Korea’s telecom titans double down on AI data centers

South Korea's leading telecom operators are embarking on an unprecedented AI infrastructure expansion, marked by ambitious multi-gigawatt scale data center projects. SK Telecom aims to develop 5 gigawatts of AI data center capacity beginning in 2029, with a long-term target of 15 gigawatts by 2035, while KT plans to increase its AI data center capacity from 152MW to 1.15GW by 2031 through a ₩5 trillion investment over five years. Meanwhile, LG Uplus is heavily investing over ₩1.3 trillion to build and expand its Paju AI data center near Seoul, targeting more than 600MW of supply capacity, positioning it as the largest AI data center hub in the Seoul metropolitan area. These phased, demand-validated expansions reflect a strategic approach to balancing capital deployment with power availability and site conditions across the industry.

To optimize AI workloads and enhance operational efficiency, South Korean telecoms are integrating heterogeneous AI chip architectures that combine Nvidia GPUs for large-scale training with domestically developed neural processing units for inference tasks. SK Telecom’s collaboration with Rebellions and Arm exemplifies this trend, deploying neural processing units alongside GPUs and developing server solutions that reduce power consumption and costs. Furthermore, innovative architectural advancements such as SKT’s Panmnesia platform leverage Compute Express Link (CXL) technology to dynamically pool CPUs, GPUs, and memory, significantly boosting resource utilization and addressing memory bottlenecks within AI data centers.

Operational excellence in cooling, power efficiency, and connectivity is emerging as a critical competitive differentiator in South Korea’s AI data center race. KT has pioneered direct-to-chip liquid cooling technology for GPUs, while LG Uplus applies a suite of group-wide technologies to enhance thermal management and power delivery at its Paju facility. Leveraging their extensive telecom networks, operators like KT emphasize data center interconnection (DCI) technologies to link multiple AI data centers via dedicated networks, enabling scalable infrastructure and robust disaster recovery capabilities. These innovations underscore a holistic approach that combines hardware, network, and environmental optimizations to sustain the massive energy demands of AI workloads.

The AI data center expansion is not only a technological endeavor but also a strategic financial and operational initiative. SK Telecom has established SK Hyper, a dedicated subsidiary with a capital injection of up to 750 billion won by 2030, to spearhead AI data center development while attracting strategic and financial investors to mitigate direct capital expenditure. KT’s strategy includes building a distributed AI compute network by interlinking central data centers, regional hubs, and edge nodes across more than 3,500 telecom central offices nationwide, creating approximately 25 AI data centers. These moves reflect a coordinated industry-wide push to scale AI infrastructure efficiently while innovating business models and operational frameworks to capture growing AI service revenues.

South Korea's leading telecom operators are aggressively innovating AI data center operations by prioritizing cooling efficiency, power stability, and network connectivity to manage escalating heat and energy demands. KT, for instance, leverages its nationwide telecom network and expertise to implement advanced data centre interconnection (DCI) technology and pioneered direct-to-chip liquid cooling in Korea, enabling scalable and high-speed AI infrastructure. Similarly, LG Uplus employs its 'One LG' strategy to integrate affiliate technologies, enhancing cooling and power systems at its expansive 200MW Paju AI data centre, while SK Telecom collaborates with global partners like Supermicro and Schneider Electric to optimize server, power, and cooling infrastructure, underscoring a collaborative industry approach to operational excellence.

SK Telecom is advancing AI data center architecture through its Panmnesia platform utilizing Compute Express Link (CXL) technology, which flexibly connects CPUs, GPUs, and memory to boost resource utilization and reduce power consumption. This architectural innovation not only enhances operational efficiency but also positions SKT to meet the surging computational demands of AI workloads more sustainably. Complementing this, SK Hyper, a dedicated SKT subsidiary, is tasked with securing critical power infrastructure and constructing substations and transmission lines, facilitating phased expansions from 5GW in 2029 to 15GW subsequently, reflecting a strategic, scalable build-out of AI data center capacity.

KT is pioneering novel operational and business models by bundling subsea cable access with AI services and introducing a token-based usage billing system through its 'Token Factory' platform, effectively linking data transmission with AI consumption billing. This integrated infrastructure chain not only streamlines operational efficiencies but also innovates financial management by aligning costs directly with actual AI usage. Furthermore, KT's ambitious plan to convert 3,500 central offices into AI edge computing facilities exemplifies its commitment to operational integration and enhanced edge capabilities, while its expansion of AI data center capacity from 152MW to 1.15GW and subsea cable bandwidth from 38Tbps to 128Tbps underscores a comprehensive approach to connectivity and scalability.

LG Uplus's operational strategy for AI data centers includes a design, build, and operate (DBO) approach to efficiently scale capacity and develop regional hubs, with plans to open four buildings at its Paju AI data centre between 2027 and 2028. This phased expansion, combined with the integration of diverse group technologies under the 'One LG' umbrella, aims to drive operational innovation and secure competitive differentiation in the rapidly evolving AI infrastructure landscape.

South Korea's leading telecom operator KT is aggressively reallocating capital expenditure from traditional network infrastructure toward expansive AI data center and subsea cable development, signaling a strategic financial pivot. With plans to scale AI data center capacity from 152MW to 1.15GW by 2031 and subsea cable bandwidth from 38Tbps to 128Tbps, KT is investing approximately ₩1 trillion (~$706.3 million) in subsea cables and self-funding 20–30% of AI data center expansion, underscoring its commitment to AI as a core growth engine. This shift is already yielding significant financial returns, as KT’s AI transformation (AX) business revenues more than doubled to ₩367.8 billion (~$259.8 million) in Q2, with projections estimating the new 1GW AI data center could generate ₩4.1 trillion (~$2.9 billion) annually and deliver operating profits near ₩500 billion (~$353.1 million), highlighting the lucrative potential of prioritizing AI infrastructure over legacy telecom services.

The broader South Korean telecom industry is mirroring KT’s financial prioritization by substantially reducing capital expenditures on traditional network infrastructure while ramping up investments in AI data centers, reflecting a coordinated strategic shift. For instance, SK Telecom and SK Broadband combined CAPEX fell by 15.8%, and KT’s CAPEX dropped 15.1% year-on-year in the first half of 2026, whereas LG Uplus increased spending driven by AI data center construction. This reallocation is paying off financially, with SK Telecom’s AI data center revenue nearly doubling to 136.2 billion won in Q2 and KT Cloud’s revenue rising almost 20%, underscoring the profitability and growth momentum of AI infrastructure within the telecom sector.

KT is pioneering innovative financial models to monetize its AI infrastructure, exemplified by the upcoming launch of the 'Token Factory'—a token-based usage billing system that directly ties customer charges to actual AI consumption. This novel approach enhances revenue predictability and customer retention by integrating billing with AI data centers and subsea cable services, forming the final piece of KT’s comprehensive bundling strategy. Such inventive billing mechanisms illustrate how telecom operators are evolving their business models to capitalize on AI infrastructure expansion, signaling a transformative shift in how AI services are commercialized in the industry.

South Korea’s AI infrastructure ecosystem thrives on a tightly integrated network of semiconductor giants like SK Telecom, Hynix, and Samsung, alongside native tech innovators such as Naver, which leverage both domestic platforms and international venture capital to fuel AI innovation. This synergy is bolstered by proactive government policies promoting sovereign AI foundation models to reduce foreign dependency and enhance national resilience, creating a robust environment where hardware, software, and services form a cohesive AI innovation 'sandwich.'

South Korean companies are strategically positioning themselves to lead comprehensive AI infrastructure projects in Central Asia, where countries like Kazakhstan aim to build sovereign AI hubs with investments up to $30 billion. This regional AI-energy-resource corridor leverages complementary strengths—Kazakhstan’s power and land, Uzbekistan’s industrial demand, and hydropower from Kyrgyzstan and Tajikistan—offering Korean firms opportunities to integrate power generation, data center construction, and AI technology into unified industrial systems rather than isolated components.

Traditional South Korean manufacturers and construction firms are rapidly transforming to meet the unique demands of AI data centers by developing specialized materials, integrated supply chains, and dedicated AI infrastructure teams. Companies like Hyundai Steel and POSCO have launched AI-focused divisions and products, while builders such as Daewoo E&C and SK Ecoplant are expanding beyond construction into land acquisition, operations, and asset management, reflecting a coordinated industry effort to build a comprehensive AI infrastructure ecosystem domestically and internationally.

The telecom sector is spearheading a collaborative approach to AI data center expansion by establishing dedicated subsidiaries like SK Telecom’s SK Hyper to manage site acquisition and business development, while integrating AI compute infrastructure with existing telecom assets across thousands of central offices. This phased, demand-validated investment strategy is supported by government commitments to invest 550 trillion won through 2029, enabling large-scale regional AI data center clusters and international expansion to meet global big tech demand, exemplifying a coordinated ecosystem involving private-public partnerships and multi-stakeholder collaboration.