NVIDIA, Rebellions, SK Telecom launch Korea sovereign AI tie-up

Rebellions plans to link its Rebel R200 NPU into NVIDIA’s platform via NVLink Fusion and to develop rack-scale systems based on NVIDIA’s MGX reference designs. The company described the R200 as a next-generation NPU optimized for low-latency, high-efficiency inference and capable of running state-of-the-art models, including Mixture-of-Experts (MoE) models with trillions of parameters. Rebellions said NVLink Fusion will connect R200 NPUs inside a rack to NVIDIA’s NVLink scale-up networking, while MGX rack references will support the system design.

SK Telecom will build the combined systems in its AI data centers, leveraging its experience in AI data-center construction and operation as well as its own foundation models and public AI services. The companies described using MGX rack architecture so GPU and NPU systems can share rack space, power, and cooling design elements. SK Telecom’s CEO Jung Jae-heon framed the project as a combination of global AI infrastructure technology and domestic AI semiconductors to validate next-generation sovereign AI infrastructure potential.

NVIDIA framed NVLink Fusion as a way to extend its open AI platform to custom XPUs, enabling semiconductor makers to attach purpose-built chips to NVIDIA’s high-bandwidth, low-latency scale-up networking and infrastructure ecosystem. The companies said that approach can raise performance, reduce deployment risk, and shorten time to market by integrating custom ASICs with the NVIDIA AI Factory platform. NVIDIA founder and CEO Jensen Huang commented that "Korea has the talent, technology and industrial capabilities to become a global leader in AI," and added that sovereign AI will let countries turn local languages, cultures and expertise into intelligence while allowing Korean semiconductor firms to connect tailored ASICs to NVIDIA’s platform via NVLink Fusion.

On technical orchestration, NVIDIA, Rebellions and SK Telecom said they will develop distributed inference that assigns each step of AI response generation to the processor architecture best suited to it. That plan includes configuring NVIDIA GPUs to handle prefill work and Rebellions NPUs to perform decode tasks in order to increase inference throughput and power efficiency. Rebellions’ co-founder and CEO Park Seong-hyun described the announcement as "a turning point" that links Korean AI semiconductors and models with global system technology; he said combining Rebellions’ custom chips and system design with NVIDIA’s NVLink Fusion and SK Telecom’s model and data-center capabilities will enable low-latency, highly efficient large-scale model inference.

South Korea’s government also commented: MSIT Second Vice Minister Ryu Je-myeong said the partnership creates an opportunity for domestically developed NPUs to enter global AI computing markets by connecting with a leading AI vendor through NVLink Fusion. The government stated it will actively support the project to help it yield tangible results and plans to expand cooperation with NVIDIA on areas including physical AI and securing GPU supply. Separately, the companies noted that deployment of Rebellions’ R200 systems could begin in 2028.