‘Operation efficiency at core of AI infrastructure capability’

A server room at Naver's GAK Sejong data center in Sejong City / Courtesy of Naver Cloud

SEJONG — Naver Cloud showcased its hyperscale data center GAK Sejong, expressing its confidence and efficiency in designing and operating data centers, the most important infrastructure for artificial intelligence (AI) services.

During a press event at the center on Monday, Naver Cloud Chief Information Officer Lee Sang-jun said the competitiveness of AI infrastructure depends not only on how many graphics processing units (GPUs) a company secures, but also on how stably and efficiently it operates those resources.

“By maintaining the balance between GPU sourcing and internalization of operational technologies, Naver Cloud is strengthening its AI infrastructure competitiveness,” he said.

“These days, doing AI business requires massive investment: large-scale data centers like this one, and the huge amount of electricity they consume,” Lee added. “But expanding those things doesn’t happen overnight. Unlike global tech giants that spend on an unimaginable scale, we have to pursue the most efficient approach possible. I believe that through technological internalization, we will eventually reach that level.”

This undated handout photo shows Naver's GAK Sejong data center in Sejong City. The larger building above is the north wing of the facility, while the smaller structure connected to it serves as the main building. Courtesy of Naver Cloud

Occupying a 294,000-square-meter plot in the west-central region of Sejong City, GAK Sejong is portal and mobile giant Naver’s second proprietary data center, widely viewed as Korea’s first data center dedicated to empowering AI services. GAK is named after Janggyeonggak, referring to a building where Buddhist or Confucian scriptures are stored.

GAK Sejong began operations in November 2023 as a high-density GPU computing facility capable of handling both AI training and inference simultaneously. GAK Sejong is equipped to accommodate up to 600,000 server units upon the completion of its expansion phase, providing a total storage capacity of 65 exabytes.

GAK Sejong currently consists of a main building and a completed North Wing, with plans underway to construct a South Wing in the coming years. The North Wing will undergo three expansion phases, followed by phases four to six in the South Wing. Currently, only the first phase is operational, while the second and third phases are scheduled for completion in 2027 and 2029.

Naver Cloud Chief Information Officer Lee Sang-jun speaks during a press conference at Naver's GAK Sejong data center in Sejong City, Monday. Courtesy of Naver Cloud

To cope with the enormous amount of electricity consumed by AI accelerators, GAK Sejong is located just 2 kilometers away from a substation, designed to supply up to 275 megawatts of power. There are also plans to use a cooling system that combines direct air, indirect air and chilled water to keep server temperatures at a desired level between 22 and 26 degrees Celsius.

This cooling system embodies Naver’s accumulated expertise in technological internalization.

The first-phase server rooms currently use an air cooling system based on Naver’s proprietary designs. The company developed its first air misting unit in 2013 for the GAK Chuncheon data center, followed by the Naver Air Membrane Unit (NAMU) I in 2014, NAMU II in 2017 and the latest NAMU III, which is installed at GAK Sejong. NAMU III is a hybrid system that uses direct air cooling in spring and fall, and indirect air cooling during summer and winter.

The interior of GAK Sejong's NAMU III air cooling system / Courtesy of Naver Cloud

The currently operating North Wing spans approximately 350 meters in length and consists of five stories — three underground and two aboveground. It was designed to withstand earthquakes of up to magnitude 7 and uses outside air for cooling around 240 days a year to maximize energy efficiency.

There are four server rooms in the first phase, each capable of housing up to 600 servers. This reporter visited one of them, where Nvidia’s H100 GPUs are in operation. A monitoring panel displayed the real-time power consumption of each server, and both walls of the room were equipped with NAMU III air cooling systems. The room maintained a pleasantly cool temperature with a faint breeze, though the noise level was quite high.

Naver's GAK Sejong data center head Roh Sang-min, left, speaks during a press conference at the data center in Sejong City, Monday. Courtesy of Naver Cloud

Roh Sang-min, head of GAK Sejong, said that the second and third expansion phases will introduce direct liquid cooling (DLC), which is required by future servers such as Nvidia’s B200 and B300.

Along with DLC, Roh said Naver is also conducting feasibility tests for immersion cooling, a method that submerges servers directly in liquid, at GAK Chuncheon. While Nvidia’s current AI servers do not support immersion cooling, the U.S. tech giant plans to introduce it starting with next-generation platforms following the Rubin Ultra series.

With one-sixth of GAK Sejong’s planned full capacity currently in operation, Roh said the facility’s annual electricity costs have already reached about 22 billion won ($15.34 million), and the number is expected to rise sharply in the coming years, requiring Naver to continue investing resources into data center operations.

Autonomous mobility robot SeRo, left, and GaRo, right, operate in an IT storage room at Naver's GAK Sejong data center in Sejong City in this undated handout photo. Courtesy of Naver Cloud

“As we enter the age of AI data centers, the power density has increased to a completely different level compared to when we used CPUs in the past,” Roh said. “Hundreds of GPUs operate together as a single cluster, consuming a highly concentrated amount of electricity, and server rooms are already using space much more compactly.”

“From the second half of this year through next year, domestic data centers will continue to face a shortage of available space. Starting in 2027, more capacity will be available, but demand will likely remain unpredictably high. We opened GAK Sejong in November 2023, and it’s already full, so we’re carrying out both the second and third expansion phases simultaneously.”

The main control room of Naver's GAK Sejong data center in Sejong City / Courtesy of Naver Cloud

CIO Lee said that considering growing demand, Naver Cloud’s GPU-as-a-service (GPUaaS) business is showing strong growth.

“Naver Cloud will leverage its experience in AI infrastructure operations to expand the GPUaaS model and create an ecosystem where companies can easily adopt AI,” he said. “Through this, we aim to make AI infrastructure not just a corporate asset but a foundation for growth across industries.”