Rebellions plans R200 inference systems with NVIDIA and SK Telecom
The AMW Read
The announced chip and rack partnership advances earlier talks and ties inference hardware to South Korea's government-supported sovereign AI plans.
Rebellions plans R200 inference systems with NVIDIA and SK Telecom
Rebellions, NVIDIA and SK Telecom announced a collaboration to build infrastructure for South Korea's sovereign AI ambitions. Rebellions plans to connect its next-generation R200 neural processing units through NVIDIA NVLink Fusion and use NVIDIA's MGX rack architecture. SK Telecom plans to deploy the systems in its AI data centers and connect them to its models and services. The companies expect R200 systems to be available for deployment in 2028.
The agreement moves Rebellions from previously reported early talks with NVIDIA to an announced technical partnership. It also shows how a domestic chip designer can pursue sovereign AI deployment while relying on a global supplier's networking and rack platform. South Korea's Ministry of Science and ICT supports the collaboration, which aims to link Korean-designed chips and models with services for local needs. The arrangement could give Rebellions a route into larger inference systems, while keeping NVIDIA infrastructure central to the design.
For builders and investors, the key test is whether the proposed split inference design delivers useful throughput and power efficiency in practice. The companies plan to use NVIDIA GPUs for prefill, when a model processes the input, and Rebellions NPUs for decode, when it generates the response. They have not reported results for that design in the announcement. Deployment, integration and performance remain milestones to assess as the 2028 availability target approaches.
