Nvidia, Rebellions and SK Telecom partner on Korean sovereign AI

Building a sovereign AI foundation

Nvidia, Rebellions, and SK Telecom have announced a formal collaboration to advance the infrastructure for sovereign AI in South Korea. Sovereign AI refers to AI systems designed to serve a specific nation's unique language, industrial needs, and national priorities. This project is backed by Korea’s Ministry of Science and ICT (MSIT) to ensure the country can produce its own intelligence.

The partnership seeks to bridge the gap between domestic semiconductor development and the massive operational scale required for modern AI. By connecting Korean hardware with established systems, the group hopes to power a new industrial revolution for the region. This effort is intended to create a full-stack ecosystem that keeps Korea's AI capabilities independent and competitive.

According to the press release, the goal is to deliver tangible results that strengthen Korea's AI ecosystem. The government plans to actively support this collaboration to further advance cooperation with Nvidia. This includes working on areas like physical AI and securing the necessary supply of GPUs for the country.

How the hardware integrates

At the technical core of this deal is Nvidia's NVLink Fusion technology. This platform allows chipmakers to connect their own custom ASICs—application-specific integrated circuits—to the broader Nvidia AI factory platform. It enables the tight coupling of specialized processors with Nvidia's high-bandwidth and low-latency networking.

Rebellions will specifically use NVLink Fusion to connect its Rebel R200 neural processing units (NPUs) to the Nvidia platform. They will also utilize the Nvidia MGX rack architecture to develop systems at a rack-scale. This combination allows the team to use specialized silicon while benefiting from Nvidia's established infrastructure ecosystem.

Sunghyun Park, cofounder and CEO of Rebellions, noted that the integration aims to deliver low-latency and energy-efficient inference. By combining purpose-built silicon with high-speed interconnects, the partners can run frontier models at scale. This approach helps improve performance while reducing the risks and time typically associated with deploying new hardware.

Data centers and model deployment

While Rebellions provides the hardware, SK Telecom provides the operational backbone for the project. SK Telecom will deploy the systems using its extensive expertise in AI data center operations. They will also integrate their own proprietary models, such as the A.X K2, into the new infrastructure.

The collaboration will also leverage public-facing AI services that were developed under the 'AI for All' project. By linking these services with the new hardware and Nvidia's platform, the group aims to create a comprehensive AI service layer. This ensures that the sovereign AI is not just a set of chips, but a functional tool for the public and private sectors.

Jung Jai-hun, president and CEO of SK Telecom, stated that the goal is to strengthen their full-stack sovereign AI competitiveness. This means the company wants to control everything from the underlying data center to the final AI service provided to users. The integration is designed to support the growth of the broader Korean AI ecosystem.

Strategic goals for Korea

The collaboration is a strategic move to ensure that Korea's AI development is not entirely dependent on external providers. By using Nvidia's NVLink Fusion, Korean chipmakers can maintain their own designs while still accessing Nvidia's high-performance compute environment. This allows for a hybrid model where local innovation meets global infrastructure standards.

Nvidia's CEO Jensen Huang highlighted that this technology allows Korea's chipmakers to connect custom processors to Nvidia's AI factory. He stated that the partners are building the computing foundation for Korea to produce its own intelligence. This focus on domestic production is a key component of the country's technological strategy.

The project aims to address specific industrial priorities that might be overlooked by general-purpose AI models. By tailoring the infrastructure to Korean language and industries, the partners hope to create more relevant tools for local businesses. This move positions Korea as a significant player in the global race for specialized AI infrastructure.