NVIDIA plans Korean AI infrastructure with Rebellions and SK Telecom

NVIDIA unveiled NVLink Fusion in May 2025 at Computex in Taipei: a way for other companies' chips to plug into NVIDIA's high-bandwidth scale-up networking instead of competing against it. The pitch was that chipmakers could build semi-custom AI infrastructure around NVIDIA's platform rather than outside it.

SK Telecom has worked with NVIDIA for years on AI infrastructure, and in June 2026 it announced plans for a gigawatt-scale AI cloud in Korea built on NVIDIA's DSX architecture, aiming for operations in 2027. Rebellions, meanwhile, already had silicon in production: Rebellions said in September 2026 that its ATOM-Max NPUs power several SK Telecom services handling about 14 million requests and more than 4 billion tokens a day — company-reported figures.

NVIDIA, Rebellions and SK Telecom announced a collaboration to advance Korea's sovereign AI infrastructure, using NVLink Fusion. Aju Press reports the plan was unveiled at NVIDIA AI Day Seoul 2026 at COEX. NVIDIA's release says the collaboration is supported by Korea's Ministry of Science and ICT. Rebellions will connect its next-generation R200 neural processing units to NVIDIA's platform through NVLink Fusion and build rack-scale systems on NVIDIA's MGX rack architecture. SK Telecom plans to deploy those systems in its AI data centers and tie them to its own models, including A.X K2, and public services under the government's AI for All project. Aju Press reports that the partners call the goal a “Sovereign AI Token Factory” — infrastructure serving the Korean language, local data and industry-specific demand.

“Korea has the talent, technology and industrial ambition to become a world leader in AI,” NVIDIA CEO Jensen Huang said in the announcement. “Sovereign AI gives nations the power to turn their language, culture and expertise into intelligence.” Rebellions CEO Sunghyun Park said the announcement “marks a turning point, opening a new chapter for sovereign AI.”

The engineering idea is disaggregated inference: split the work of generating an answer between the processor best suited to each stage. Under the proposed approach, NVIDIA GPUs would take prefill, the compute-heavy first pass over a prompt; Rebellions NPUs would take decode, generating the answer token by token. The partners say that split should raise throughput and power efficiency at once. The R200 is designed for low-latency, energy-efficient inference on frontier models, including multitrillion-parameter mixture-of-experts models — design goals for a chip still in development, not measured results.

Rebellions' R200 systems are expected to be ready for deployment in 2028, according to the companies. Between now and then, the work is integration: getting Korean-designed NPUs running inside NVIDIA racks at scale. The joint press release does not specify financial terms, purchase commitments or deployment volumes.

NVIDIA released Dynamo deployment recipes for four Korean open models on November 8 — SK Telecom's A.X K2, LG AI Research's K-EXAONE 2.0, Upstage's Solar Open 2 and Motif Technologies' Motif-3 — and LG opened free API access to its newer K-EXAONE 2.5 model on Monday. NVIDIA's release says Korea's Ministry of Science and ICT is supporting collaboration with Korean NPU companies, beginning with Rebellions and SK Telecom. The joint release does not state whether other Korean chipmakers will join, and the 2028 readiness date is a company expectation, not a completed deployment.