NAVER’s 55-Megawatt NVIDIA Buildout Tests Sovereign AI Cloud Demand

NAVER is expanding its sovereign AI infrastructure with NVIDIA, beginning with a 55-megawatt buildout and an ambition to scale toward gigawatt-class capacity.

The announcement puts Korea’s AI cloud ambitions into a concrete data-center frame: power, platform design, model localization, and enterprise demand all have to line up before sovereign AI becomes a production business rather than a policy slogan.

A 55-Megawatt Starting Point For Sovereign AI

NVIDIA and NAVER said the expansion will start at 55 megawatts and use the NVIDIA DSX platform to design, build, and scale end-to-end AI platforms for enterprises, industries, and government users.

The companies described the plan as a route toward gigawatt scale, which makes the initial figure less a final capacity target than a first phase in a larger AI factory program.

The infrastructure will expand NAVER’s GAK Sejong data center in Sejong, South Korea.

That site is presented as a next-generation hyperscale facility for NAVER’s AI and cloud services, with high-density accelerated computing, automation, sustainability features, and disaster-response capabilities.

For customers, the relevant point is not just GPU availability.

It is whether the platform can support production workloads while meeting local data-sovereignty and regulatory expectations.

NVIDIA framed DSX as a full-stack system for AI factories, covering chips, systems, software, facilities, and partner technologies.

NAVER brings its own hyperscale data-center operations and proprietary GPU cluster experience.

The combination is meant to reduce deployment friction for AI infrastructure that has to serve models, agents, cloud customers, and regulated enterprise users.

Korea’s AI Cloud Push Moves Beyond Model Training

The plan is not limited to raw compute.

NAVER said the infrastructure will support sovereign AI models, agentic AI services, and its AI data-center business.

The company also said it serves sovereign AI demand across Europe and the Middle East, giving the expansion an international angle beyond Korea’s domestic market.

Model work remains part of the strategy.

NAVER is advancing HyperCLOVA X by fine-tuning NVIDIA Nemotron 3 Ultra with proprietary data and training expertise.

It also joined the NVIDIA Nemotron Coalition, becoming the first Korean company to participate in that open model development effort across pretraining, post-training, and reinforcement learning.

Using NVIDIA NemoClaw blueprints, NAVER has put a Korea AI Agent Platform on the calendar for the second half of the year.

Those details matter because sovereign AI infrastructure only becomes commercially useful if customers can run localized models and services on it.

A data center buildout creates the physical base.

Model adaptation, orchestration software, and customer deployment tools determine whether enterprises and public-sector users can move from experimentation into production.

The Enterprise Test Is Compliance, Capacity And Token Cost

NVIDIA’s DSX software pitch centers on token economics and operational management.

DSX MaxLPS is described as software for maximizing token throughput per megawatt, while DSX OS is positioned as an operating layer for lifecycle management, runtime operations, health automation, resiliency, and multi-tenant AI factory management.

That language points to the pressure facing AI cloud providers.

They need enough power and accelerated compute to support demand, but they also need predictable cost per token, uptime controls, and tenant separation for customers that cannot treat AI workloads as experimental side projects.

NAVER’s claim of a trusted alternative for secure, high-performance digital services will depend on those operational measures, especially for government and enterprise customers with data-sovereignty constraints.

The announcement also links the infrastructure expansion to physical AI.

NAVER is developing a Seoul World Model using proprietary urban street-view data and spatial modeling technology based on NVIDIA Cosmos world foundation models.

If that work progresses, the same infrastructure story could extend from model services into city-scale simulation and spatial AI services.

What To Watch As NAVER Scales The Platform

The first watchpoint is whether the 55-megawatt starting phase converts into usable cloud capacity for named government, enterprise, and AI customers.

The second is whether NAVER can move HyperCLOVA X, Nemotron-based work, and its planned AI Agent Platform into services that customers can deploy at scale.

The third is the power trajectory: a move from 55 megawatts toward gigawatt scale would require not only GPUs and software, but also data-center expansion, energy availability, and sustained demand.

the strategic signal is clear.

Sovereign AI is becoming a data-center and cloud execution contest, not only a model-building race.

NAVER and NVIDIA are presenting Korea as a production hub for AI factories, but the proof will come from capacity delivered, workloads migrated, and customers willing to run sensitive AI systems on the platform.