As global demand for sovereign artificial intelligence surge, internet giant NAVER and semiconductor leader NVIDIA have formalized a massive infrastructure partnership. Utilizing the modular NVIDIA DSX platform, the joint venture will scale physical AI factories toward gigawatt capacity starting with a 55-megawatt deployment at the GAK Sejong facility in South Korea.
The global race for artificial intelligence hardware is expanding beyond standard cloud services. Hyperscale operators are shifting their focus to local infrastructure that respects data sovereignty and local cultural contexts. Under a major partnership formalized in early June 2026, South Korean internet giant NAVER and chip maker NVIDIA are joining forces to build large-scale AI factories.
This initiative leverages the newly introduced NVIDIA DSX platform, designed to simplify and accelerate the deployment of high-density data centers. Starting with an initial expansion phase at NAVER's GAK Sejong data center, the collaboration establishes a long-term roadmap to scale infrastructure toward gigawatt-level capacity. This post breaks down the technical and strategic elements of this global partnership.
Hyperscale data centers require high-density power and cooling architectures to support next-generation accelerated compute clusters.
- Gigawatt-scale Roadmap: NAVER and NVIDIA will develop AI factories designed to scale toward a target capacity of 1 gigawatt.
- GAK Sejong Expansion: The partnership begins with a 55-megawatt deployment at NAVER's flagship GAK Sejong data center in early 2027.
- DSX Platform Integration: The project utilizes NVIDIA's full-stack DSX platform, integrating compute, modular software, and digital twins.
- Model Advancements: NAVER will leverage NVIDIA's Nemotron 3 Ultra open model to refine its proprietary HyperCLOVA X large language model.
- Global Outreach: NAVER will act as a core global partner to deploy sovereign AI infrastructure in the APAC region, Europe, and the Middle East.
The Sovereignty Directive: Why Nations Are Reclaiming Their Digital Intelligence
The concept of sovereign AI has transitioned from an academic discussion to a critical commercial and political directive. Governments and regional enterprises are increasingly reluctant to entrust their national data to foreign hyperscalers. They recognize that artificial intelligence models codify a society's language, history, and cultural nuances, making data autonomy a national asset. The strategic necessity of local control has driven the adoption of sovereign AI platforms.
By establishing localized AI factories, countries can ensure that their data remains within physical and legal borders. This protects against policy shifts by foreign corporations or governments while ensuring compliance with local data privacy mandates. NVIDIA has positioned itself as a key enabler of this trend, providing the hardware and software foundations necessary for regional partners to build independent systems. Citing this strategic shift, the CEO of NVIDIA shared the following perspective:
“Useful AI has arrived, and demand for AI factories is extraordinary. Every country needs to own the production of their own intelligence. You cannot allow that to be done by other people. It codifies your culture, your society's intelligence, your common sense, your history – you own your own data.”
— Jensen Huang, CEO, NVIDIA Keynote Address, June 2026
This political shift has created a growing market for specialized, high-performance infrastructure. Regional operators are no longer merely leasing cloud capacity from global giants; instead, they are investing in domestic facilities designed to run customized local models. NAVER's collaboration with NVIDIA represents a practical blueprint for how regional leaders can execute this strategy, combining local operational expertise with advanced accelerated computing architectures.
The Gigawatt-Scale Roadmap: Phased Capacity Upgrades at GAK Sejong
The collaboration between NAVER and NVIDIA will center on GAK Sejong, NAVER's flagship hyperscale data center in South Korea. Spanning 294,000 square meters, GAK Sejong is currently the largest data center facility in the country, designed to accommodate up to 600,000 server units. The site provides the physical space and power capacity necessary to host high-density AI factories. The expansion project will proceed in several key phases:
- Phase One (H1 2027): Deployment of an initial 55 megawatts of power capacity optimized for the NVIDIA DSX platform.
- Phase Two (Late 2027): Scaling the deployment to 100 megawatts to support broader commercial enterprise workloads.
- Phase Three (2028): Reaching a capacity of 200 megawatts, focusing on physical AI and digital twin simulations.
This phased roadmap is designed to test and optimize the infrastructure before scaling to its long-term target of 1 gigawatt of capacity. A 1-gigawatt deployment represents approximately four times the total current maximum capacity of the GAK Sejong site, requiring significant capital investment. The scale of this plan highlights the surging demand for AI computing resources in the Asian market.
By scaling capacity in increments, NAVER can manage its capital expenditures while ensuring that the grid infrastructure can support the massive power requirements. Hyperscale clusters generate extreme thermal loads, meaning that each phase of the expansion must be accompanied by advancements in facility-level cooling and power distribution technologies to prevent operational failures.
The DSX Platform Playbook: Decoupling Compute, Power, and Cooling
To construct and manage these gigawatt-scale AI factories, NAVER is integrating NVIDIA's newly introduced DSX platform. The DSX platform represents a full-stack architectural framework that treats the entire data center as a single system. This approach replaces the traditional method of designing compute, network, power, and cooling layers in isolation, which often leads to inefficiencies. The core components of the DSX platform include:
- NVIDIA DSX OS: An open-source, modular software suite for lifecycle management, multi-tenancy, and health automation.
- DSX Air and Digital Twins: Simulation software that utilizes the Omniverse platform to create physics-accurate digital twins of data centers.
- Reference Designs: Co-designed system specifications that align compute clusters with local facility constraints.
- DSX MaxLPS: Optimization software designed to maximize the volume of tokens generated per megawatt of power.
The integration of digital twin simulations allows NAVER to model the thermal and electrical behavior of the GAK Sejong expansion before physical installation begins. Engineers can simulate airflow patterns and cooling efficiency under maximum load, reducing the risk of hot spots. This simulation-first playbook accelerates design cycles and minimizes costly post-installation modifications.
Furthermore, the modular nature of the DSX OS ensures runtime consistency across different hardware configurations. This software layer automates health checks and resiliency protocols, allowing the data center to maintain operational stability even during hardware failures. The result is a highly resilient infrastructure that can support continuous training runs for large-scale AI models.
Blackwell at Scale: Training HyperCLOVA X With 4,000 GPUs
The hardware foundation of NAVER's AI infrastructure is built on NVIDIA's advanced GPU architectures. Earlier in 2026, NAVER completed the installation of a 4,000-unit cluster powered by NVIDIA Blackwell (B200) GPUs. This deployment significantly enhanced NAVER's internal computing resources, providing the performance necessary to train its proprietary HyperCLOVA X model. Key specifications of the Blackwell cluster include:
- GPU Architecture: NVIDIA B200 Blackwell chips utilizing advanced packaging and high-bandwidth memory.
- Interconnect Fabric: High-speed networking systems designed to minimize communication latency between nodes.
- Energy Efficiency: Liquid-cooling loops that reduce the power required to maintain optimal operating temperatures.
The Blackwell GPUs offer substantial improvements in processing efficiency compared to previous architectures, allowing NAVER to train larger models in less time. The integration of this 4,000-unit cluster into the broader DSX platform represents the foundation of NAVER's high-performance computing services, allowing local enterprises to deploy advanced generative models.
The efficiency gains of the B200 cluster are particularly important given the high power consumption of hyperscale data centers. By generating more tokens per watt, NAVER can reduce the operational costs of its AI services, making advanced computing resources more accessible to local startups and government agencies that operate under strict budget constraints.
The Seoul World Model: Enhancing Physical AI Through Cosmos and Omniverse
Beyond standard text and language processing, the NAVER-NVIDIA partnership focuses on the emerging field of physical AI. This refers to artificial intelligence systems that interact with the physical world, such as autonomous robots, self-driving delivery vehicles, and smart manufacturing systems. NAVER is using NVIDIA's Cosmos platform to build the "Seoul World Model," a digital twin of the metropolitan area. The project is detailed in the callout below:
The Seoul World Model and Spatial Computing: The Seoul World Model is a physics-accurate digital replication of Seoul's physical infrastructure, compiled using high-resolution spatial mapping data. By running this model within NVIDIA's Omniverse simulation environment, NAVER can train autonomous systems in a safe, virtual space, simulating real-world physics, lighting, and weather conditions before deploying physical robots onto city streets.
Physical AI requires a combination of high-precision spatial data and real-time physical simulation. By training autonomous systems in a simulated digital twin, developers can test edge cases that would be dangerous or impossible to replicate in the real world. This approach accelerates the development of autonomous delivery networks and smart factory systems in South Korea.
The collaboration on physical AI highlights the growing convergence of digital models and physical operations. As AI factories scale, the spatial data processed in facilities like GAK Sejong will play a central role in driving automation across industrial sectors, transforming logistics, construction, and urban planning.
Beyond the Peninsula: NAVER and NVIDIA's Joint Global Push
The strategic partnership between NAVER and NVIDIA is not confined to the South Korean domestic market. During high-level meetings at NAVER's "1784" headquarters in Seongnam, NAVER CEO Choi Soo-yeon and NVIDIA CEO Jensen Huang outlined a roadmap to jointly enter global markets. Under this business model, NAVER will act as a key partner, sharing the risks and returns of building sovereign AI infrastructure worldwide.
The target markets include regions that seek to build domestic AI capabilities while maintaining data sovereignty. NAVER's operational experience with GAK Sejong and its proprietary HyperCLOVA X model positions the company as a valuable partner for governments and enterprises in the Asia-Pacific region, the Middle East, and Europe. Commenting on the role of hyperscale data centers, NAVER's CEO shared the following perspective:
“Hyperscale data centers like GAK Sejong represent the core infrastructure for technological stability and efficiency. By integrating our operational expertise with NVIDIA's full-stack DSX platform, we are building a state-of-the-art foundation capable of scaling sovereign AI solutions globally.”
— Choi Soo-yeon, CEO, NAVER Press Briefing, June 2026
By partnering with NVIDIA, NAVER can offer a validated, turn-key solution for sovereign AI factories. This global push allows NAVER to export its cloud services and model architectures to new markets, diversifying its revenue streams while contributing to the expansion of a decentralized, regional AI ecosystem.
Comparing NAVER's GAK Hyperscale Data Center Generations
To understand the technological evolution that enables gigawatt-scale AI factories, it is helpful to compare NAVER's first dedicated data center, GAK Chuncheon, with its latest flagship facility, GAK Sejong. The table below evaluates the two generations of infrastructure across several key metrics, showcasing the scale of the transition:
| Infrastructure Metric | GAK Sejong (Hyperscale Generation) | GAK Chuncheon (First Generation) |
|---|---|---|
| Operational Commencement | 2023 ▲ Leading | 2013 ▼ Behind |
| Designed Power Capacity | Up to 270 MW ▲ Leading | ~40 MW ▼ Behind |
| Physical Campus Area | ~294,000 m² ▲ Leading | ~54,229 m² ▼ Behind |
| Max Server Capacity | ~600,000 units ▲ Leading | ~100,000 units ▼ Behind |
| Primary Cooling System | Advanced Hybrid/Chilled Water ▲ Leading | Eco-friendly Natural Air Cooling ≈ Parity |
| AI Hardware Support | NVIDIA Blackwell / DSX OS ▲ Leading | Legacy High-Performance Compute ▼ Behind |
The comparison highlights the dramatic increase in scale and complexity represented by GAK Sejong. With nearly seven times the power capacity and six times the server density of GAK Chuncheon, GAK Sejong was designed from the ground up to support the massive data and power demands of modern hyperscale computing. This infrastructure foundation is what enables NAVER to host NVIDIA's advanced DSX-driven AI factories.
Conclusion: Defining the Next Era of Decentralized AI Infrastructure
The partnership between NAVER and NVIDIA represents a significant shift in the deployment of global AI resources. As the demand for sovereign AI and local data control grows, the industry is moving away from a centralized model dominated by a few global providers. By scaling AI factories to gigawatt capacity, regional leaders can establish local digital ecosystems that support economic growth and protect national data autonomy. The GAK Sejong expansion and the integration of the DSX platform establish a practical playbook for this transition, defining the infrastructure models of the coming decade.
Ultimately, the success of this decentralized model will depend on the ability of regional operators to deliver cost-effective, high-performance computing services. If successful, this strategy will pave the way for a more diverse and resilient global technology landscape, ensuring that AI development reflects local cultures and interests.
Sources and References
- NVIDIA Newsroom - Strategic Partnership With NAVER to Build Gigawatt-Scale AI Factories: nvidia.com
- NAVER Corporate - Hyperscale Data Center Footprint and GAK Sejong Infrastructure: navercorp.com
- The Korea Herald - Regional Infrastructure Expansions and Sovereign AI Strategies: koreaherald.com
- Pulse News - High-Level Leadership Meetings and Global Market Expansion Plans: pulsenews.co.kr
- Chosun Ilbo - Technical Specifications and Phased Upgrade Timeline for GAK Sejong: chosun.com
Post a Comment