South Korea to invest 2.8 trillion won in national AI GPU clusters
South Korea’s government is expanding its investment in domestic AI infrastructure, aiming to install and operate large-scale GPU clusters through a 2.8 trillion won package to support research, industry, and public sector AI projects.
The Ministry of Science and ICT announced the 2026 AI Computing Resource Utilization Foundation project, known as GPU 확보·구축·운용지원, on the 12th. The public call runs from now until the 13th of the following month. Officials say the plan will accelerate the AI ecosystem and move toward turning Korea into a leading AI nation, with service delivery expected within the year and ongoing operations through 2031.
The budget focuses on buying GPU servers, racks, cooling systems, storage, networking gear, software licenses, and technical support. In addition to simply increasing GPU numbers, the government emphasizes high performance per cost, large-scale clustering, direct in-house clustering capabilities, prioritizing the latest GPUs, and early service deployment. Proposals that propose at least one cluster of 256 servers (2048 GPUs) or more and include Nvidia’s Blackwell-class GPUs or higher will be prioritized; proposals mentioning next-generation Verarubin GPUs will also be favored.
Eligible participants are domestic cloud providers that can offer and operate GPU-as-a-Service (GPUaaS). Either solo firms or consortia may bid, but proposals must include relevant revenue and operating performance from the past three years. Bidders must have data-center space in Korea and demonstrate plans for integrated power, cooling, temperature and humidity control, and networking infrastructure to support large-scale GPU clusters. The government appears to be evaluating not just GPU supply, but the ability to operate and manage a high-performance, integrated infrastructure.
The government will hold an information session on the 20th at POSCO Tower Yeoksam in Seoul. Applications are due by early next month, with the evaluation committee selection and on-site data-center inspections planned for April and May. The winner is to be announced in May, followed by procurement, government funding, GPU orders, installation, testing, and an end-of-year service launch. Mid-term reporting is scheduled for September, with final assessments in January and settlement in February.
This year’s push builds on last year’s extension of a GPU expansion program funded by a supplementary budget of about 1.4 trillion won to secure around 13,000 GPUs. Naver Cloud, NHN Cloud, and Kakao were selected in that round. Since then, government and industry groups have formed a practical coordination body to align GPU provisioning, deployment, and an integrated support platform for academia and industry. GPU supplies began flowing to research and industry partners earlier this year under the initial program.
Experts say the new bid signals that Korea intends more than just adding GPUs—it seeks a sustained, government-backed GPU service ecosystem capable of supporting large-scale AI training and inference for national projects, industry R&D, and startup AI services. For U.S. readers, the plan matters because it reflects wider global competition to secure AI infrastructure, with implications for multinational cloud providers, hardware suppliers like Nvidia, and cross-border collaboration on AI standards, security, and innovation ecosystems.