NVIDIA, Rebellions and SK Telecom Partner to Advance South Korea's Sovereign AI Infrastructure
NVIDIA, South Korean AI semiconductor company Rebellions and telecommunications provider SK Telecom have announced a collaboration to advance South Korea's sovereign artificial intelligence infrastructure.
Supported by South Korea's Ministry of Science and ICT (MSIT), the initiative aims to strengthen the country's domestic AI computing ecosystem and support the development of AI services tailored to local languages, industries and national priorities.
Rebellions to Integrate Its R200 Processors With NVIDIA NVLink Fusion
A key part of the collaboration involves Rebellions adopting NVIDIA NVLink Fusion to connect its next-generation Rebel R200 neural processing units (NPUs) with NVIDIA's open AI infrastructure platform.
The integration will also use NVIDIA's MGX rack architecture and reference designs to support rack-scale AI inference systems.
Rebellions is developing the R200 NPU to support low-latency and energy-efficient inference for advanced AI models, including large mixture-of-experts models.
Through NVLink Fusion, the company aims to combine its specialised processors with NVIDIA's high-speed networking technology and established infrastructure ecosystem.
SK Telecom Plans AI Data Centre Deployment
SK Telecom plans to deploy the integrated systems in its AI data centre, combining computing infrastructure with its proprietary foundation models and public-facing AI services.
The company intends to connect these capabilities with initiatives under South Korea's government-backed AI for All project.
Its existing AI model portfolio includes A.X K2, which forms part of its broader strategy to develop a full-stack AI ecosystem spanning infrastructure, models and applications.
The collaboration is expected to support AI services designed for South Korean businesses, public institutions and consumers.
Combining GPUs and NPUs for Efficient AI Inference
The companies also plan to explore disaggregated inference, an approach that assigns different stages of AI response generation to different processor architectures.
Under the proposed approach, NVIDIA GPUs would handle the prefill stage, while Rebellions NPUs would manage the decode stage.
This configuration is intended to improve inference throughput and power efficiency by assigning computing tasks to hardware suited to each operation.
The NVIDIA MGX architecture will also provide a common infrastructure foundation for GPU and NPU systems, supporting shared approaches to rack design, power delivery and cooling.
South Korea Strengthens Its Sovereign AI Strategy
The collaboration is supported by South Korea's Ministry of Science and ICT, which is working to strengthen domestic semiconductor development and AI infrastructure capabilities.
Sovereign AI refers to a country's ability to develop, operate and control AI systems using infrastructure, data, models and technology aligned with its national priorities.
By connecting locally developed processors and AI models with NVIDIA's infrastructure platform, the partnership aims to strengthen South Korea's ability to build and operate its own AI systems.
The initiative also reflects the government's broader ambition to develop domestic AI capabilities while maintaining cooperation with global technology companies.
Deployment Expected in 2028
Rebellions' R200 systems are expected to be ready for deployment in 2028.
The collaboration highlights the growing role of specialised AI processors alongside GPUs as organisations look for more efficient ways to run large-scale AI workloads.
For South Korea, the partnership represents an effort to connect domestic semiconductor innovation with established global AI infrastructure technologies.
As demand for AI computing continues to increase, collaborations between chip developers, cloud infrastructure providers and telecommunications companies could play an important role in shaping the next generation of sovereign AI ecosystems.
