Validation of an NVLink Fusion-based GPU-NPU system
GPUs analyze questions; NPUs generate answers
"We will strengthen full-stack sovereign AI"

NVIDIA, SK Telecom and Rebellions are joining forces to build sovereign AI infrastructure. Using NVIDIA's AI chip interconnect technology, they will integrate Rebellions' neural processing units (NPUs) and NVIDIA's graphics processing units (GPUs) into SK Telecom's AI data centers. The plan is to efficiently supply "tokens," the units of data processed by AI, and expand the sovereign AI ecosystem.

Park Seonghyeon, CEO of Rebellions (from left); Jeong Seokgeun, chief technology officer (CTO) of SK Telecom; and Ashish Karandikar, vice president of NVLink Fusion at NVIDIA, pose for a photo after signing an MOU between NVIDIA, SK Telecom and Rebellions at COEX in Gangnam-gu, Seoul, on Nov. 9, 2026. Photo by Kim Hyunmin

Park Seonghyeon, CEO of Rebellions (from left); Jeong Seokgeun, chief technology officer (CTO) of SK Telecom; and Ashish Karandikar, vice president of NVLink Fusion at NVIDIA, pose for a photo after signing an MOU between NVIDIA, SK Telecom and Rebellions at COEX in Gangnam-gu, Seoul, on Nov. 9, 2026. Photo by Kim Hyunmin

View original image

NVIDIA, SK Telecom and Rebellions signed a "NVLink Fusion Cooperation Memorandum of Understanding" at the "NVIDIA AI Day" held at COEX in Gangnam-gu, Seoul, on Nov. 9. The signing ceremony was attended by Lee Do-gyu, director general of the Information and Communications Policy Office at the Ministry of Science and ICT; Raymond Te, senior vice president at NVIDIA; Jung Seok-geun, chief technology officer (CTO) of SK Telecom; and Park Sung-hyun, CEO of Rebellions.


The three companies will use NVIDIA's NVLink Fusion technology, which connects AI chips, to connect Rebellions' Rebel R200 NPU to the NVIDIA platform. Their plan is to improve the efficiency of AI inference by enabling heterogeneous computing that uses GPUs and NPUs together. SK Telecom will introduce the integrated system at its AI data centers and verify its performance.


The goal is to increase processing speed and power efficiency through "disaggregated inference": assigning the "prefill" stage, in which a question is analyzed during inference, to GPUs, which are suited to high-performance computing, and the "decode" stage, in which an answer is generated sequentially, to NPUs.


"By using NVLink Fusion to connect specialized chips to mature infrastructure, we can build customized AI factories and, ultimately, sovereign AI," said Jesse Clayton, senior product marketing manager for AI infrastructure at NVIDIA.


Domestic AI chip companies are expected to validate their chips' performance and potential applications while expanding overseas business opportunities. Park Sung-hyun, CEO of Rebellions, said the collaboration was chosen as a way to take NPU validation to the next stage. "This will be the first step toward setting the standard for Korea's sovereign AI ecosystem," Park said. "We will help technology developed in Korea grow into a key driver of AI around the world."


SK Telecom plans to complete a full-stack sovereign AI system spanning infrastructure, models and services by connecting its proprietary AI model and public-facing services such as AI for Everyone. "It is the role of infrastructure providers to create the optimal environment in which global GPUs and NPUs can both be used," said Jung Seok-geun, CTO of SK Telecom. "The idea is to create sovereign tokens." He added, "We will create an environment where people can use tokens without worrying about the hardware."


In line with the Ministry of Science and ICT's policies to promote sovereign AI and domestic AI chips, the three companies will establish a Korean sovereign AI infrastructure model and lay the foundation for sovereign AI services across Korea's public and private sectors.

SK Telecom Chief Technology Officer (CTO) Jeong Seokgeun speaks at a signing ceremony for an MOU between Nvidia, SK Telecom and Rebellions on NVLink at COEX in Gangnam-gu, Seoul, on Nov. 9, 2026. Photo by Kim Hyunmin

SK Telecom Chief Technology Officer (CTO) Jeong Seokgeun speaks at a signing ceremony for an MOU between Nvidia, SK Telecom and Rebellions on NVLink at COEX in Gangnam-gu, Seoul, on Nov. 9, 2026. Photo by Kim Hyunmin

View original image

A forum held afterward discussed ways for domestic AI startups to collaborate with NVIDIA. NVIDIA was represented by Vice Presidents Deepu Talla and Jonathan Cohen, among others. Twenty-two leading domestic AI startups, including Upstage, Realworld, TwelveLabs, FuriosaAI and Rebellions, also took part.



"Competition in AI chips today has evolved beyond chip performance. It is now about how effectively diverse semiconductors and technologies can be connected and AI infrastructure efficiently deployed in real-world markets," said Lee Do-gyu, director general of the Information and Communications Policy Office at the Ministry of Science and ICT. "I hope this will be a turning point that establishes Korea as an important pillar of the global AI chip ecosystem."


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing