea’s primary AI infrastructure operator and to scale the initial Ulsan AI Data Center beyond 1 GW through partnerships with global tech firms. Lee Chang‑yang, the former Minister of Trade, Industry and Energy, has gone further, calling large‑scale data centers “national strategic facilities” critical to Korea’s digital exports.
The government had already acted on that logic. In September 2024, the Act on Promotion of Data Centers for Vitalization of the Digital Economy came into force, designating hyperscale facilities as national infrastructure and unlocking tax breaks, streamlined permitting, and land‑use incentives for projects exactly like the Ulsan cluster. SK Telecom’s pivot into AI infrastructure is accelerating, building on the IOWN AI Fund it launched in June 2026 with NTT and Chunghwa Telecom to back optical networking and cooling startups.
For a Western AI firm that needs to train a model on Asian user data, a data center in Ulsan offers latency under 30 milliseconds to Tokyo and Singapore—and it sidesteps the legal and geopolitical risks of hosting in mainland China. That is the practical payload behind the gigawatt numbers. But the International Energy Agency warns that rapid data center growth, particularly for AI, requires careful grid planning to avoid local capacity constraints, and South Korea’s own data center council has flagged that sites around Seoul already face grid‑connection bottlenecks, pushing new projects toward coastal industrial zones.
The ambition’s scale becomes plain when set against existing hubs.
The kilowatt geopolitics behind the cranes
Behind the announcement sits a simple observation: the global AI buildout is running into a structural power deficit. McKinsey & Company projects that AI and cloud demand will create a persistent shortage of high‑capacity, low‑latency data center infrastructure in major markets through 2030. The US alone faces a projected 15‑GW supply gap by then. In Asia, hyperscale capacity is concentrated in a handful of Chinese clusters and the tightly regulated island of Singapore, which allocates just 300 MW of additional IT load per year. Japan and India are scaling, but neither combines the domestic high‑bandwidth memory supply chain, stable nuclear‑backed power, and integrated group‑level engineering that SK Group can bring.
Yet the regulatory picture cuts both ways. South Korea subjects AI data centers to the Personal Information Protection Act and Network Act, which impose strict rules on cross‑border data transfers and security. That clarity can reassure Western cloud tenants accustomed to GDPR‑like frameworks, but it also means foreign operators cannot simply ship data abroad without navigating a compliance apparatus overseen by the Personal Information Protection Commission and Korea Communications Commission.
The first real test of the strategy arrives in 2027, when SK Telecom’s “AI Factory” is scheduled to enter commercial service. If it launches on time at multi‑hundred‑megawatt scale and secures GPU‑cluster leases from a major US or Asian cloud provider, the signal will be strong enough to pull forward follow‑on investment. If it slips materially, financiers will demand tighter phasing and proof of pre‑commitments before backing anything near 15 GW. The International Energy Agency’s own data shows that nuclear and LNG provided over 60% of South Korea’s electricity in 2023—a stable baseload that data centers prize—but it also warns that rapid growth in AI‑driven consumption requires grid upgrades that lag behind permit timelines.
South Korea is not betting on a single data center. It is betting on turning the entire industrial geography of the southeast—the shipyards, the LNG terminals, the memory‑fab clusters—into the floor that global AI stands on. The first concrete of that floor will be poured in 2027.
Beyond the headline
The bigger picture
South Korea’s bet on AI data centers shifts the focus of the global AI race from algorithms to physical compute, power and cooling capacity. Instead of trying to out‑compete US firms on frontier models, Seoul is positioning itself as the industrial backbone that everyone else must rent, turning electricity, industrial land and HBM supply chains into strategic assets comparable to oil terminals or semiconductor fabs.
The power behind it
Control over this buildout will not sit only with SK Telecom but with a triangle of ministries, grid operators and SK Group energy affiliates that ultimately decide where gigawatts of load can physically connect. Their siting, pricing and permitting choices will determine which regions become AI boomtowns and which Western partners are allowed into strategic clusters hosting sensitive national compute.
The reach
For Western cloud and AI companies, a Korean AI hub offers a way to serve East Asian users with lower geopolitical and sanctions risk than relying solely on mainland China or a single Southeast Asian city‑state. That diversification could reshape where large training runs and latency‑sensitive inference are hosted, altering global traffic patterns and weakening the incumbency of existing US‑centric hyperscale regions.
What Korea’s infrastructure gambit demands of the West
With the first commercial AI Factory due in 2027 and the 5‑GW initial phase targeting 2029, Western companies must decide whether to commit capital and workloads to a hub that is legislated as critical infrastructure but remains operationally untested.
- Western AI and cloud companies
Review South Korea’s National Strategy for Artificial Intelligence on the Ministry of Science and ICT website. The document signals how the government plans to allocate compute infrastructure and semiconductor resources—details that will determine whether your model‑training workloads can reliably reserve GW‑scale capacity at fixed power rates. Without that clarity, you risk committing to a cluster that cannot deliver steady baseload during a three‑month training run.
- Infrastructure investors
Monitor the International Energy Agency’s periodic commentary on data center electricity demand at iea.org. Korea’s nuclear‑and‑LNG grid is strong, but the IEA notes that local bottlenecks can emerge quickly. If grid expansion in the Gyeongsang region lags behind SK Telecom’s construction schedule, project‑finance terms will tighten sharply—and the difference between a 2030 revenue start and a 2033 delay is the difference between a financing that closes and one that does not.
- Supply chain and procurement managers
The Ulsan cluster sites AI hardware yards within kilometers of SK Hynix’s HBM production lines. That physical proximity can shorten lead times for GPU‑server integration and maintenance, but it also ties your hardware procurement to a single‑geography supply chain. Map your own total bill of materials against where Ulsan sources its cooling, fiber, and power‑distribution units before assuming a net resilience gain.
Explainer
- AI data center
- A facility purpose‑built for training and running large AI models using tens of thousands of GPUs or specialized accelerators. Unlike enterprise data centers, an AI center concentrates extreme power and cooling loads per rack, often requiring liquid cooling and direct grid co
RELATED STORIES
