The Heart Sustaining the AI Era: Data Centers

Behind every AI-generated output lies a data center. Even AI services that appear instantaneous—like answering questions, generating images, or recognizing speech—actually run on physical infrastructure that aggregates countless servers, storage systems, and network equipment.

Data centers form the foundation that supports everything from photo and video storage to search, shopping, payments, navigation, and AI model training and inference. Especially since the spread of AI, they are being redefined not as simple server storage spaces but as production sites where high-density computation occurs continuously.

The reason data centers are drawing renewed attention in the AI era is GPUs. Unlike traditional CPU-centric environments, GPU-based computation generates far greater power consumption and heat, making it difficult to handle with existing cooling and power structures alone. As a result, the core of data center competitiveness has shifted from the number of servers to how stably high-density computation can be processed.

Team NAVER operates its own data center, 'GAK', having internalized planning, design, construction, and operation entirely in-house. Building on its self-owned data centers in Chuncheon and Sejong, it has accumulated AI infrastructure capabilities, with GAK Sejong in particular introduced as a hyperscale AI data center that stably operates large-scale GPU clusters. The company also has experience being the world's first to apply NVIDIA's SuperPOD in an actual production service environment.

The core technologies of AI data centers can be condensed into three main areas.

• Cooling technology: Air cooling faces growing constraints in high-density GPU environments due to air's low heat capacity and limited heat transfer, so liquid cooling methods such as Direct Liquid Cooling and Immersion Cooling are spreading to compensate.
• Power stability: Since power outages are not permissible, systems feature layered power structures combining UPS, large-capacity batteries, and emergency generators.
• Integrated control systems: Servers, power, cooling, fire safety, and security are managed together on a single screen, enabling early detection of failure signs and minimizing service disruptions.

Team NAVER employs its self-developed hybrid air conditioning system NAMU (NAVER Air Membrane Unit) to combine outside air utilization with contamination blocking, and with NAMU III, it flexibly applies both direct and indirect outside air cooling. As a result, it maintains a PUE of around 1.1x, achieving higher energy efficiency than the global data center average of approximately 1.58.

Ultimately, the data centers of the future will evolve beyond simple storage spaces into systems that self-optimize power, cooling, networking, and control to maximize AI performance. How quickly companies can internalize capabilities in renewable energy utilization, high-efficiency cooling, and facility construction and operational optimization will determine future competitiveness.