AI news, problem solutions & tool picks
DigestX

AI News

NVIDIA, Rebellions and SK Telecom Team Up on Korean Sovereign AI

· 2 min read

Seoul skyline reflected in the Han River under a clear blue sky
Photo via Pexels

NVIDIA, Korean chip startup Rebellions and SK Telecom announced a deal on November 9 to build Korea's sovereign AI infrastructure, at NVIDIA AI Day Seoul. Rebellions will connect its next-generation Rebel R200 neural processing units to NVIDIA's platform through NVLink Fusion and the MGX rack architecture, and SK Telecom will deploy the combined systems in its AI data centers.

The systems will run SK Telecom's own models, including A.X K2, and its services, under the Korean government's "AI for All" project. The work is backed by Korea's Ministry of Science and ICT. The announcement was made at the COEX convention centre in Seoul, where NVIDIA is holding its first AI Day in Korea.

What the deal actually builds

The plan splits AI inference work between the two kinds of chip. NVIDIA GPUs handle the prefill step — reading and processing the prompt — while Rebellions' NPUs handle decode, generating the answer token by token. Rebellions says this disaggregated setup cuts the cost of each token, and its chief executive said the goal is to make Korea "a token exporter": a country that sells AI computing power abroad instead of buying it.

Jensen Huang, NVIDIA's chief executive, framed the deal in national terms: "Sovereign AI gives nations the power to turn their language, culture and expertise into intelligence." For Korea, that means AI systems trained on Korean language and data, running on infrastructure inside the country.

The honest catch: nothing ships until 2028

The R200 systems are not expected to be ready for deployment until 2028. The announcement is a collaboration agreement and a product roadmap, not shipping hardware. The R200 itself is Rebellions' next-generation chip; no pricing, order volumes or performance figures for the finished systems were published with the announcement.

That matters because sovereign AI deals are easy to announce and slow to build. The test of this one will be whether the R200 arrives on schedule and whether the disaggregated inference setup delivers the cost savings Rebellions promises.

Why Korea, and why now

Korea is one of the most aggressive sovereign AI builders. The government has made AI infrastructure a national project, and SK Telecom is its largest carrier, with data centers and its own models to run on them. Rebellions is the country's best-known AI chip startup, and NVIDIA's NVLink Fusion lets outside chipmakers connect their own processors to NVIDIA's platform — a door NVIDIA opened to partners rather than keeping its interconnect closed.

The same AI Day Seoul keynote also carried NVIDIA's wider physical AI push in Korea, with Hyundai, Doosan, LG and SK hynix. The sovereign AI deal is the concrete agreement to come out of the day; the rest is a statement of direction.

What to watch next

The next signals are simple: a firm R200 launch date, the first SK Telecom data center deployment, and any price or performance numbers for the combined systems. Until then, the deal is a plan with strong backing and a 2028 date on it.