Korean AI chips plan to share NVIDIA racks, with deployment expected in 2028
NVIDIA, Rebellions and SK Telecom plan a mixed-processor inference system using GPUs for prompt processing and Korean NPUs for token generation, with R200 systems expected to be deployable in 2028.
The November 9 Seoul agreement connects domestic chip ambitions to an existing international systems platform.
Rebellions plans to connect its developing Rebel R200 neural processing units through NVIDIA’s NVLink Fusion and use the MGX rack architecture. SK Telecom says the parties signed a memorandum to develop and validate an integrated GPU–NPU system. [1][2][3]
Their proposed division of work is specific. NVIDIA GPUs would perform prefill, the computation that processes an incoming prompt; Rebellions NPUs would perform decode, producing the response token by token. The companies aim to improve throughput and power efficiency by assigning the two stages to different processors. No measured improvement is reported in the announcement. [2][3]
This is a 2028 planning story, not capacity available to rent today. NVIDIA’s release says R200 systems are expected to be ready for deployment in 2028. SK Telecom plans to introduce the system into its data-center environment to test performance and stability and assess commercial use. A timetable and an MOU do not establish a delivered production service. [2][3][4]
Sources & details 4 sources
Discuss in your voice app
Take the reporting, visual explanations and original sources into a conversation. Copy the brief, paste it into your preferred assistant, then start its voice mode.
Paste the brief into your voice app to discuss this reporting.Reporting sources
- NVIDIA · sovereign-infrastructure collaboration
Newsroom accessed through its board-statement link after direct URL failed. Header November 8; syndicated release and Seoul event dated November 9.
- NVIDIA · November 9 syndicated announcement
Original issuer release via GlobeNewswire; explicitly states expected R200 deployment readiness in 2028.
- SK Telecom · November 9 Seoul MOU
Separate counterparty account in Korean: development/validation MOU, GPU prefill, NPU decode and planned testing. Commercial participant, not independent performance evaluation.
- Benzinga · Korean infrastructure agreement
November9 independently authored report corroborates plannedR200/NVLink/MGX integration and expected2028 readiness; repeats announcement, not an independent systems benchmark.
Images & figures
Planned GPU prefill and NPU decode share a rack-level architecture; readiness is expected in 2028.
Euangelion · original explanatory figure · Source · Usage terms