Elice launches world-first B300 GPU modular data center

A new standard for AI hardware

Elice recently announced the completion of an eco-friendly modular artificial intelligence (AI) data center. This new facility is specifically equipped with Nvidia B300 graphics processing units (GPUs), which are the specialized chips used to train and run complex AI models. The company officially started a beta service for this infrastructure on the 30th of last month.

The project was developed as part of a government initiative by the Ministry of Science and ICT. This program is designed to help secure, build, and operate advanced GPUs within the country. Elice was selected as a project operator in June to lead the construction of this specific B300 GPU infrastructure.

The company intends to move from this testing phase to a full service launch before the end of the year. By providing an end-to-end offering, Elice handles everything from the initial design and construction of the data center to the actual cloud operations. They use their own cloud-optimized solution, known as ECI, to manage these services for clients.

Modular construction and speed

A core component of this project is the use of Elice's in-house modular AI data center, which they call AI PMDC. Unlike traditional data centers that are built piece-by-piece on a construction site, this method involves pre-manufacturing key facilities in modular units. These units are then transported and assembled on-site, which significantly shortens the overall construction timeline.

This modular approach allows for more predictable building schedules and potentially lower costs during the development phase. By standardizing the components of the data center, Elice can deploy high-performance computing environments more rapidly than conventional methods. It represents a shift toward industrializing the way large-scale computing hubs are built.

The company has already established a footprint in this space. Last year, Elice became the first in Asia to build an AI PMDC based on B200 GPUs. This year's move to the B300 chips continues their trajectory of integrating the latest hardware into their modular framework.

Innovative cooling technologies

One of the most significant technical features of this new data center is its cooling system. It combines warm-water cooling, which utilizes water at temperatures above 40°C, with an air-side economizer. The air-side economizer is a system that lowers the temperature of the cooling water by using outside air to facilitate heat exchange.

The primary goal of combining these two technologies is to drastically reduce the amount of power consumed by cooling systems. High-performance GPUs generate immense amounts of heat, and traditional cooling can be one of the largest expenses and energy drains for a data center. By using warmer water and outside air, the facility aims to achieve a power usage effectiveness (PUE) of 1.1 or lower.

PUE is a standard metric used to measure how efficiently a data center uses energy. Achieving a 1.1 or lower is a high target for a facility housing powerful B300 GPUs. As Chosunbiz reported, this makes it the first data center in the world to combine these specific cooling technologies with B300-based modular design.

Strengthening regional infrastructure

The development of this center is tied to a broader goal of improving the competitiveness of Korea's AI infrastructure. By creating a domestic hub for advanced GPUs, the company aims to provide a stable environment for local and international AI development. This infrastructure is intended to support the growing demand for high-speed, reliable computing power.

Kim Jae-won, the CEO of Elice, noted that the start of the beta service demonstrates how the company's technical capabilities are turning into tangible results. He emphasized that the company will continue to advance its modular and cloud technologies. The focus remains on providing the necessary tools for the next generation of AI applications.

For the industry, this move signals a trend toward more specialized, high-efficiency data centers. Rather than building massive, general-purpose facilities, companies are increasingly looking for modular solutions that can be scaled and cooled more effectively. Elice's model provides a blueprint for how these requirements can be met simultaneously.