Rebellions Will Plug R200 Chips Into NVIDIA NVLink Fusion by 2028

NVIDIA, Rebellions and SK Telecom will connect Rebellions' next-generation R200 processors to NVLink Fusion, with systems ready for deployment in 2028, the companies announced on November 8, 2026. The deal gives Korean-designed silicon a defined route into NVIDIA-based AI factories — and gives NVIDIA's ecosystem a sovereign-AI foothold in Korea.

The agreement is a development and deployment plan, not infrastructure already in operation. Rebellions will link its Rebel R200 neural processing units through NVLink Fusion and build rack-scale systems around NVIDIA's MGX architecture. SK Telecom plans to deploy those systems in its AI data center and connect them to its proprietary models and services.

How does the R200 fit into NVIDIA racks?

Rebellions is designing the R200 for low-latency, energy-efficient inference on frontier models, including multitrillion-parameter mixture-of-experts systems. Those are design objectives for a chip still in development, not measured performance from deployed hardware.

NVLink Fusion extends NVIDIA's AI platform to custom XPUs, letting specialized processors use NVLink scale-up networking built for high bandwidth and low latency. Rebellions plans to use that networking to connect R200 NPUs inside a rack, slotting its own silicon into NVIDIA's wider infrastructure ecosystem.

The interconnect also enables a planned disaggregated-inference design, which assigns separate stages of AI response generation to different processor architectures:

• NVIDIA GPUs handle the prefill stage.
• Rebellions NPUs handle the decode stage.

The companies are targeting improved inference throughput and power efficiency from that split. The arrangement keeps NVIDIA GPUs in the system while carving out a specialized inference role for Rebellions' NPUs. In plain terms: nobody's replacing NVIDIA here — they're partitioning the workload.

MGX covers the physical layer. Its reference designs and ecosystem give Rebellions access to original design manufacturers and component suppliers, and GPU and NPU systems can share data center elements such as rack footprint, power delivery and cooling. That shared foundation gives SK Telecom a basis for facilities planning while Rebellions keeps developing the processor.

What is SK Telecom building?

SK Telecom's role goes beyond housing hardware. It plans to connect the infrastructure to proprietary foundation models, including A.X K2, and to public-facing services developed under the Korean government's AI for All project. The stated goal is AI services tailored to Korea's language, industries and priorities, eventually spanning public and private sectors.

The full-stack strategy covers data center operations, proprietary models and the services built on top. President and CEO Jung Jai-hun described combining those capabilities with Korean AI semiconductors and global infrastructure technology as a way to strengthen SK Telecom's sovereign AI competitiveness.

Government backing is explicit. Korea's Ministry of Science and ICT is supporting collaboration between NVIDIA and Korean NPU companies, starting with Rebellions and SK Telecom. Bae Kyung-hoon, Korea's deputy prime minister and science and ICT minister, said NVLink Fusion would enable Korean-developed NPUs to "enter the computing infrastructure market alongside NVIDIA." He added that the government would back tangible results and further cooperation with NVIDIA on physical AI and GPU supply.

What came before?

The November 8 announcement follows a separate June 7, 2026, deal in which SK Telecom outlined a gigawatt-scale AI Cloud in Korea built on NVIDIA's DSX architecture. The first AI factory under that plan was scheduled to come online in 2027, and SK Telecom was set to become an NVIDIA Cloud Partner. The R200 work runs alongside that broader relationship on its own timeline.

What has to happen before 2028?

Three things stand between the announcement and running services. The companies must finish the R200 processor and its rack systems, integrate them into SK Telecom's infrastructure, and prove that the proposed division of inference work between NVIDIA GPUs and Rebellions NPUs actually delivers the intended efficiency gains. Until then, the 2028 date is a target — not a shipped product.