NVIDIA, South Korean chip designer Rebellions and SK Telecom announced a collaboration on November 9 in Seoul to build sovereign AI infrastructure for Korea, linking Korean-designed processors directly into NVIDIA's rack-scale platform. The agreement, supported by Korea's Ministry of Science and ICT, was unveiled during NVIDIA AI Day Seoul and is structured as a development collaboration rather than a product launch.
Under the plan, Rebellions will adopt NVIDIA NVLink Fusion to connect its next-generation Rebel R200 neural processing units, or NPUs, to NVIDIA's open infrastructure platform. Rebellions will also build its rack-scale systems on NVIDIA's MGX rack architecture, which gives it access to an established ecosystem of server makers and component suppliers and a shared design for rack footprint, power delivery and cooling. SK Telecom plans to deploy the resulting systems in its own AI data centers.
NVLink Fusion is NVIDIA's program for extending its scale-up networking to chips it did not design. It lets custom processors, which NVIDIA calls XPUs, plug into the same high-bandwidth, low-latency NVLink fabric that connects NVIDIA's own GPUs inside a rack. For Rebellions, that removes the need to build a competing rack interconnect from scratch, and it places a domestic Korean inference chip inside the dominant AI rack design instead of alongside it.
The division of labor the companies describe is a split inference pipeline. NVIDIA GPUs would handle prefill, the stage in which a model processes a prompt, while Rebellions NPUs would handle decode, the stage in which the model generates its reply token by token. Rebellions says the R200 is designed for low-latency, energy-efficient inference on frontier models, including multitrillion-parameter mixture-of-experts models. SK Telecom plans to connect the systems to its proprietary foundation models, including A.X K2, and to public-facing services under Korea's "AI for All" project.
The sovereign AI framing is central to the announcement. The three companies said the goal is full-stack sovereign capability spanning infrastructure, proprietary models and services, with AI services tailored to Korea's language, industries and priorities. The Ministry of Science and ICT is backing the effort as a way to connect domestic semiconductor innovation with the systems and operational capabilities needed to deliver AI at scale.
Enterprise and data center operators should read the announcement as a roadmap commitment, not shipping hardware. Rebellions' R200 systems are expected to be ready for deployment in 2028, and the companies published no performance figures, power numbers, prices, rack counts or memory specifications for the R200. The technical question that will decide whether the split pipeline works in production, how quickly attention state moves between GPU and NPU during generation, was not answered with benchmarks in the release.
Even with those open questions, the deal matters for the wider AI infrastructure market. NVLink Fusion is becoming NVIDIA's answer to customers and governments that want local or specialized silicon without leaving its ecosystem, and each new XPU partner makes that ecosystem harder to displace. For SK Telecom, the collaboration offers a path to run national AI services on a mix of global and domestic chips. For Rebellions, it is a distribution opening that a standalone chip launch could not provide. Whether that opening turns into production systems now depends on the R200 arriving on schedule in 2028 and on measured results that have yet to be published.