Test report DSG-7166 · Rev C · tested November 14, 2026

AI Datacenter InfrastructureDevice under test

Elice Group Pilots GPU Data Center Cooled by Hot Water and Outside Air

Elice Group started trial operations Oct. 30 of a modular data center cooling Nvidia B300 GPUs with hot water and outside air, targeting a PUE of 1.1 or below ahead of commercial launch this year.

Read
2 min
Words
444
Node
3nm
Operator
Grace Kim

Spec summary

  1. Elice Group began trial operations of its hot-water and outside-air cooled GPU data center on Oct. 30
  2. The facility targets a power usage effectiveness (PUE) of 1.1 or below
  3. Cooling liquid operates above 40 degrees Celsius, with outside air rejecting the heat
  4. Elice Group claims the world's first use of both cooling methods on Nvidia B300 servers
  5. Commercial service is planned within 2025 following the beta phase under the Ministry of Science and ICT's GPU program
Elice Group Starts Pilot Run of GPU Data Center Cooled by Hot Water and Outside Air
Fig. AElice Group Starts Pilot Run of GPU Data Center Cooled by Hot Water and Outside Air — Akela999 / pixabay

Elice Group began trial operations on Oct. 30 of a modular data center that cools Nvidia B300 GPU servers using liquid above 40 degrees Celsius combined with outside air — a combination the company says is a world first for B300 hardware.

The pilot runs under the Ministry of Science and ICT's program supporting the procurement, construction and operation of advanced GPUs, the company announced on Nov. 14. Elice Group plans to launch commercial service within this year once the beta phase ends.

It marks the first time Elice Group has built and operated hot-water cooling and outside-air cooling together since entering the data center business.

How does the cooling system work?

The design absorbs heat from GPUs using liquid at temperatures above 40 degrees Celsius, then uses outside air to lower the coolant's temperature. That dual approach cuts the electricity needed for cooling.

Elice Group targets a power usage effectiveness (PUE — the ratio of total facility energy to computing energy) of 1.1 or below for high-efficiency operation. Lower is better; the metric is the industry's standard yardstick for data center efficiency.

What does this change for Korean AI infrastructure?

The ministry's program aims to expand South Korea's AI computing infrastructure and strengthen the foundation for using high-performance GPU resources. It's a direct bet on domestic capacity for training and inference workloads that currently lean on foreign cloud capacity.

Elice Group has been building the infrastructure to run B300 GPUs since being selected as an operator in June this year. The company's tool for the job is AI PMDC, its in-house modular AI data center platform.

Why the timing matters

B300 GPUs sit at the top of Nvidia's data center lineup, and cooling them is the hard part. Hot-water liquid cooling handles the dense heat load at the chip; outside-air cooling then rejects that heat without running compressors year-round. Combining the two is what Elice Group claims no other B300 operator has done.

The pilot's success criteria are straightforward: hold the targeted PUE through the beta phase, then flip to commercial service before year-end. If it works, the design becomes a template for future Korean GPU builds under the ministry's program.

Key facts at a glance:

  • Trial operations started Oct. 30; commercial launch planned within the year
  • Cooling liquid runs above 40 degrees Celsius, with outside air as the second stage
  • Target PUE: 1.1 or below
  • Operator selection came in June under the Ministry of Science and ICT's GPU program
  • Facility uses Elice Group's in-house AI PMDC modular data center platform

via en.sedaily.com (Source)

Filed under

  • nvidia-b300
  • ai-data-centers
  • liquid-cooling
  • data-center-energy-efficiency
  • modular-infrastructure
Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering regulation, policy and their impact on hardware layer of artificial intelligence at Die Signal.

565 articles

Same lot · LOT-362A

« Previous articleNext article »