Nvidia, Rebellions and SK Telecom announced on November 9 in Seoul that they will build Korean sovereign AI infrastructure around Nvidia NVLink Fusion, the interconnect Nvidia licenses so third-party processors can work inside its systems.
Under the agreement, Rebellions will use NVLink Fusion to connect its Rebel R200 neural processing units to Nvidia's platform, linked within a rack through NVLink scale-up networking. The companies say the R200 is designed for low-latency, energy-efficient inference on large models, including multitrillion-parameter mixture-of-experts models. Rebellions will build its rack-scale systems on Nvidia's MGX reference architecture, which also gives it access to the MGX supply chain.
The companies plan to develop disaggregated inference in which prefill runs on Nvidia GPUs and decode runs on Rebellions NPUs, with the aim of improving throughput and power efficiency. SK Telecom plans to deploy the systems in its AI data center and connect them to its own models, including A.X K2, and to public services under Korea's "AI for All" project. Rebellions' R200 systems are expected to be ready for deployment in 2028.
Korea's Ministry of Science and ICT is backing the collaboration. Deputy Prime Minister and Science Minister Bae Kyung-hoon said it "enables Korean-developed NPUs to enter the computing infrastructure market alongside NVIDIA." Nvidia CEO Jensen Huang said sovereign AI "gives nations the power to turn their language, culture and expertise into intelligence."
The arrangement is a concrete example of mixed-vendor inference racks: Nvidia keeps its networking and rack design at the center while a domestic chipmaker supplies part of the compute. For operators, splitting prefill and decode across different chips is a way to tune cost per token, since the two phases stress compute and memory differently.