NAVER to Build Korea’s Sovereign AI “Factories” Using NVIDIA DSX, Starting at 55 MW

NAVER, one of South Korea’s largest internet and cloud providers, said it will expand its sovereign AI infrastructure using NVIDIA’s DSX platform, aiming to move from AI experimentation to production-scale “AI factories” designed to train and run increasingly capable models and agentic services. The announcement, made June 7, comes as companies worldwide compete for the hardware capacity and power efficiency needed to support high-volume AI workloads.

Under the plan, NAVER will begin with a 55-megawatt expansion of its GAK Sejong hyperscale data center in Sejong, South Korea, and intends to scale up toward gigawatt-class deployments. NVIDIA described DSX as an end-to-end, co-designed infrastructure stack that covers the full AI factory lifecycle, spanning compute, systems, infrastructure software, and facilities considerations such as power and operations.

NVIDIA said DSX is intended to help reduce “token cost,” which is a way of describing the cost to generate or process each unit of AI output at scale. In NAVER’s case, NVIDIA said the DSX-based approach pairs NAVER’s experience building and operating large proprietary GPU clusters and hyperscale data centers with NVIDIA’s platform aimed at speeding time to first production while supporting reliability and efficient operations.

The companies also linked the infrastructure expansion to NAVER’s regional model and agent roadmap. NVIDIA said NAVER will use NVIDIA full-stack AI platforms, models, and software to advance regional AI efforts, including next-generation HyperCLOVA X models, work on NAVER’s “Seoul World Model,” and agentic AI services that can be connected to real-world digital and physical workflows.

HyperCLOVA X is NAVER’s next-generation large language model effort, while the Seoul World Model is described as a foundation model development initiative using NAVER’s proprietary urban street-view and spatial modeling data. NVIDIA further said NAVER will fine-tune an open model, Nemotron 3 Ultra, using its proprietary data to improve the model’s cultural and language fit for Korean and enterprise customers.

NVIDIA also said NAVER is a member of the NVIDIA Nemotron Coalition, an industry effort intended to support open model development across stages like pretraining, post-training, and reinforcement learning. Separately, NVIDIA said NAVER plans to launch an AI Agent Platform in Korea in the second half of 2026, and that the platform will be powered by NVIDIA NemoClaw “blueprints,” which NVIDIA positioned as reference materials for building agent platforms at production scale.

The DSX announcement focuses on infrastructure and software enablement rather than a detailed delivery contract. NAVER and NVIDIA did not disclose the financial terms of the initiative, the number or type of GPUs that will be deployed, the expected duration of the 55 MW buildout, or a public schedule of subsequent gigawatt-stage milestones.

For the market, the deal underscores how “AI factories” are becoming a central battleground for sovereign and enterprise AI, with infrastructure design and power efficiency treated as strategic differentiators rather than background utilities. It also indicates that NVIDIA’s DSX platform is being positioned not just as hardware support, but as a way for cloud providers to run large-scale models and agent workloads more consistently.

What to watch next is whether NAVER publishes deployment details for GAK Sejong and whether other South Korean cloud and industrial players align their AI infrastructure plans to similar DSX-based architectures. Additional disclosures could also come through partner announcements around model operations, agent tooling, and physical AI deployments that depend on the factory buildout.