Nvidia, Rebellions and SK Telecom to pair Korean AI chips with Nvidia GPUs

Story So Far

Nvidia, Rebellions and SK Telecom said on Monday they will work together to build sovereign AI infrastructure in South Korea, pairing Rebellions' locally designed processors with Nvidia's computing platform inside SK Telecom's data centres.

Rebellions will connect its next-generation Rebel R200 neural processing units to Nvidia's platform through NVLink Fusion and build rack-scale inference systems on Nvidia's MGX rack design. SK Telecom will deploy the systems in its AI data centre, linking them to its own models, including A.X K2, and to public services under the Korean government's AI for All project. The R200 systems are expected to be ready for deployment in 2028. The plan was unveiled at Nvidia AI Day Seoul 2026 at COEX in southern Seoul, with backing from South Korea's Ministry of Science and ICT.

The R200 is designed for low-latency, energy-efficient inference on frontier models, including multitrillion-parameter mixture-of-experts models. NVLink Fusion is the technology Nvidia offers to outside chipmakers so their own processors can plug into its high-speed scale-up networking, and the MGX design lets GPU and NPU systems share rack footprint, power delivery and cooling. The plan builds on an earlier June announcement of a gigawatt-scale AI cloud in Korea using Nvidia technology.

The three companies' joint release is the source for the plan, and it states plainly that the R200 systems are expected to be ready in 2028, so this is a development programme, not hardware already running. The release gives no order value, deployment size or performance figures, and none of the independent coverage found any either, so there is no price tag or benchmark to report. The ministry's backing is confirmed in the release and in Korean reporting of the signing ceremony at COEX, which names the officials and executives present. Claims about the chip's low latency and energy efficiency are the companies' design goals for a processor still being built, not measured results.

The deal gives Korean-designed silicon a defined route into Nvidia-based AI infrastructure. For Rebellions, a startup backed by investors including Samsung, SK Hynix, KT and Aramco, it is a path from chip design into live data-centre deployment. For SK Telecom, the systems are meant to anchor a full stack of Korean infrastructure, models and services, rather than relying entirely on imported hardware. For Nvidia, it extends its platform to a domestic chipmaker in a market where the government is pushing hard for AI self-reliance.

The ministry says the Rebellions project is the start of broader cooperation between Nvidia and Korean neural-processing-unit companies. The next test will be whether the mixed GPU and NPU racks, once built, deliver the efficiency gains the partners are aiming for when deployment begins in 2028.