Rebellions Plans R200 Integration for Korean Sovereign AI

By Rebecca Stone

NVIDIA announced on November 8, 2026, that Rebellions and SK Telecom expect systems using Rebellions’ next-generation R200 processors and NVLink Fusion to be ready for deployment in Korea in 2028.

The collaboration gives Korean-developed NPUs a defined route into NVIDIA-based AI factories and gives SK Telecom an operating environment for its proprietary models and public-facing services. It remains a development plan: the R200 is still under development, and the announcement provides no performance data from deployed systems.

Under the agreement, Rebellions takes responsibility for the R200 and the rack-scale design, while NVIDIA would provide NVLink Fusion and the MGX framework. SK Telecom would provide the data center and connect the systems with proprietary models and services.

That division would keep NVIDIA GPUs in the deployment while giving Rebellions a place inside NVIDIA’s wider infrastructure ecosystem. It would also make SK Telecom the operator linking Korean-developed silicon to models and public-facing services.

What will SK Telecom deploy in 2028?

Under the agreement, Rebellions will link its Rebel R200 neural processing units (NPUs) through NVLink Fusion and develop rack-scale systems around NVIDIA’s MGX architecture. SK Telecom plans to deploy those systems in its AI data center and connect them with its proprietary models and services.

The announcement sets 2028 as the target for deployment readiness, not a completed rollout. It establishes a development and deployment plan rather than infrastructure already in operation.

What has the Korean government committed?

The Ministry of Science and ICT is supporting collaboration between NVIDIA and Korean NPU companies, beginning with Rebellions and SK Telecom. Bae Kyung-hoon, Korea’s deputy prime minister and science and ICT minister, framed the agreement as an opening for domestic processors.

Bae said the collaboration would enable Korean-developed NPUs to “enter the computing infrastructure market alongside NVIDIA” through NVLink Fusion. He also said the government would support tangible results and further cooperation with NVIDIA.

Bae identified physical AI and securing GPU supply as areas for that further cooperation. The policy backing therefore sits alongside the technical and deployment commitments rather than replacing them.

Why does Rebellions need NVLink Fusion?

According to NVIDIA, NVLink Fusion extends its AI platform to custom XPUs, allowing specialized processors to use NVLink scale-up networking designed for high bandwidth and low latency. Rebellions intends to use that networking to connect R200 NPUs within a rack.

The arrangement plugs Rebellions’ own silicon into NVIDIA’s wider infrastructure ecosystem. Rebellions is developing the R200 for low-latency, energy-efficient inference with frontier models, including multitrillion-parameter mixture-of-experts models.

Those latency and efficiency goals are design objectives, not performance measurements from deployed systems. The announcement provides no speed, power or benchmark result for the proposed configuration.

How will NVIDIA GPUs and Rebellions NPUs share inference?

The planned disaggregated-inference design assigns separate stages of AI response generation to different processor architectures. NVIDIA GPUs would handle the prefill stage, while Rebellions NPUs would handle the decode stage.

The companies target improved inference throughput and power efficiency from the division. The design keeps NVIDIA GPUs in the system while giving Rebellions’ NPUs a specialized inference role.

No measured throughput or power result accompanies the announcement. The proposed R200, NVLink Fusion and rack design must therefore clear three gates before SK Telecom can use the processor with its models and services.

What does MGX add?

MGX addresses the physical infrastructure surrounding the processors. According to NVIDIA, its reference designs and ecosystem give Rebellions access to original design manufacturers and component suppliers.

GPU and NPU systems can also share data-center design elements such as rack footprint, power delivery and cooling. Those shared elements give SK Telecom a basis for facilities planning while Rebellions continues work on the processor and system.

How does SK Telecom build its sovereign AI stack?

SK Telecom plans to connect the infrastructure with proprietary foundation models, including A.X K2, and public-facing services developed under the Korean government’s AI for All project. Its stated goal is to support AI services tailored to Korea’s language, industries and priorities.

The company wants those capabilities to form a foundation for services across the public and private sectors. Its full-stack strategy spans data-center operations, proprietary models and the services built on them.

SK Telecom President and CEO Jung Jai-hun said combining those capabilities with Korean AI semiconductors and global infrastructure technology would strengthen the company’s sovereign AI competitiveness.

How does this fit SK Telecom’s larger AI Cloud plan?

The processor collaboration follows a separate June 7, 2026, announcement in which SK Telecom outlined a gigawatt-scale AI Cloud in Korea using NVIDIA’s DSX architecture. SK Telecom had scheduled the first AI factory under that plan to come online in 2027.

The company also planned to become an NVIDIA Cloud Partner. The R200 work sits alongside that broader infrastructure relationship, but it carries its own development and deployment schedule.

The pairing gives Korean-developed silicon a defined route into NVIDIA-based AI factories. SK Telecom supplies the operating environment and links the processors to proprietary models and services, while Rebellions develops the specialized hardware.

What must happen before deployment?

The announcement sets three conditions before SK Telecom’s services can use the R200: the companies must finish the hardware, integrate it and test the proposed efficiency gains.

• Rebellions must complete the R200 processor and the rack-scale systems built around NVIDIA’s MGX architecture.
• SK Telecom and its partners must integrate those systems into its AI data center and connect them with proprietary models and services.
• The companies must establish whether assigning prefill to NVIDIA GPUs and decode to Rebellions NPUs delivers the targeted inference throughput and power-efficiency gains.

The shared rack design makes those steps interdependent. MGX covers facilities such as rack footprint, power delivery and cooling, while NVLink Fusion would supply the processor interconnect and the R200 would take the specialized decode role.

Until the processors, racks and data-center integration are complete, the 2028 date remains a target for deployment readiness. The decisive test is whether the finished systems produce the targeted throughput and power-efficiency gains while operating with SK Telecom’s models and public-facing services.