Nvidia, Rebellions and SK Telecom agree NVLink Fusion chip deal

Nvidia, SK Telecom and Korean chipmaker Rebellions signed a memorandum of understanding to connect Rebellions’ neural processing units with Nvidia GPUs inside Nvidia data-center systems, South Korea’s Ministry of Science and ICT said via The Korea Herald. The deal uses Nvidia’s NVLink Fusion interconnect and targets Rebellions’ next-generation Rebel R200 NPUs, with R200 systems expected for deployment in 2028, according to Nvidia’s newsroom release.

The Korea Herald, citing the ministry, said the MOU was signed Monday at Nvidia’s AI Day Seoul event at Coex. Nvidia’s newsroom release is dated November 8, 2026; Korean outlets including the Herald and Seoul Economic Daily report the signing as November 9.

Under the agreement, Rebellions will use NVLink Fusion to link its R200 NPUs to Nvidia’s platform and the MGX rack architecture for rack-scale system development, Nvidia said. SK Telecom plans to deploy the systems in its AI data center. Seoul Economic Daily reported that SK Telecom will verify performance and stability of a combined Nvidia GPU and Rebellions NPU setup at its AIDC. Nvidia said the companies plan a disaggregated-inference approach that uses Nvidia GPUs for prefill and Rebellions NPUs for decode.

NVLink Fusion is Nvidia’s path for third-party accelerators into its rack-scale networking. The ministry told the Herald the arrangement could help Korean chipmakers move beyond standalone product development, and that it wants to extend similar partnerships to other domestic NPU makers. Seoul Economic Daily said the MOU is Nvidia’s first formalized cooperation with a Korean NPU company, following earlier NVLink Fusion work with chipmakers such as Marvell and MediaTek.

The 2028 target is an availability plan, not a shipment already completed. Nvidia’s release states: “Rebellions’ R200 systems are expected to be ready for deployment in 2028.” Codema’s earlier NVIDIA DSX POD coverage explains how Nvidia packages rack-scale AI factories; the Rebellions deal is about attaching a Korean NPU to that class of infrastructure via NVLink Fusion, not a DSX POD announcement. Separate Codema reporting on the HBM4 race among SK hynix, Samsung and Micron and AI data-centre power demand covers adjacent Korean memory supply and facility limits; neither is part of this MOU. Our AI chips and memory guide explains the GPUs, NPUs and HBM these rack-scale systems rely on.