Elice Group will build a modular AI data center equipped with 2,560 Nvidia B300 graphics processing units (GPUs). It plans to create an eco-friendly data center environment by applying a hot-water cooling method.

Elice Group said on Monday it had been selected as a participating company in the "AI computing resource utilization capability strengthening project" promoted by the Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA).

The project focuses on expanding the national AI computing infrastructure and supporting AI research and development by domestic companies and research institutions. Elice Group will deploy the 2,560 B300 GPUs across four modular AI data center buildings, with 640 units in each.

To match the B300's high-power, high-density environment, it designed the GPU cluster and the cooling, network and power systems. The data center's power usage effectiveness (PUE) target is about 1.1. It will also apply direct liquid cooling (DLC) and an ultra-high-speed network architecture for large GPU clusters.

The cooling system will adopt a hot-water cooling method using cooling water of 40 degrees Celsius or higher, along with outside-air cooling technology. By reducing the power used in cooling, it plans to improve operating efficiency and pursue eco-friendly operations. To reduce leakage risks in a liquid-cooled data center, it designed separated cooling pipes based on the company's patented technology.

To address load issues in high-density liquid-cooled infrastructure, it will apply a modular container method with high-load and high-insulation performance. The company said conventional data centers require large-scale structural reinforcement work when building liquid-cooling infrastructure, but a modular structure allows GPU infrastructure to be built relatively quickly. It said.

Elice Group plans to use the project to strengthen cooling technology that can respond to next-generation high-power GPUs, including the B300 and future Nvidia Vera Rubin, as well as capabilities to design and build portable modular data centers (PMDCs).

Elice Group CEO Jae-won Kim (김재원) said, "We will build a modular AI data center optimized for high-performance GPU environments to provide a stable and efficient AI computing infrastructure," adding, "Through K-PMDC completed based on domestic technology, we will contribute to strengthening domestic AI computing competitiveness."