NVIDIA and Rebellions plan shared inference racks for SK Telecom
NVIDIA, South Korea’s chip developer Rebellions and operator SK Telecom agreed to develop shared infrastructure for running artificial intelligence models. The plan connects Korean-designed processors through NVLink Fusion and divides response generation between processor types. SK Telecom intends to use the systems in its data centre, with Rebellions’ R200 systems expected to be ready for deployment in 2028.
Artificial Intelligence··Midday
Korean processors enter a shared rack design
NVIDIA, South Korean processor developer Rebellions and telecom operator SK Telecom announced an infrastructure collaboration on November 8. Rebellions plans to connect its next-generation R200 neural processing units, chips designed to run neural networks, through NVIDIA’s NVLink Fusion. SK Telecom intends to deploy the resulting systems in its artificial intelligence data centre. The agreement joins processor development, server design and operation within the same project.[1], [2]
Prompt processing and response generation are divided
The partners plan to investigate disaggregated inference: dividing the work of generating an answer between different processor architectures. In their proposed arrangement, NVIDIA graphics processors handle prefill, the stage that processes a user’s prompt. Rebellions processors handle decoding, which produces the response step by step. Rebellions is developing R200 for large models, including mixture-of-experts systems, which use specialised model components rather than activating every component for each task.[1]
Deployment readiness is targeted for 2028
The systems will use NVIDIA’s MGX rack architecture and reference designs. These provide common arrangements for rack footprint, power distribution and cooling, along with access to design manufacturers and component suppliers. SK Telecom plans to connect the infrastructure to its own foundation models and public-facing services. Rebellions’ R200 systems are expected to be ready for deployment in 2028. South Korea’s Ministry of Science and Information and Communication Technologies supports the collaboration.[1]
