Rebellions will connect its R200 NPUs to NVIDIA’s rack-scale platform through NVLink Fusion, with SK Telecom deploying the systems from 2028. Futurum examines what the decode-focused design gives each party and where Korea’s sovereign AI strategy now depends on NVIDIA’s fabric.
What Is Covered in This Article:
- NVLink Fusion collaboration for Korean sovereign AI infrastructure
- Rebel R200 NPUs on decode with NVIDIA GPUs on prefill in MGX racks
- SK Telecom deployment plans for A.X K2 and the AI for All project, with R200 systems expected in 2028
- MSIT backing for a broader Korean NPU program on NVIDIA’s platform
- Competitive position against Groq 3 LPX, d-Matrix Raptor, FuriosaAI, and the UALink and Ethernet camp
The News: NVIDIA, Rebellions, and SK Telecom announced a collaboration on November 8 to advance Korea’s sovereign AI infrastructure using NVIDIA NVLink Fusion. Rebellions will connect its next-generation Rebel R200 neural processing units to the NVIDIA platform and build rack-scale systems on the NVIDIA MGX architecture. The companies plan disaggregated inference, with NVIDIA GPUs handling prefill and R200 NPUs handling decode for frontier models, including multitrillion-parameter mixture-of-experts models. SK Telecom will deploy the systems in its AI data center and connect them to its A.X K2 model and to services under the government’s AI for All project. Korea’s Ministry of Science and ICT (MSIT) is supporting the effort as the first of its planned collaborations between NVIDIA and Korean NPU companies. R200 systems are expected to be ready for deployment in 2028.
“We will deliver low-latency and energy-efficient inference for frontier models at scale,” said Sunghyun Park, Cofounder and CEO of Rebellions. “With NVLink Fusion, Korea’s chipmakers can connect their custom ASICs to NVIDIA’s AI factory platform, tightly coupling their processors with NVIDIA compute to run AI workloads,” said Jensen Huang, Founder and CEO of NVIDIA.
NVIDIA Puts Korea’s Sovereign AI Ambitions on NVLink Fusion With Rebellions and SK Telecom
Analyst Take: NVIDIA has placed a Korean national champion inside its own rack. Rebellions took the first direct investment from the Korea National Growth Fund, roughly $166 million, in its March pre-IPO round, and MSIT now sponsors its entry into the NVIDIA platform. NVIDIA gains a decode tray inside a government-backed program, with every rack built to this design still shipping NVIDIA GPUs for prefill and NVLink switching for scale-up. Rebellions gains a validated MGX rack and a 2028 deployment commitment from a strategic shareholder. Missing from the announcement is any R200 specification or power figure. The deal settles the architecture for Korea’s sovereign inference racks two years before anyone can measure whether the disaggregated system inside them pays for itself.
SK Telecom Supplies the Production Record That R200 Will Need
Rebellions brings more carrier operating history than most inference startups can show. SK Telecom merged its Sapeon chip unit into Rebellions in 2024, and by September 2026 Rebellions ATOM-Max NPUs ran four commercial SK Telecom services handling more than 4 billion tokens a day, according to DIGITIMES. Those services cover compact workloads on a mixed GPU and NPU fleet. Multitrillion-parameter decode is harder, since every request moves the KV cache from NVIDIA GPU memory to Rebellions NPU memory over NVLink, and neither company has described how that handoff will work.
Disaggregated Decode Gives Rebellions a Defined Socket in NVIDIA’s Inference Lineup
NVIDIA now hosts decode specialists segmented by memory technology. The SRAM-based Groq 3 LPX is in full production for compact models and d-Matrix’s 3D-DRAM Raptor reaches MGX racks in Q4 2027. Rebellions adds an HBM design. Its current Rebel100 pairs four Samsung SF4X chiplets with 144 GB of HBM3E and 4.8 TB/s of bandwidth for 2 PFLOPS at FP8, and Rebellions claims H200-class performance at lower power. Futurum’s 1H 2026 Data Center Semiconductors Market Sizing and Five-Year Forecast projects agent- and reasoning-first inference silicon growing from $35.9 billion in 2025 to $546.0 billion by 2030, with the XPU sub-market expanding from $37.4 billion to $237.2 billion. NVIDIA books GPU and switch revenue on every disaggregated rack whoever wins the decode tray. Rebellions will build on production experience after Groq and Raptor have built installed bases.
Korea’s NPU Startups Split Across Rival Fabrics as the Economics Wait for 2028
The deal puts Korea’s two leading NPU startups in rival scale-up camps. FuriosaAI, valued at $2.9 billion after a $620 million Series D, plans to develop its next Renegade chip with Broadcom on Ethernet, the approach AMD and Intel also favor over NVLink Fusion. MSIT’s phrase “beginning with Rebellions” leaves room for Furiosa to follow. Bae Kyung-hoon, Korea Deputy Prime Minister and Minister of Science and ICT, tied the program to “securing GPU supply,” folding Korean NPUs into the same relationship as Korea’s 2025 commitment to more than 260,000 NVIDIA GPUs. A sovereign stack that licenses its fabric and prefill compute from one US vendor confines Korean silicon to the decode tray. Futurum’s hard ROI test also stays open, because NVIDIA’s release claims better throughput and power efficiency without numbers. Layered coexistence with a multi-silicon Sovereign AI strategy is the default outcome to monitor.
What to Watch:
- Whether Rebellions publishes R200 specifications and decode power figures before its IPO filing
- Whether SK Telecom commits a named R200 rack count or capacity figure for 2028
- Whether FuriosaAI or another XPU vendor joins the MSIT-backed NVLink Fusion program
- Whether R200 systems reach deployment on the 2028 schedule against Raptor’s Q4 2027 MGX availability
- Whether SK Telecom expands standalone RebelPOD deployments outside the MGX design
Read the announcement details on the NVIDIA website.
Sources
- NVIDIA, Rebellions and SK Telecom Announce Collaboration to Advance Sovereign AI Infrastructure With NVIDIA NVLink Fusion, NVIDIA
- 1H 2026 Data Center Semiconductors Market Sizing & Five-Year Forecast
Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.
Other Insights From Futurum:
Oracle Makes the Supercluster Multi-Silicon at AI World 2026
Cadence Maestro Puts Frontier-Class Reinforcement Learning Behind Chip Design Agents
Author Information
Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers.
Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research.
Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.
