NVIDIA, the South Korean chip startup Rebellions and the telecoms group SK Telecom have announced a joint effort to build the computing backbone for South Korea's sovereign artificial intelligence ambitions, pairing Korean-designed processors with NVIDIA's open infrastructure platform.
The collaboration, announced in Seoul on Monday, 9 November 2026, centres on Rebellions' next-generation R200 neural processing units. Rebellions plans to connect the chips to NVIDIA's platform using NVLink Fusion, the high-bandwidth interconnect that lets chipmakers combine their own specialised silicon with NVIDIA's scale-up networking, and to build the systems on NVIDIA's MGX rack architecture. SK Telecom plans to deploy the resulting systems in its own AI data centres.
The project is backed by South Korea's Ministry of Science and ICT, which is supporting the work as the starting point for broader cooperation between NVIDIA and Korean chipmakers. The ministry framed the deal as a way for Korean-designed processors to enter the global computing infrastructure market alongside the world's leading AI hardware company, and said the government would actively support the collaboration, including in physical AI and in securing supplies of graphics processors.
At the heart of the announcement is the idea of sovereign AI: the notion that a country should be able to build, train and run its own AI systems on its own infrastructure, using models tuned to its own language, culture and industries, rather than relying entirely on foreign platforms. NVIDIA's chief executive, Jensen Huang, said Korea had "the talent, technology and industrial ambition to become a world leader in AI", and that sovereign AI gave nations "the power to turn their language, culture and expertise into intelligence".
Rebellions, founded in Seoul and backed by investors including Samsung, SK Hynix, Arm and Aramco, describes the R200 as designed for low-latency, energy-efficient inference on frontier models, including models with multiple trillions of parameters. Its existing flagship chip, the Rebel100, is already deployed in enterprise and government settings. The R200 systems are expected to be ready for deployment in 2028, giving the partners roughly two years of integration and system development.
The technical plan splits AI work between the two kinds of processor. NVIDIA's graphics processors would handle the "prefill" stage, in which a model first reads and processes a prompt, while Rebellions' NPUs would handle "decode", the stage in which the model generates its answer token by token. The companies say this disaggregated approach should improve both inference throughput and power efficiency, two of the most pressing constraints as AI models grow larger and data centres consume more electricity.
The MGX rack architecture is meant to make that pairing practical. It gives GPU and NPU systems a shared rack footprint, along with common power delivery and cooling, and gives Rebellions access to an established supply chain of original design manufacturers and component suppliers. SK Telecom, which has operated AI data centres and built its own foundation models, plans to connect the infrastructure with its proprietary models and public-facing services, including those under the government's "AI for All" project.
The collaboration reaches beyond hardware into models and skills. Under the AI for All programme, NVIDIA will work with three selected consortia and four Korean foundation-model builders, using its open-source Dynamo inference framework to optimise and scale their models. Deployment recipes have been prepared for SK Telecom's A.X K2, LG AI Research's K-EXAONE 2.0, Upstage's Solar Open 2 and Motif Technologies' Motif-3, with NVIDIA reporting up to twice the out-of-the-box performance on B200 graphics processors. Training and certification programmes, hackathons and technical seminars are also planned to build the workforce needed to run GPU-based infrastructure.
Agent safety is part of the package as well. Hyundai Motor Group is exploring NVIDIA's NemoClaw blueprints and OpenShell runtime, which provide sandboxed execution and policy-based controls for autonomous AI agents, while the Korean security company AIM Intelligence is integrating its Starfort agent-control platform with OpenShell, combining runtime access controls with detection of malicious instructions and data leakage. AIM Intelligence is also considering a proof of concept for the public energy sector.
The announcement came during NVIDIA's AI Day in Seoul, where the company also set out a wider push into physical AI — the technology that lets robots understand and act in the real world — with partners including Hyundai Motor Group, Doosan Robotics, LG Electronics and SK Hynix. NVIDIA said AI agents could now build robot simulation environments in about a week, work that used to take several engineers several months, accelerating Hyundai's digital-twin factory programme and Doosan's precision polishing work.
For South Korea, the deal is a step towards turning its semiconductor strength into AI independence. The country is home to two of the world's largest memory chipmakers, yet it has so far depended heavily on foreign computing platforms to run its most advanced AI systems. If the R200 programme delivers on schedule, Korean chips, Korean models and Korean data centres could for the first time form a complete, domestically controlled AI stack — and Rebellions' chief executive, Sunghyun Park, said the company intended to lead the charge in establishing Korean technology "as a core pillar powering AI worldwide".
What happens next will depend on execution: the mixed GPU-NPU racks still need processor integration and system development before SK Telecom can deploy them, and the 2028 target leaves little room for delay. But with government backing, a committed customer in SK Telecom and NVIDIA's open platform behind it, the project now has the pieces in place to test whether a mid-sized chipmaker can stand beside the industry's dominant supplier — and whether South Korea can turn its chip-making muscle into genuine AI sovereignty.
Figures in this report were checked at 11:55 PKT on 09 November 2026 against the joint announcement carried by GlobeNewswire, and reports from Unite.AI and Chosun Biz.
