The Korean government is launching a "GPU Acquisition, Construction, and Operation Support Project" with a budget of 1.459 trillion KRW (approximately 1.5 trillion USD) to secure advanced Artificial Intelligence (AI) infrastructure. The primary goal is to provide critical AI computing resources to domestic academia, research institutions, and startups, thereby fostering national AI competitiveness.

The core methodology of this initiative involves a unique public-private partnership model:

1. Government Procurement: The National IT Industry Promotion Agency (NIPA), under the Ministry of Science and ICT (MSIT), will utilize the allocated budget to purchase state-of-the-art GPUs, such as NVIDIA's next-generation Blackwell (B200) architecture. NIPA will retain ownership of these GPU assets.
2. Private Sector Operation: Private cloud service providers (CSPs) and other qualified entities will be selected as project operators. These operators will be responsible for the physical construction of the necessary data center infrastructure to host these GPUs, their subsequent clusterization, and their operational management for a contractual period of five years. This includes ensuring robust power supply, cooling, and network connectivity crucial for large-scale GPU operations.
3. Shared Resource Utilization: While NIPA owns the GPUs, the participating private operators gain significant operational advantages. After allocating a designated portion of the GPU compute capacity for public projects involving academia, research bodies, and startups, the operators are granted the right to utilize the remaining GPU resources for their own commercial services or internal AI development initiatives. This model aims to incentivize private sector participation by offering access to high-value AI infrastructure without direct capital expenditure on GPU procurement.

A critical technical challenge for prospective participants is the stringent infrastructure requirement. To effectively support and cluster a large volume of cutting-edge GPUs like the B200, data centers must be AI-optimized, capable of handling significant power loads and advanced cooling. Industry estimates suggest a minimum power reception capacity of 17-20 megawatts (MW) is necessary to meet the government's expected scale. This requirement severely limits the number of eligible domestic companies.

Key contenders and their positions include:

• Naver: Considered the frontrunner due to its proactive expansion of data center capacity, including leasing significant spaces at LG CNS Jukjeon (approximately 19 MW) and LG U+ Gasan (approximately 20 MW) data centers, along with a recent third 10 MW lease. These acquisitions are widely interpreted as strategic moves to accommodate the project's demanding infrastructure needs.
• Coupang: Despite being an "all-in cloud" enterprise currently utilizing Amazon Web Services (AWS), Coupang has shown keen interest. It reportedly attended the project briefing and is exploring colocation data center options for its own internal AI infrastructure development, indicating a potential shift in its cloud strategy. Participation in this project would align with its strategic pivot towards building proprietary AI capabilities.
• Samsung SDS: Expressed interest, but faces limitations as most of its available data center capacity is allocated to Software-Defined Data Centers (SDDC) serving group affiliates, making them unsuitable for external GPU integration.
• NHN Cloud: Considered participation but reportedly withdrew interest due to the operational burden of already managing the Gwangju National AI Data Center.
• Other entities: Some companies are actively forming consortia with colocation specialists to address the infrastructure deficit.

This project was initially conceptualized as an adjunct to the larger "National AI Computing Center Construction Project," a public-private special purpose corporation (SPC) initiative to build an AI data center capable of 1 exaFLOPs. However, the larger project faced a lack of private sector interest due to its perceived long-term, uncertain profitability, leading to its cancellation. Consequently, the GPU Acquisition and Operation Support Project was separated as a distinct, more appealing opportunity for private enterprises, offering a simpler 5-year operational term and access to government-funded GPUs for private use.

The application deadline for the project is June 23, 2025. NIPA plans to select operators by next month, with GPU procurement and cluster construction commencing within the year.