Full-scale deployment of modular AI data centers optimized for next-generation B300 GPUs
AI full-stack company Elice Inc. (CEO Jae Won Kim), providing AI infrastructure, cloud, and industry-specific solutions, has been selected as a participating company in the “AI Computing Resource Utilization Reinforcement (GPU Server Expansion & Integrated GPU Operation Environment Establishment)” project led by the Ministry of Science and ICT (MSIT) and the National IT Industry Promotion Agency (NIPA). Elice Inc. will build and operate modular AI data centers optimized for the B300 GPU. This project aims to expand national AI computing infrastructure and support AI R&D for domestic companies and research institutes.
Elice Inc. will deploy a total of 2,560 B300 GPUs across four modular AI data center buildings, 640 GPUs each. The company has completed data center designs optimized for ultra-high-power, high-density environments, fine-tuning everything from cluster architecture and cooling to networking and power efficiency for the latest B300 GPUs. The goal is to maintain a Data Center Power Usage Effectiveness (PUE) of 1.1. Direct Liquid Cooling (DLC) and an ultra-high-speed network architecture optimized for large-scale GPU clusters will also be applied.
High-efficiency operation via warm-water and free-air cooling… Leak-risk mitigation through patented separate-piping design
The cooling system adopts Warm Water Cooling (utilizing coolant at 40°C or higher) and free-air cooling. This reduces cooling-related power consumption, improves operational efficiency, and enables eco-friendly data center operations. Additionally, leveraging its patented technology, Elice Inc. designed a separate piping structure to address leak risks inherent in liquid-cooled data centers.
High-load, high-insulation modular containers applied… Targeting PUE 1.1-level efficiency
Structural load challenges from liquid cooling in next-generation AI semiconductor environments are resolved through modular containers with high-load-bearing and high-insulation performance. While conventional data centers require massive structural reinforcement for high-density liquid-cooling infrastructure, Elice Inc.’s proprietary high-load modular structure enables rapid, flexible deployment of large-scale GPU infrastructure.
Through this project, Elice Inc. will contribute to expanding national AI computing infrastructure while strengthening high-efficiency cooling technology and PMDC design/construction capabilities ready for next-generation, high-power, high-density GPUs—including the B300 and future architectures such as NVIDIA Vera Rubin. Building on this, the company plans to advance its end-to-end AI infrastructure capabilities spanning large-scale clustering, cooling and power optimization, and AI cloud service operations.
Elice Inc. CEO Jae Won Kim stated:
“Through this project, we will build modular AI data centers optimized for high-performance GPU environments, providing stable and efficient AI computing infrastructure. We will establish AI infrastructure optimized from GPU to network and storage, and contribute to strengthening Korea’s AI computing competitiveness through K-PMDC realized with domestic technology.”