The U.S. government is launching a 2.8 trillion KRW (approximately 1.85 billion USD) project to secure graphics processing units (GPUs) in partnership with Naver Cloud, Samsung SDS, and Ellice Group. This initiative aims to build the necessary computing infrastructure for developing proprietary artificial intelligence (AI) foundation models and supporting national AI projects in collaboration with private cloud companies.
On Monday, the Ministry of Science and Information and Communications (ICT) announced the selection of three cloud service providers—Naver Cloud, Samsung SDS, and Ellice Group—to participate in the advanced GPU acquisition, construction, and operational support project. This initiative is a crucial component of the government’s AI Highway program. The ministry intends to secure 9,704 advanced GPUs to meet AI development needs in both private and public sectors.
The GPU acquisition includes 2,016 units of the next-generation Veralubin and 7,688 units of the B300. Naver Cloud will procure 1,008 Veralubin units and 3,112 B300 units, Samsung SDS will acquire 1,008 Veralubin units and 2,016 B300 units, while Ellice Group will establish 2,560 B300 units.
A key aspect of the policy is not merely the quantity of GPUs, but their actual computational power. The government initially targeted 15,000 B200 units but now reports that the Veralubin and B300 combination will deliver performance equivalent to about 19,000 B200 units. Despite the lower unit count, this represents a 30% increase in performance over the original goal.
Of the 9,704 GPUs, the Ministry will allocate 2,016 Veralubin units and 4,360 B300 units for government use. These will support the development of proprietary AI foundation models, national AI projects, and the advancement of AI models and services in collaboration with academia and industry.
The remaining 3,328 B300 units will be utilized by the cloud service providers for their own purposes, enabling them to enhance cloud-based GPU services and improve their AI models and services.
The Ministry plans to issue GPU purchase orders in June. Cloud companies will begin offering B300 services within the year as they complete their installations. Veralubin services will be rolled out gradually starting in the first half of 2027, in line with the product launch schedule.
Deputy Prime Minister and Minister of Science and ICT Lee Kwang-hoon stated that the advanced GPUs they’re securing, including Veralubin, will be a game-changer in accelerating AI research and development. The goal is to establish world-class AI infrastructure to drive innovation and growth for Korean companies and research institutions.
This ambitious project underscores the government’s commitment to positioning Korea at the forefront of AI technology and ensuring its competitiveness in the global AI race.