Naver builds Korea’s largest AI computing cluster with Nvidia chips

Naver’s headquarters in Seongnam, Gyeonggi Province (Newsis)

Naver said Thursday it has completed construction of Korea’s largest artificial intelligence computing cluster, a move aimed at sharply accelerating development of its in-house AI models.

The nation’s biggest online portal operator added the system is built on about 4,000 of Nvidia’s next-generation B200 graphics processing units, giving it computing power comparable to that of the world’s top 500 supercomputers.

Naver said the new infrastructure is expected to speed up AI model development by about 12 times, based on internal simulations. The company’s research team reported that training a model with 72 billion parameters, which previously took about 18 months on an Nvidia A100-based cluster with 2,048 chips, now takes roughly six weeks on the new infrastructure.

The company said the cluster lays the groundwork for advancing its proprietary foundation models and expanding the use of AI across its services and industries. It plans to scale up training of its Omni model, which processes images, video and audio as well as text, with the goal of achieving global-level performance and gradually deploying the technology across its platforms.

“We have secured key assets that support national AI competitiveness and sovereignty,” Naver Corp. CEO Choi Soo-yeon said. “With infrastructure that enables rapid learning and iterative experimentation, we will apply AI technology more flexibly across services and industrial fields.”