Bae said the government has secured approximately 21,000 of the roughly 35,000 GPUs it intends to obtain directly from Nvidia, measured in Nvidia B200 equivalents, with the remaining 14,000 expected to arrive by early 2027. South Korea also plans to secure the equivalent of another 15,000 GPUs through the National AI Computing Center, bringing the government's total allocation to more than 50,000 units.
The broader supply plan was first announced in October 2025. Under the arrangement, Samsung Electronics, SK Group and Hyundai Motor Group would each build AI infrastructure using more than 50,000 GPUs, while NAVER Cloud planned to deploy more than 60,000. The government portion would support sovereign AI projects, cloud providers, research institutes, startups and universities. A Korean government account of the original agreement said the public and private sectors would use the systems for national AI computing, foundation models, industrial applications and physical AI.
Nvidia described the planned buildout as a mix of sovereign cloud capacity and company-operated AI factories. The announced private deployments cover semiconductor manufacturing, automotive systems, robotics, telecommunications and enterprise cloud services, linking the hardware program to several of South Korea's largest technology and industrial groups.
Bae said the two sides also intend to continue technical cooperation beyond the hardware deliveries. Planned work includes connecting Korean-designed AI accelerators with Nvidia infrastructure through NVLink, developing AI computing capacity, conducting physical AI research and training specialized workers. The ministry will continue monitoring GPU supply and demand across the public and private sectors, and said the government-procured capacity is intended for companies, universities and research institutions, including teams working on South Korea's sovereign foundation-model and AI-for-all initiatives.
The reaffirmation comes as South Korea expands several Nvidia-linked infrastructure programs. Nvidia, SK Telecom and Korean chip developer Rebellions plan to connect locally designed neural processing units with Nvidia's rack-scale infrastructure for sovereign AI systems targeted for deployment in 2028, further deepening the country's integration with Nvidia's technology stack.
