NVIDIA, Rebellions and SK Telecom Build South Korea's Sovereign AI Infrastructure
NVIDIA, South Korean AI chipmaker Rebellions and telecommunications company SK Telecom are collaborating to develop South Korea's sovereign AI infrastructure using NVIDIA's NVLink Fusion platform. The partnership will connect Rebellions' Korean-designed Rebel R200 neural processing units (NPUs) to NVIDIA's computing ecosystem, with SK Telecom planning to deploy the integrated systems in its AI data centres.
The collaboration aims to combine locally developed AI processors, domestic AI models and data-centre operations with NVIDIA's infrastructure technology. Supported by South Korea's Ministry of Science and ICT, the initiative is intended to strengthen the country's ability to develop and operate AI systems around its own technological and national priorities. Rebellions' R200 systems are expected to be ready for deployment in 2028.
South Korea's Sovereign AI Infrastructure Gains New Partners
Under the collaboration, Rebellions plans to use NVIDIA NVLink Fusion to connect its next-generation Rebel R200 NPUs to NVIDIA's open infrastructure platform. The companies will also use NVIDIA's MGX rack architecture to develop integrated systems that combine GPU and NPU computing.
NVLink Fusion provides high-speed, low-latency connections between specialised processors and NVIDIA's computing infrastructure. The technology allows chipmakers to integrate their own processors into systems built around NVIDIA's networking and infrastructure ecosystem.
For Rebellions, the partnership creates a route to connect Korean-developed AI silicon with an established rack-scale computing platform. The MGX architecture also provides standardised designs for rack space, power delivery and cooling, which can simplify infrastructure development and deployment planning.
NVIDIA GPUs and Rebellions NPUs to Share AI Workloads
One of the central technical elements is a planned approach called disaggregated inference. Rather than assigning every stage of an AI response to the same processor, the companies intend to divide the workload according to the strengths of different chips.
Under the proposed arrangement, NVIDIA GPUs will handle prefill, the stage that processes an incoming prompt and prepares the information needed for generation. Rebellions' R200 NPUs will handle decoding, where the model generates its response token by token.
This division is intended to improve inference throughput and power efficiency. The approach could also help operators use specialised processors more effectively instead of relying on one type of accelerator for every computing task. However, the expected performance benefits will need to be established through system development and testing.
SK Telecom Connects AI Infrastructure With Local Models
SK Telecom will bring its AI data-centre operating experience, proprietary foundation models and public-facing services into the project. The company plans to deploy the integrated GPU-NPU systems in its AI data centres and assess their performance and practical use.
Its proprietary A.X K2 model and services developed under South Korea's government-backed AI for All initiative are among the capabilities that could connect to the infrastructure. This gives the collaboration a broader scope than processor integration alone, linking semiconductor design with computing facilities, AI models and services.
South Korea's Ministry of Science and ICT is supporting the collaboration as part of its efforts to strengthen domestic AI semiconductor development and sovereign computing capabilities. The government has also indicated support for wider cooperation involving Korean NPU companies and NVIDIA.
R200 Deployment Expected in 2028
The companies expect Rebellions' R200 systems to be ready for deployment in 2028. Before then, the partners must develop and validate the integrated systems and establish how the processors will operate together in data-centre environments.
The agreement marks a step towards connecting South Korea's domestic AI chip industry with global infrastructure technology. It does not mean the country will become independent of foreign computing platforms. Instead, the project aims to give Korean-developed processors and AI models a place within a broader ecosystem while allowing local companies to retain a stronger role in infrastructure and services.
For South Korea, the partnership offers a practical route to expand its sovereign AI capabilities. Its success will depend on the performance, efficiency and commercial adoption of the integrated systems as they move towards the expected 2028 deployment.
