SK Telecom is constructing an integrated computing system that combines NVIDIA Graphics Processing Units (GPUs) with domestic Neural Processing Units (NPUs) in its AI Data Center (AIDC). By integrating GPUs and NPUs, the company aims to enhance AI infrastructure efficiency while laying the groundwork for the global expansion of domestic AI semiconductors.
SK Telecom, Rebellions, and NVIDIA signed a business agreement on the 9th at the 'NVIDIA AI Day Seoul 2026' held at COEX in Seoul to develop a GPU-NPU integrated system based on NVLink Fusion.
NVLink Fusion is a technology that extends NVIDIA's GPU interconnect capabilities to external partner accelerators. The three companies will connect Rebellions' NPU 'Rebellion R200' to the NVIDIA platform and develop a rack-level integrated system that operates alongside GPUs in data centers.
With the rise in AI inference demand, heterogeneous computing using both GPUs and NPUs has emerged as a key next-generation AI infrastructure solution. In this hybrid structure, GPUs handle prefill (question analysis), which requires heavy computation, while NPUs manage decode (answer generation), which demands rapid data retrieval from memory, thereby improving speed and power efficiency.
SK Telecom will introduce the integrated system into its AIDC to validate performance and stability and assess commercialization potential.
Jeong Seok-geun, head of SK Telecom's AI CIC, presented a blueprint for building a Korean sovereign AI infrastructure. The plan involves constructing an optimal AIDC that combines the strengths of GPUs and NPUs and serving as a token production hub accessible via APIs for public and private sectors.
The Ministry of Science and ICT also anticipates that this collaboration will help domestic companies enter the global AIDC market and secure new business opportunities in the AI inference market, which prioritizes cost and power efficiency.
The government previously collaborated with AMD to establish a foundation for expanding the ecosystem of domestic NPUs integrated with both UA-Link and NVLink. It will support the spread of these collaborative effects to other domestic NPU companies.
Ashish Karandikar, NVIDIA's VP of NVLink Fusion, stated, “NVLink Fusion enables high-performance AI infrastructure by combining Korea's ASICs with NVIDIA's AI Factory platform.” He added, “NVIDIA will contribute to strengthening Korea's sovereign AI capabilities and building the next-generation AI computing ecosystem.”
