[Edaily Reporter Kim Hyun-ah ] Rebellion, a South Korean AI semiconductor company, is joining NVIDIA’s XPU (multi-type AI semiconductor) ecosystem to target the global AI data center market. Utilizing NVIDIA’s high-speed chip interconnect technology, NVLink Fusion, the company is developing a system that combines Rebellion’s neural processing units (NPUs) with NVIDIA’s graphics processing units (GPUs) in the AI data centers being built by SKTelecom.

By partnering with Rebellion following its collaboration with U.S. AI semiconductor company d-Matrix, NVIDIA is expanding its GPU-centric AI infrastructure into a structure that combines various computing semiconductors, such as CPUs and NPUs. Rebellion is seeking to enter the global market with its next-generation chip, Rebellion 2 (tentative name), the successor to the Rebellion 100.

According to the Ministry of Science and ICT on the 9th, NVIDIA, SKTelecom, and Rebellion signed a Memorandum of Understanding (MOU) for NVLink Fusion collaboration at NVIDIA AI Day, held that day at COEX in Seoul. This is part of the collaboration to integrate domestically produced NPUs with NVIDIA systems, which the Ministry of Science and ICT and NVIDIA have been discussing.

The signing ceremony was attended by Lee Do-kyu, Director General of the Information and Communications Policy Bureau at the Ministry of Science and ICT; Raymond Teh, Senior Vice President of NVIDIA; Jeong Seok-geun, Chief Technology Officer (CTO) of SKTelecom(017670); and Park Seong-hyun, CEO of Rebellion.

Entering the NVIDIA Ecosystem with “Rebel 2,” the Successor to Rebel 100

This collaboration is significant in that it paves the way for domestically developed NPUs to be integrated into and utilized within NVIDIA’s AI computing systems.

NVLink Fusion is a platform that enables third-party CPUs and AI accelerators to be connected to NVIDIA’s AI computing systems. By leveraging this platform, Rebellion’s NPU can be integrated into NVIDIA’s rack-scale systems to create a heterogeneous computing environment where it operates alongside GPUs.

The collaboration focuses not on the current product, the Rebel 100, but on its successor, the Rebel 2 (tentative name). Therefore, this agreement does not mean that the Rebel 100 will be immediately integrated into NVIDIA systems. Details regarding actual data center deployment, supply volume, and the timing of commercialization will need to be finalized in the future.

Shin Sung-kyu, Chief Financial Officer (CFO) of Rebellion, said in a phone interview with E-Daily, “Following Dematrix, this collaboration positions Rebellion as a key partner in the XPU sector,” adding, “Rebellion is the only AI semiconductor company in Korea participating in such a partnership.”

He emphasized that this collaboration differs from NVIDIA simply bringing in multiple XPU companies as partners. He explained that since this agreement is a Memorandum of Understanding (MOU), it is not a binding supply contract with specific business terms or roles finalized.

Targeting the Global Market While Maintaining Proprietary Technology

The collaboration with NVIDIA does not mean that Rebellion will abandon its proprietary chip design or interconnect technology. Rebellion plans to maintain its own chip interconnect technology while simultaneously collaborating with the NVIDIA ecosystem.

The business interests of all three companies are intertwined in this collaboration. NVIDIA can expand its infrastructure business opportunities in the South Korean AI data center market by combining not only GPUs but also various computational semiconductors. SKTelecom secures the opportunity to implement heterogeneous computing systems—combining NVIDIA technology with domestically produced NPUs—in its data centers. Rebellion can validate next-generation products and gain business experience in large-scale data centers.

SKTelecom and SKSQUARE indirectly hold stakes in Rebellion through Safion Inc., the largest shareholder of the merged entity formed by Safion and Rebellion. Saphion Inc. holds approximately 26.05% of Rebellion’s shares. Based on this, SKTelecom’s indirect stake is approximately 16.3%, while SKSQUARE’s is approximately 3.26%. This collaboration is also significant because it allows SKTelecom to utilize Rebellion’s domestically produced NPU as it expands its data center business.

In particular, the implementation experience gained from domestic AI data centers could serve as a foundation for Rebellion’s entry into overseas markets. CFO Shin stated, “We are not abandoning our proprietary inter-chip connectivity technology to switch to the NVIDIA model,” adding, “We will maintain our own technology while also collaborating with the NVIDIA ecosystem.” Regarding the timeline for implementation in domestic and international AI data centers, he noted, “It would be more accurate to say 2 to 3 years rather than 3 to 4 years.”

Government to Support Global Collaboration for Domestic AI Companies

The government plans to expand this collaboration into an opportunity for the entire domestic AI semiconductor industry to enter overseas markets. Following its previous collaboration with AMD, the Ministry of Science and ICT (MSIT) intends to finalize plans for integrating domestically produced NPUs with NVIDIA’s technology and support global partnerships for other domestic AI semiconductor companies besides Rebellion.

Lee Do-kyu, Director General of the Information and Communications Policy Bureau at the Ministry of Science and ICT, said, “The competitiveness of the AI industry can grow even further when diverse capabilities are interconnected and converged, going beyond the development of individual technologies,” adding, “We hope this will lead to domestic companies identifying opportunities for technical and business cooperation with NVIDIA and successfully entering the global market together.”

On the same day, the Ministry of Science and ICT and NVIDIA also held a partnership forum with 22 domestic AI startups to discuss joint research, technology validation, commercialization, and strategies for overseas expansion. Domestic AI semiconductor companies such as FuriosaAI, DeepX, Mobilint, and HyperExcel participated in the forum.