TEAM NAVER Builds Korea’s Largest AI Cluster With 4,000 B200 GPUs, Accelerating AI Development 12-Fold
TEAM NAVER Builds Korea’s Largest AI Cluster With 4,000 B200 GPUs, Accelerating AI Development 12-Fold
TEAM NAVER Builds Korea’s Largest AI Cluster With 4,000 B200 GPUs, Accelerating AI Development 12-Fold
- Unrivaled expertise in designing ultra-high-performance GPU clusters secures a globally top-tier scale
- Training period reduced from 18 months to 1.5 months, enabling a 12-fold faster development cycle
- Lays the foundation for advancing proprietary foundation models and applying them across services and industries
January 8, 2026
TEAM NAVER announced on the 8th that it has completed Korea’s largest AI computing cluster, built with 4,000 next-generation NVIDIA B200 (Blackwell) GPUs. With this infrastructure, TEAM NAVER has secured global-level computing power while laying a core foundation for advancing its proprietary foundation models and flexibly applying AI technologies across services and industries.
Moving beyond simple hardware adoption, TEAM NAVER has unrivaled expertise in clustering, which connects large-scale GPU resources into a unified system to deliver optimal performance. Since becoming the world’s fastest company to commercialize NVIDIA’s SuperPOD supercomputing infrastructure in 2019, TEAM NAVER has accumulated proven experience in directly designing and operating ultra-high-performance GPU clusters.
The newly built B200 4K Cluster brings together cooling, power, and network optimization technologies based on this experience. Designed for large-scale parallel computing and high-speed communication, the cluster is considered to have a computing scale comparable to that of top-ranked supercomputers on the global Top500 list.
This overwhelming infrastructure performance directly translates into faster AI model development. According to the company, internal simulations showed that the B200 4K Cluster can reduce the training period for a 72-billion-parameter (72B) model from approximately 18 months on its existing main A100-based infrastructure with 2,048 GPUs to about 1.5 months. The figures are based on internal simulations, and actual training time may vary depending on specific training tasks and configurations.
With training efficiency improving by more than 12-fold, TEAM NAVER has established a development and operational framework that enables more experimentation and iterative training, improves model completeness, and allows the company to respond more nimbly to a changing technological environment. With infrastructure capable of rapidly repeating large-scale training now in place, the overall speed and flexibility of AI model development are expected to be further strengthened.
Leveraging this cluster, TEAM NAVER plans to accelerate the advancement of its proprietary foundation models. TEAM NAVER will expand large-scale training of Omni models, which can process text, images, video, and audio simultaneously, to raise AI performance to global standards and gradually apply these models across a wide range of services and industries.
“Establishing this AI infrastructure goes beyond a simple technology investment,” said Choi Soo-yeon, NAVER CEO. “It is significant in that we have secured a core asset that supports national AI competitiveness, self-reliance, and sovereignty. Based on infrastructure that enables rapid training and iterative experimentation, TEAM NAVER will create practical value by applying AI technology more flexibly across services and industrial fields.”
