Nvidia, Naver partner on gigawatt-scale AI infrastructure

Seoul, South Korea - Nvidia Corp. and Naver Corp. announced an agreement Monday to build artificial intelligence infrastructure using Nvidia's DSX platform.

The project is scheduled to begin operations in 2027 with an initial capacity of 55 megawatts, expanding to 100 megawatts later that year, 200 megawatts in 2028, and eventually reaching the gigawatt scale. 

One gigawatt represents approximately four times the maximum capacity of Naver's existing GAK Sejong hyperscale data center. 

The companies said they will jointly manage the venture's operational risks and financial performance.

The DSX platform combines Nvidia hardware and modular operating system software with Naver's graphics processing unit clusters. 

The infrastructure is intended to support sovereign AI requirements for corporate, industrial, and government clients in the Asia-Pacific region, Europe, and the Middle East, while complying with local data privacy regulations.

According to the companies' joint statements, the deployment uses Nvidia's DSX MaxLPS software to regulate token processing efficiency per megawatt and the DSX operating system for multi-tenant data center lifecycle management.

As part of the collaboration, Naver is fine-tuning Nvidia's Nemotron 3 Ultra open models using its proprietary datasets to update its HyperCLOVA X large language model. 

Naver is the first South Korean company to join the Nemotron Coalition, which focuses on open-source pre-training, post-training, and reinforcement learning development. 

The company plans to launch an AI agent platform in South Korea in the second half of 2026, built on Nvidia's NemoClaw blueprint.

Naver is also using Nvidia's Cosmos world foundation model alongside its internal street view and spatial modeling datasets to develop a simulated environment it calls the Seoul World Model.