NAVER expands sovereign AI infrastructure with NVIDIA DSX platform
NAVER has announced a significant expansion of its AI infrastructure in partnership with NVIDIA, deploying the NVIDIA DSX platform to build out what both companies describe as sovereign AI factory capacity. The rollout begins with 55 megawatts of new capacity at NAVER's GAK Sejong hyperscale data centre in Sejong, South Korea, with a stated ambition to scale to gigawatt levels over time.
The NVIDIA DSX platform is positioned as a full-stack, co-designed blueprint for AI factories, covering chips, systems, software, facilities and partner integrations. NVIDIA's DSX MaxLPS software is designed to maximise token throughput per megawatt — a metric that is becoming a primary competitive variable as inference workloads dominate enterprise AI spending. NAVER will also run NVIDIA DSX OS, which provides lifecycle management, multi-tenant orchestration and health automation across the factory infrastructure.
Model development and the Nemotron coalition
Beyond raw infrastructure, the announcement deepens a software and model development collaboration between the two companies. NAVER is fine-tuning NVIDIA's Nemotron 3 Ultra open model using its proprietary Korean-language data and training pipelines, producing the next generation of its HyperCLOVA X models. The company says the resulting models will serve both domestic Korean enterprise customers and sovereign AI programmes in Europe and the Middle East.
NAVER is noted as the first Korean company to join the NVIDIA Nemotron Coalition, a partner programme spanning pretraining, post-training and reinforcement-learning contributions to open model development. NAVER also plans to launch an AI Agent Platform in the second half of 2026, built on NVIDIA's NemoClaw blueprints, and is separately developing a Seoul World Model — a physical AI application drawing on NAVER's urban street-view data and spatial modelling capabilities, built on NVIDIA's Cosmos world foundation models.
Haejin Lee, founder and chairman of NAVER, said the GAK Sejong expansion would allow customers "to move from AI experimentation to production-scale AI factories that power models, agents and real-world services."
Market and competitive context
The announcement is part of a broader pattern of NVIDIA cementing relationships with regional sovereign AI programmes ahead of hyperscaler competition. South Korea's government has been explicit about building domestic AI capacity, and NAVER — which operates one of Asia's largest indigenous large language model programmes — is the natural anchor tenant for that ambition. Comparable sovereign AI factory deals have been announced in France (with Mistral and state-backed investors), the UAE (G42 and the Falcon programme) and Saudi Arabia.
For NAVER, the deal also strengthens its competitive positioning against Korean rival Kakao and against global cloud providers such as AWS, Azure and Google Cloud, which are all expanding Korean data-centre footprints. NAVER Cloud Platform's ability to offer customers an end-to-end stack — from LLM training through inference and agentic services — on domestically operated infrastructure is a meaningful differentiator for government and regulated-industry buyers.
The digital sovereignty angle carries regulatory weight: South Korea's data-localisation requirements and emerging AI governance frameworks — influenced by both the EU AI Act and domestic legislation — make on-shore, trusted-operator infrastructure a commercial as well as a compliance proposition. NAVER's existing presence in Europe and the Middle East suggests the company is positioning GAK Sejong as a template it can replicate in other jurisdictions with similar sovereignty requirements.
Neither company disclosed contract value, capital expenditure commitments, a timeline for reaching gigawatt-scale capacity, or specific customer commitments beyond the general enterprise and government categories. Investors and analysts will look for those metrics in NAVER's forthcoming earnings disclosures and NVIDIA's partner pipeline updates.