On Nov. 9, SK Telecom, Nvidia and Rebellions sign MoU to link chips in Korean AI infrastructure
Key Facts
SK Telecom announced on Nov. 9 that it had signed a memorandum of understanding with Nvidia and Rebellions to develop a computing system combining Nvidia graphics processing units with Rebellions neural processing units, as part of efforts to build sovereign AI infrastructure in South Korea. The announcement sets a framework for cooperation and development; it is not a declaration that a system is complete or infrastructure has been built. The companies plan to connect Rebellions’ “Rebel R200” unit to Nvidia’s platform using NVLink Fusion technology. They will also use Nvidia’s MGX rack design to develop an integrated rack-scale system. The partnership aims to test the combination of Korean chips and Nvidia’s platform in data-center applications.
The proposed design distributes the two stages of inference between the processors: the GPU handles question analysis during prefill, while the NPU generates the answer during decoding. SK Telecom said the division is intended to increase processing speed and improve energy efficiency. The idea comes as attention grows on the cost of operating data centers and their ability to handle AI workloads. The company said the spread of agentic AI is increasing inference requests, making performance, energy efficiency and operating costs competitive factors for data centers. South Korea’s ministry sees the combination of NPUs and GPUs as potentially helping data centers balance performance with energy consumption.
SK Telecom’s chief technology officer said the collaboration tests the possibility of building sovereign Korean infrastructure that combines Nvidia’s GPU capabilities with Rebellions’ NPUs. He added that the goal is to verify whether this combination can improve inference performance and energy efficiency. Nvidia’s vice president responsible for NVLink Fusion technology said the technology enables Korean application-specific integrated circuits to be integrated with the company’s AI factory platform. The collaboration places Rebellions chips in a design based on Nvidia’s platform, with different roles assigned to each type of processor. The memorandum remains a development and testing step, while the companies have not announced operational results or commercial contracts arising from it in the details available.
SK Telecom will introduce the proposed system at an AI data center to test its performance and stability, then consider using it in live services and the possibility of marketing it commercially. The company said it intends to connect the system to its independent “A.X K2” model and public AI services as part of its “AI for Everyone” initiative. The plan aims to build an ecosystem spanning infrastructure, models and services, according to the company. Nvidia said Rebellions’ R200 systems are expected to be ready for deployment in 2028. That date sets a later timeline for development, while the company did not give an earlier date for the start of testing or a final decision on commercial use.
The memorandum comes against the backdrop of earlier cooperation between Nvidia and SK Telecom on AI infrastructure in Korea. In an announcement dated June 7, 2026, Nvidia said SK Telecom planned to build a gigawatt-scale AI cloud using the DSX platform, with the first AI factory scheduled to begin operations in 2027. The company envisioned the cloud supporting training, inference and agentic AI workloads, as well as sovereign, industrial and enterprise services in Korea, with a view to expanding to broader parts of Asia. The latest trading data available in the evidence show Nvidia shares closed at $186.64 on Nov. 6, 2026. That price relates to a session before the memorandum announcement and does not represent the market’s reaction on Nov. 9.