Naver — B200 4K AI Cluster

Completed in January 2026, Naver’s “B200 4K Cluster” is the largest AI computing cluster in South Korea at launch, featuring 4,000 NVIDIA B200 Blackwell GPUs in a tightly integrated high-density cluster. Located within Naver’s “GAK Sejong” (Global AI Campus, Sejong City) data center facility in Korea’s administrative capital, the cluster is specifically optimized for training and evolving HyperCLOVA X — Naver’s flagship Korean-language sovereign large language model.

The B200 Blackwell chips (each GPU draws up to 1,000W under load) require both advanced air cooling and direct liquid cooling infrastructure. GAK Sejong was purpose-built to support these requirements, with direct liquid cooling loops embedded in the server racks and high-voltage power delivery at the rack level. The cluster’s 40 MW power draw is significant for a single cluster within a larger campus.

HyperCLOVA X is South Korea’s primary domestically-trained large language model, developed to perform at the level of leading US models on Korean-language tasks while supporting Korean cultural context, legal and regulatory compliance requirements, and data sovereignty. Naver has positioned it as the foundation for enterprise AI services in Korea and for Korean-language AI applications globally. The B200 4K cluster represents a step-change in Naver’s domestic training capacity, enabling more frequent model iteration and larger training runs.

The GAK Sejong campus also houses Naver’s cloud infrastructure for CLOVA Studio (enterprise AI platform) and Naver’s AI services including search, image generation, and code assistance. South Korea’s government has supported domestic AI model development through the Korea AI Initiative (KAI), and Naver’s infrastructure investments align with national AI sovereignty goals.

Hardware: 4,000 NVIDIA B200 Blackwell GPUs · ~40 MW active power draw · NVLink interconnect · InfiniBand network fabric