SK Telecom will build an integrated computing system combining Nvidia graphics processing units (GPUs) with domestically developed neural processing units (NPUs) in its artificial intelligence data centers (AIDCs). Combining GPUs and NPUs would improve AI infrastructure efficiency while providing a foundation for Korean AI chipmakers to expand globally. SK Telecom, Rebellions and Nvidia signed a memorandum of understanding to develop an NVLink Fusion-based integrated GPU and NPU system at Nvidia AI Day Seoul 2026, held at COEX in Seoul on November 9. NVLink Fusion extends Nvidia’s GPU interconnection technology to accelerators from external partners. The three companies will connect Rebellions’ REBEL R200 NPU to Nvidia’s platform and develop a rack-level integrated system that operates alongside GPUs in data centers. As demand for AI inference increases, heterogeneous computing that combines GPUs and NPUs has emerged as a key component of next-generation AI infrastructure. The hybrid architecture improves speed and power efficiency by assigning compute-intensive prefill, or question analysis, to GPUs, and decode, or answer generation, which requires rapid retrieval of data from memory, to NPUs. SK Telecom will introduce the integrated system into its AIDCs to demonstrate performance and stability and is reviewing its commercial viability. Jung Suk-geun, head of SK Telecom’s AI CIC, presented a blueprint for Korean sovereign AI infrastructure at the event. The concept is to build an optimized AIDC combining the strengths of GPUs and NPUs and serve as a token production facility that public and private organizations can access through application programming interfaces (APIs). The Ministry of Science and ICT also said the collaboration is expected to help domestic companies enter the global AIDC market and secure new business opportunities in the AI inference market, where cost and power efficiency are priorities. Alongside this Nvidia collaboration, the government’s earlier cooperation with AMD has provided a foundation for expanding ecosystems that integrate domestic NPUs in both the UALink and NVLink camps. The government will support efforts to extend the benefits of the cooperation to other domestic NPU companies. “NVLink Fusion enables Korean application-specific integrated circuits (ASICs) to be combined with Nvidia’s AI factory platform to deliver high-performance AI infrastructure,” said Ashish Karandikar, Nvidia’s vice president for NVLink Fusion. “Nvidia will strengthen Korea’s sovereign AI capabilities and contribute to building a next-generation AI computing ecosystem.”
Original Korean article: ETNews
English brief by Korea Wind Intelligence, koreawindintel.com
