NAVER Cloud to Drive GPUaaS Era with World-Class AI Infrastructure Capabilities
NAVER Cloud to Drive GPUaaS Era with World-Class AI Infrastructure Capabilities
NAVER Cloud to Drive GPUaaS Era with World-Class AI Infrastructure Capabilities
– Unveils full-stack AI infrastructure integrating data centers, AI resources, and AI platform technologies
– Ensures 24/7 operational stability through predictable infrastructure operations and automation systems
– Expands AI infrastructure services via GPUaaS model, establishing a broad-based AI utilization ecosystem across industries
October 27, 2025
At a tech meetup held at “GAK Sejong,” Korea’s first AI data center, NAVER Cloud (CEO Kim Yu-won) unveiled its differentiated capabilities to maximize GPU operational efficiency and independently design and operate AI infrastructure. The company plans to expand AI adoption across industries by leveraging its world-class GPU operation capabilities.
AI infrastructure competitiveness hinges on “operational efficiency,” not just “GPU acquisition”—NAVER possesses world-class operational capabilities
NAVER Cloud CIO Lee Sang-jun stated, “The competitiveness of AI infrastructure lies not in how many GPUs are acquired but in how reliably and efficiently those resources are operated.” He added, “NAVER Cloud is strengthening its AI infrastructure by balancing GPU acquisition with in-house operational technology.”
In 2019, NAVER became the first company in the world to commercialize NVIDIA’s supercomputing infrastructure, SuperPOD, and has since gained extensive experience in designing and operating ultrahigh-performance GPU clusters. Building on this experience, the company has directly designed and currently operates a large-scale GPU cluster at “GAK Sejong,” independently developing key data center infrastructure, such as cooling, power, and networking, while internalizing technologies optimized for AI workloads.
Lee added, “NAVER’s data centers serve as a full-stack AI infrastructure capable of comprehensively managing all AI workloads,” emphasizing, “This capability to design and operate infrastructure as a single integrated system is recognized as one of the best, not only in Korea but globally.”
‘GAK Sejong’: Beyond data storage—A space where intelligence operates
GAK Sejong was designed not merely as a traditional Internet Data Center (IDC) for storage and processing, but as a high-density GPU compute space where both AI training and inference coincide. To support this, NAVER Cloud has established a 24/7 stable operating environment engineered to ensure power efficiency, cooling performance, and uninterrupted service.
Internalizing next-gen cooling technologies for the high-density, high-power era
In data centers dedicated to AI computation, thermal management is the most critical factor. As GPU power density increases, cooling has become the key determinant of both operational efficiency and system stability.
Based on real-world data and thermal environment analysis accumulated from operating “GAK Chuncheon,” “GAK Sejong” has implemented a hybrid cooling system that integrates direct outdoor air cooling, indirect outdoor air cooling, and chilled water. This system automatically switches cooling modes depending on the season—using direct outdoor air in winter and combining indirect air cooling with chilled water during transitional seasons and summer. As a result, even in high GPU density environments, the facility maintains stable thermal control and high energy efficiency.
NAVER Cloud is also testing a liquid immersion cooling container infrastructure to verify the stability of the cooling fluid, energy efficiency, and operational reliability. Building on these efforts, the company has developed a concrete road map for next-generation cooling technologies designed for high-power, high-density environments and has already filed a patent for its proprietary water-cooled server systems.
GAK Sejong’s design enables uninterrupted operations
“GAK Sejong” has been designed with an ‘Active–Active Architecture’ that ensures uninterrupted service even during system failures. It achieves this by separating power, cooling, and server operations into isolated systems while maintaining cohesive operation.
To address the high-power demands of GPU servers, uninterruptible power supply (UPS) units and power distribution systems were strategically reconfigured to block fault propagation structurally. This architecture serves as the foundation for “GAK Sejong’s” 24/7 stable operation.
Ensuring 24/7 stability through predictable infrastructure operation and automation systems
Leveraging its experience managing hundreds of thousands of servers, NAVER has built a standardized infrastructure and automated operations system that remains stable even during system failures. All servers undergo predeployment verification for performance, power efficiency, and operability, and are configured to standardized specifications. High-performance resources such as GPUs are continuously monitored in real-time and supported by automated recovery systems, ensuring stable service continuity even in the event of disruptions.
The operation of GPU clusters also incorporates automation technologies for resource management, allocation, and recovery, enabling consistent efficiency and stability even in large-scale environments. The system is designed to automatically handle many routine tasks, including repetitive failure responses and resource adjustments. NAVER is extending these operational technologies beyond GPUs to the broader infrastructure management of its data centers, achieving predictable infrastructure operations even under complex AI workload conditions. Looking ahead, the company plans to continue enhancing operational efficiency by integrating advanced AI technologies.
AI platform: “The brain powering AI operations”
Built on this robust operational foundation, NAVER’s AI platform functions as a unified operating system that connects the entire AI lifecycle—from model development and training to inference and deployment. Internally, both the training and operation of HyperCLOVA are conducted on this platform, which enables efficient control over GPU resource allocation, model management, and scheduling.
This enables developers to conduct training and experimentation without infrastructure constraints, while operators can monitor and analyze GPU usage and power efficiency in real-time, applying optimization strategies as needed. As a result, NAVER’s AI platform is not merely an execution environment—it serves as the integrated “brain” of AI development and operations.
NAVER’s AI Infrastructure Expands Through GPUaaS
NAVER Cloud is leveraging its accumulated technology and operational expertise to offer AI infrastructure services to major Korean companies through its GPU as a Service (GPUaaS) model, which is already being applied across various industrial fields.
CIO Lee stated, “NAVER Cloud will evolve its accumulated AI infrastructure capabilities into a GPUaaS model to create an ecosystem where Korean companies can easily adopt AI.” He added, “Through this, we aim to ensure that AI infrastructure grows beyond a single corporate asset to become a foundation for industry-wide growth.” (/End)
