NAVER, NVIDIA and Brookfield plan to expand Korea AI factory to 200 megawatts

NAVER, NVIDIA and Brookfield are planning to expand a Korean AI factory from an initial 55 megawatts to 200 megawatts by 2028, creating a large multi-tenant facility for companies building models, agents and AI-powered services. The proposed project would be located at NAVER's GAK Sejong hyperscale data centre and use NVIDIA's DSX platform. NAVER says the long-term ambition is to extend its sovereign AI infrastructure toward one gigawatt of capacity.

The announcement combines technology deployment with a financing plan. NVIDIA intends to invest 1 billion dollars in NAVER, while Brookfield has signed a nonbinding term sheet to fund up to 9 billion dollars. NAVER would fund the remaining amounts. NVIDIA's investment is subject to customary closing conditions and to NAVER securing at least 9 billion dollars of committed financing from other sources. Those conditions mean the published figures describe a proposed expansion rather than money already spent or capacity already online.

The facility is designed around a national infrastructure model. Instead of serving a single company, a multi-tenant AI factory would offer compute, software and support to Korean and US innovators, enterprises, industries and government organisations. The planned stack is expected to include NVIDIA Vera Rubin and Blackwell platforms, with DSX providing management for compute, facilities, resilience and multiple customers. The goal is to make large-scale AI resources available without requiring every organisation to build its own data centre.

For South Korea, the plan is part of a broader push to develop sovereign AI capability. NAVER operates one of the country's largest internet platforms and has experience running hyperscale cloud systems. The company is also adapting its HyperCLOVA X models with NVIDIA's Nemotron open models and plans to launch an agent platform in Korea later this year. A separate Seoul World Model project is intended to use urban street-view and spatial data for physical-AI applications.

Brookfield brings a different part of the equation: long-term capital and experience financing data centres, power generation and other infrastructure. AI factories require more than accelerators. They need electricity, cooling, networking, land, construction and operations teams, and the time required to assemble those pieces can be as important as chip supply. The proposed financing structure reflects how the AI build-out is increasingly being treated as an infrastructure investment rather than only a software project.

The expansion will still face practical questions. The partners must close the financing, obtain permits and power, deliver the equipment and prove that the facility can operate efficiently for several customers. A large capacity number also does not guarantee that startups will receive affordable access or that models trained there will compete globally. The value of a sovereign facility will depend on utilisation, software support and the ability to turn compute into products used outside the data centre.

If completed, the project would make Korea a notable test case for national AI factories and multi-tenant infrastructure. It would also deepen the relationship between a local platform company, a global chip provider and an international investor. The next milestones are financing commitments, construction progress and the first production workloads; those details will show whether the proposed 200-megawatt expansion becomes operating capacity or remains an ambitious plan.