GAK Sejong Hybrid Cooling System Evolution
NAVER Cloud · Data Center Engineering
GAK Sejong Hybrid Cooling System
Three generations of energy-efficient cooling architecture — from natural airflow to AI-controlled hybrid liquid systems
Generation 1 · 2013
GAK Chuncheon
Air Management Unit (AMU)
  • Direct outdoor air (free cooling) — natural airflow through server halls
  • Evaporative mist cooling during peak temperature periods
  • Mechanical cooling (CRAC) used < 10% of operational hours annually
Target PUE 1.09
Generation 2 · 2023
GAK Sejong
NAMU I / II / III
  • Fan Wall array (BLDC motors): no air duct friction losses
  • High-volume / low-power airflow design
  • AI-controlled bypass dampers — indirect outdoor air integration
Operational PUE 1.13
Generation 3 · 2027+
AI Factory
Next-Gen Hybrid (DLC + Immersion)
  • Direct Liquid Cooling (DLC) plates on Vera Rubin / Blackwell chips
  • Immersion Cooling: servers submerged in dielectric fluid
  • AI-managed coolant flow rate — responds to real-time GPU thermal telemetry
Design Target PUE < 1.05
Cooling Power Overhead by Technology (% of total IT load)
Traditional CRAC
~60% overhead
60%
GAK Chuncheon (AMU)
~15%
15%
GAK Sejong (NAMU III)
~13%
13%
AI Factory Hybrid (2027)
<5%
<5%
AI Factory 3-Stage Hybrid Cooling — How It Works
①
Air + Bypass Damper
NAMU III fan wall maintains ambient airflow. Indirect outdoor air bypass activates when external temp allows free cooling.
②
Direct Liquid Cooling
Liquid cooling plates attach directly to GPU die surfaces. Chilled water loop absorbs heat at the chip level before it enters the room air.
③
Immersion Cooling
Full server submersion in non-conductive dielectric fluid. Highest-density GPU racks. AI controls fluid circulation based on real-time thermal telemetry.
Source: NAVER Data Center Integration Center · Noh Sang-min · GAK Sejong Technical Whitepaper