NAVER Goes Gigawatt: A Sovereign-AI Factory Build on NVIDIA DSX, Starting at 55 MW

NVIDIA and South Korea's NAVER announced that the search-and-cloud giant will build sovereign AI factories on NVIDIA's DSX platform, beginning with a 55-megawatt deployment at its GAK Sejong data center and scaling toward gigawatt capacity. The plan, unveiled during Jensen Huang's visit to Seoul, is one of the largest sovereign-AI infrastructure commitments announced in Asia this year — a bet that the demand for domestically controlled compute is about to outrun anything a single nation has built before.

The ramp is deliberately staged. NAVER expects to bring 55 megawatts online in the first half of 2027, expand to 100 megawatts later that year, and reach 200 megawatts in 2028, with gigawatt scale as the stated long-term goal. The build runs on NVIDIA's DSX stack — an end-to-end blueprint spanning chips, systems, facilities, and software, including DSX MaxLPS to optimize token throughput per megawatt and a DSX OS layer for lifecycle management and multi-tenant operations. GAK Sejong, a next-generation hyperscale facility, is being engineered for the high-density, energy-efficient operation that frontier-scale clusters demand.

The capacity is pointed at NAVER's own model stack. The company will refine its next-generation HyperCLOVA X models using NVIDIA's Nemotron 3 Ultra, advance its Seoul World Model for urban simulation on the Cosmos foundation models, and launch an agentic AI platform in the second half of 2026 built on NVIDIA NemoClaw. Taken together, it is a full-stack sovereign-AI play: the silicon, the data center, the foundation models, and the agent layer all assembled under one national operator rather than rented piecemeal from a foreign hyperscaler.

NAVER is also explicit that GAK Sejong is not just for Korea. The company plans to serve AI demand across Asia, the Middle East, and Europe — positioning itself as a regional sovereign-AI provider for governments and enterprises wary of routing sensitive workloads through US clouds. "NAVER is building sovereign AI infrastructure that can serve Korea's industries," said founder and chairman Haejin Lee, while Huang framed the moment bluntly: "Useful AI has arrived, and demand for AI factories is extraordinary."

The announcement lands amid a global scramble for sovereign compute, from the Gulf states' gigawatt campuses to Europe's push for homegrown clouds. What separates NAVER's move is vertical integration: most national AI projects assemble these pieces from different vendors, but NAVER is wiring its models, its world simulator, and its agents directly into a single NVIDIA-powered fabric it owns and operates — and then offering the surplus to neighbors. If the 2027 milestones hold, GAK Sejong becomes one of the clearest tests yet of whether a national champion can stand up frontier-scale infrastructure on its own terms.