AI news ·
Nvidia, Rebellions and SK Telecom partner to build Korean sovereign AI infrastructure with NVLink Fusion
NVIDIA, Rebellions, and SK Telecom will jointly build AI data center infrastructure in South Korea, with Rebellions' R200 NPUs connecting to NVIDIA's platform via NVLink Fusion for deployment by 2028.
NVIDIA, Rebellions and SK Telecom will jointly build AI data center infrastructure in South Korea, the companies said November 8, 2026. The deal connects Rebellions' Korean-designed neural processing units with NVIDIA's rack-scale platform and SK Telecom's data center operations, creating a path for locally developed silicon to power sovereign AI services at scale.
Rebellions plans to adopt NVIDIA NVLink Fusion to link its next-generation R200 NPUs to NVIDIA's open infrastructure platform. SK Telecom will deploy the resulting systems in its AI data center, drawing on proprietary foundation models and public-facing services developed under Korea's "AI for All" project. South Korea's Ministry of Science and ICT (MSIT) is backing the collaboration as part of a broader push to connect domestic semiconductor innovation with national AI infrastructure.
How the technology fits together
NVLink Fusion lets chipmakers combine custom silicon with NVIDIA's high-bandwidth, low-latency scale-up networking. Rebellions will use it to connect R200 NPUs within a rack through NVIDIA NVLink scale-up networking, while the NVIDIA MGX rack architecture provides standardized power delivery, cooling and footprint designs. The MGX ecosystem also gives Rebellions access to an established supply chain of original design manufacturers and component suppliers.
The R200 NPU targets low-latency, energy-efficient inference for frontier models, including multitrillion-parameter mixture-of-experts architectures. The companies plan to develop a disaggregated inference approach that assigns different stages of AI response generation to the processor best suited for each task - NVIDIA GPUs for prefill and Rebellions NPUs for decode - to improve throughput and power efficiency.
What the leaders said
"Korea has the talent, technology and industrial ambition to become a world leader in AI," said Jensen Huang, founder and CEO of NVIDIA. "Sovereign AI gives nations the power to turn their language, culture and expertise into intelligence. 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."
Sunghyun Park, cofounder and CEO of Rebellions, said the announcement "marks a turning point" for sovereign AI by linking Korean silicon with global systems technology. "By integrating Rebellions' purpose-built silicon and system design with NVIDIA's NVLink Fusion, high-speed interconnects and rack-scale design expertise, and SKT's model and data center operations capabilities, we will deliver low-latency and energy-efficient inference for frontier models at scale."
SK Telecom president and CEO Jung Jai-hun framed the collaboration as validation of next-generation sovereign AI infrastructure. The carrier plans to connect its AI data centers, proprietary models including A.X K2, and services from the "AI for All" project into a full-stack offering. For technology leaders navigating similar infrastructure decisions, AI Technology Leadership Courses cover the architectural tradeoffs involved in GPU-NPU hybrid deployments.
Government backing and timeline
Deputy Prime Minister Bae Kyung-hoon, who also serves as Minister of Science and ICT, said the government "will actively support this collaboration to deliver tangible results and further advance cooperation with NVIDIA across a range of areas, including physical AI and securing GPU supply." The ministry is supporting broader ties between NVIDIA and Korean NPU companies, starting with Rebellions and SK Telecom.
Rebellions' R200 systems are expected to be ready for deployment in 2028. The timeline gives SK Telecom room to align data center planning with processor development, using the standardized MGX rack architecture as a common foundation.
Why this matters for IT and operations leaders
The deal signals that custom AI silicon can enter the data center alongside GPUs through standardized interconnects and rack designs, rather than requiring entirely separate infrastructure. For teams planning AI capacity, the disaggregated inference approach - splitting prefill and decode across different processor types - offers a concrete template for improving throughput and power efficiency without waiting for a single chip to handle every workload. The 2028 deployment target also provides a realistic benchmark for how long silicon-to-rack integration takes when multiple vendors and a national government are involved.