South Korea Pushes Its AI Factory to 200 Megawatts as Capital Moves Into Sovereign Compute
NAVER, NVIDIA, and Brookfield plan to expand the AI factory in Sejong, South Korea, from 55 megawatts to 200 megawatts, backed by a financing commitm…
On July 24, at an AI summit in San Francisco, a data center in Sejong, South Korea, was redefined: it is not just NAVER’s cloud computing facility, but also the compute foundation of South Korea’s “sovereign AI” strategy. NAVER, NVIDIA, and Brookfield announced plans to expand the NVIDIA DSX AI factory inside the GAK Sejong data center from the previously disclosed 55 megawatts to 200 megawatts, with the goal of completing the expansion by 2028.
What happened
The capital structure of the plan is also worth noting. NVIDIA plans to invest $1 billion in NAVER; Brookfield has signed a non-binding term sheet and intends to provide up to $9 billion in financing for the project; NAVER will cover the remaining funding. According to the official announcement, the project is worth about $10 billion, and the expanded facility will provide production-grade compute for AI companies in South Korea and the United States to develop models, AI agents, and intelligent services.
This is not simply a data center expansion. NAVER said it will continue pushing AI infrastructure toward the 1 gigawatt scale and will provide a complete system from chips and networking to software through the NVIDIA DSX platform. The project is still in the planning and financing finalization stage, NVIDIA’s investment remains subject to closing conditions, and Brookfield’s term sheet is not a final binding financing agreement.
Why it matters
Over the past two years, the main themes in AI infrastructure have usually been chip procurement, power targets, and cloud revenue. This deal adds another layer: infrastructure investors provide long-term capital for compute factories, chip companies participate in asset and ecosystem development, local platforms handle operations and customer distribution, and governments provide demand and strategic direction through industrial policy.
This means compute is gaining infrastructure characteristics similar to electricity, communications networks, and industrial parks. For South Korea, a 200-megawatt scale could help local model companies, manufacturers, and government agencies reduce dependence on overseas cloud resources. For NVIDIA, direct investment is not just about selling GPUs; it is also about promoting an AI factory standard defined by its hardware and software platform. For Brookfield, AI compute becomes a new asset class that digital infrastructure capital can allocate to.
But “sovereign” does not mean fully domestic. The project still depends on NVIDIA’s accelerated computing platform, overseas capital, and the global supply chain. It is more like deploying controllable compute nodes locally, rather than severing external technology ties. That distinction will determine whether South Korea can convert infrastructure advantages into advantages in models, robotics, and industrial applications.
What to watch
First, whether the financing commitment of up to $9 billion can be converted into a formal agreement, and whether the total investment and construction timeline change. Second, whether the power connection, land, network, and equipment deliveries required to expand from 55 megawatts to 200 megawatts can be completed in sync. Third, who will use the compute once it comes online, at what price it will be rented out, and whether NAVER can raise the data center utilization rate to a level sufficient to support capital returns.
More long term, will South Korea see multiple “national AI factories” run by different companies but dependent on similar chips and platforms? If the answer is yes, then the core of AI competition will no longer be model performance alone, but whether power, financing, customers, and the software ecosystem can be organized into a sustainably operating infrastructure.
Sources
Information only. No investment, legal, tax, or financial advice.