South Korea's Ministry of Science and ICT on 16日 announced three national mega-projects totaling 550 trillion won in AI data center investments, alongside Physical AI and K-semiconductor development initiatives, during a briefing to the president at Cheong Wa Dae Yeongbinguan. The government aims to support 8.4GW of AI data center capacity staged from 2029, develop indigenous Physical AI world models by 2027, and establish 1-nanometer semiconductor technology roadmaps by December to secure national technological leadership. The initiative responds to accelerating global demand for AI infrastructure and aims to transform domestic equipment and component industries into export sectors through integrated solution packages.
SK, GS, and Naver to Build 8.4GW AI Data Centers by 2029
SK will construct 5 gigawatts of capacity across five regions including Ulsan, GS will build 2.4GW in Donghae, and Naver will develop 1GW across multiple regions centered on Sejong, according to the ministry's plan. These gigawatt-scale facilities will begin staged operation from 2029. The government established a three-axis support system comprising an inter-ministry task force, a public-private alliance, and a dedicated support team to resolve bottlenecks in power supply, site acquisition, and permitting processes. From 2027, the ministry will support localization and advancement of core data center solutions, IT equipment including servers, and power and cooling facilities, with validation in operational data centers for export readiness. A ministry official stated the government will develop solutions and equipment industries as export sectors given global data center construction growth, noting that HBM and IT equipment, transformers, and cooling systems are already being exported, with plans to expand production and export capacity through packaged solution offerings rather than individual equipment sales.
Government Allocates 34 Billion Won for Physical AI World Model Development
The government will invest 34 billion won from 2026 to 2027 for 10 corporate and institutional consortia to develop an indigenous Physical AI world model, targeting world leadership by 2030. The world model technology simulates real-world motion and physical laws in virtual environments to produce synthetic data for robot and autonomous machine learning. From 2027, the ministry will establish a system to generate large-scale synthetic data through the world model and create learning and training frameworks. Based on this foundation, the government plans to develop a Physical AI foundation model capable of perceiving and acting in environments like humans, along with a domestic full-stack encompassing chips, data infrastructure, and services. Consortium selection for the Physical AI project will evaluate participants based on pre-verification projects conducted in North Jeolla and South Gyeongsang provinces to build a domestic technology-centered ecosystem. A ministry official emphasized that while foreign companies may participate in consortia, the goal is to localize Physical AI technology with domestic capabilities. Developed technologies will first be demonstrated in regional manufacturing sites, then expanded to defense, care, agriculture, and public safety applications for cultivation as export industries.
Ministry Plans 1-Nanometer Semiconductor Roadmap by December
The government will establish a 'K-AI Semiconductor Full Package' ecosystem composed of domestic technologies spanning AI accelerators, infrastructure and networks, software, and AI services. The ministry will support initial market establishment of domestic AI semiconductors through application in major corporate products and services, expanded demonstration, designation as innovative public procurement products, and lead purchasing. Development of next-generation core technologies to overcome physical limits of semiconductors will also proceed, with preliminary planning in the second half of this year leading to a roadmap for 1-nanometer-class devices and stacked next-generation HBM technology by December, and full-scale R&D starting from 2027. A ministry official explained the goal is to achieve performance improvements of more than 10 times compared to existing technologies using next-generation materials below 1 nanometer, noting that unlike simple stacked memory developed by the private sector, the approach will heterogeneously stack computing devices and memory to enhance both performance and energy efficiency.
South Korea to Complete 5G Dedicated Networks by Year-End
To support communication infrastructure for AI data centers, the government will complete 5G dedicated networks by year-end and expand intelligent base stations. Development of 6G technology demonstrations and low-orbit satellite communication networks will proceed in parallel to secure ultra-high-speed, ultra-low-latency network infrastructure required for data centers and Physical AI.
FAQ
What is the total investment scale for South Korea's AI data center projects?
The total investment is 550 trillion won, including foreign investment attraction. SK will build 5GW across five regions including Ulsan, GS will construct 2.4GW in Donghae, and Naver will develop 1GW across multiple regions centered on Sejong, totaling 8.4GW of capacity with staged operation beginning in 2029.
When will South Korea's Physical AI world model development begin?
The government will invest 34 billion won from 2026 to 2027 for 10 corporate and institutional consortia to develop an indigenous Physical AI world model. From 2027, the ministry will establish systems to generate large-scale synthetic data and create learning frameworks, aiming to develop foundation models and domestic full-stack solutions within three years.
What is the timeline for South Korea's 1-nanometer semiconductor technology development?
The ministry will conduct preliminary planning in the second half of this year and establish a roadmap for 1-nanometer-class devices and stacked next-generation HBM technology by December. Full-scale research and development will commence from 2027, targeting performance improvements of more than 10 times through next-generation materials and heterogeneous stacking of computing devices and memory.
