SEOUL — NVIDIA, Rebellions and SK Telecom announced on Monday, 9 November 2026, a collaboration to advance South Korea's sovereign AI infrastructure, pairing Rebellions' Korean-designed AI processors with NVIDIA's open infrastructure platform and SK Telecom's AI data centres, models and services. Under the agreement, Rebellions will connect its next-generation Rebel R200 neural processing units (NPUs) to NVIDIA's platform through NVIDIA NVLink Fusion and build rack-scale systems on the NVIDIA MGX rack architecture, with the first R200 systems expected to be ready for deployment in 2028. SK Telecom plans to deploy those systems in its AI data centre, linking them to its proprietary foundation models — including A.X K2 — and to public-facing services under the South Korean government's "AI for All" project. The collaboration is supported by Korea's Ministry of Science and ICT (MSIT).
Illustration: An AI-generated illustration of a modern AI data center hall in Seoul, with server racks and Korean-designed AI accelerator chips. This is an artist's impression, not a photograph of the event.
What Happened
The three companies announced the collaboration in Seoul on Monday, 9 November 2026, in a joint release distributed by GlobeNewswire. The deal brings together three distinct pieces of an AI system: NVIDIA's open infrastructure platform, Rebellions' Korean-designed AI processors, and SK Telecom's data centre, model and service capabilities. According to the announcement, the aim is to build computing infrastructure that lets South Korea run advanced AI services on hardware and models shaped around the country's own language, industries and priorities.
At the centre of the agreement is Rebellions' next-generation Rebel R200 neural processing unit. Rebellions will use NVIDIA NVLink Fusion to connect the R200 chips to NVIDIA's platform, and will harness the NVIDIA MGX rack architecture for rack-scale system development. The R200 is designed for low-latency, energy-efficient inference on frontier AI models, including multitrillion-parameter mixture-of-experts models. Through NVLink Fusion, Rebellions will connect its R200 NPUs within a rack using NVIDIA NVLink scale-up networking.
SK Telecom, meanwhile, plans to deploy the resulting systems in its AI data centre, drawing on its experience building and operating AI data centres, its proprietary foundation models and its public-facing AI services. The carrier has a long-standing collaboration with NVIDIA in AI development and sovereign infrastructure, and the new systems will use the standardised MGX rack architecture, which gives GPU and NPU systems a shared foundation for rack footprint, power delivery and cooling.
The companies also plan to develop an approach called disaggregated inference, in which different stages of generating an AI response are assigned to the processor best suited to each task: NVIDIA GPUs for the "prefill" stage, which processes a user's prompt, and Rebellions NPUs for the "decode" stage, in which the model generates its answer. According to the announcement, splitting the work this way is intended to improve inference throughput and power efficiency.
NVIDIA founder and CEO Jensen Huang said in the release: "Korea has the talent, technology and industrial ambition to become a world leader in AI. Sovereign AI gives nations the power to turn their language, culture and expertise into intelligence. Together with MSIT, Rebellions and SK Telecom, we're building the computing foundation for Korea to produce its own intelligence and power its next industrial revolution."
Sunghyun Park, cofounder and CEO of Rebellions, said: "Today's announcement marks a turning point, opening a new chapter for sovereign AI by connecting Korea's AI silicon and models with global systems technology. By integrating Rebellions' purpose-built silicon and system design with NVIDIA's NVLink Fusion, high-speed interconnects and rack-scale design expertise, and SKT's model and data center operations capabilities, we will deliver low-latency and energy-efficient inference for frontier models at scale."
Jung Jai-hun, president and CEO of SK Telecom, said the collaboration "validates the potential of next-generation sovereign AI infrastructure by combining global AI infrastructure technology with Korean AI semiconductors," adding that linking SKT's AI data centres, proprietary models including A.X K2, and public-facing AI services under the "AI for All" project would "strengthen our full-stack sovereign AI competitiveness and contribute to the growth of Korea's AI ecosystem."
Bae Kyung-hoon, Korea's Deputy Prime Minister and Minister of Science and ICT, said: "This collaboration with NVIDIA marks a significant step forward, as it enables Korean-developed NPUs to enter the computing infrastructure market alongside NVIDIA, a global leader in AI computing, through NVLink Fusion. The government will actively support this collaboration to deliver tangible results."
Background
The announcement lands at a moment when South Korea is investing heavily in becoming a major AI power, and when governments around the world are asking how much of their AI infrastructure they should control themselves. "Sovereign AI" describes the idea that a country should be able to build, run and govern AI systems using its own computing capacity, data, models and expertise, rather than depending entirely on foreign providers. For South Korea, that means AI that understands Korean language and culture, serves Korean industries, and runs on infrastructure the country can operate and secure itself.
Rebellions is a South Korean startup that designs inference accelerators — chips built specifically to run trained AI models, as opposed to training them. Its flagship product, the Rebel100, is built for inference, with a focus on maximising performance relative to cost and power use, using a chiplet architecture and a full software stack built entirely on open-source technologies and open standards. Rebellions has commercial deployments in enterprise and government environments. Based in South Korea and operating in the United States, the company is backed by Aramco, Arm, Kindred Ventures, KT, Mirae Asset Group, Samsung, SK Hynix and SK Telecom.
Illustration: An AI-generated illustration of a next-generation AI processor chip with high-speed interconnect links. This is a conceptual image, not a photograph of the Rebel R200 chip.
NVLink Fusion is the technology that makes the partnership possible. It extends NVIDIA's platform to custom XPUs — processors designed by other companies — so that chipmakers can combine their specialised silicon with NVIDIA's high-bandwidth, low-latency scale-up networking and established infrastructure ecosystem. According to NVIDIA, that combination is meant to improve performance, reduce deployment risk and accelerate time to market. In practice, it allows a Korean-designed chip to sit inside the same rack, on the same high-speed fabric, as NVIDIA's own processors.
The MGX rack architecture plays a supporting role. MGX is NVIDIA's modular reference design for building AI servers and racks, and its ecosystem gives Rebellions access to an established supply chain of original design manufacturers and component suppliers. For a startup bringing a new chip to market, that access matters: designing a competitive processor is only part of the challenge, and the rest lies in turning it into a complete, reliable, manufacturable system.
SK Telecom brings the other half of the equation. The company has led South Korean telecommunications since 1984, and today it works across the full-stack AI ecosystem, from infrastructure to models to services. Its proprietary foundation model family includes A.X K2, and its public-facing services include work under the government's "AI for All" project, which aims to spread AI services across the country's public and private sectors.
The announcement also arrived against a reassuring backdrop for South Korea's AI hardware plans. Reuters reported on Monday that South Korea expects contracted deliveries of NVIDIA graphics processing units to proceed without disruption. Vice Minister of Science and ICT Ryu Jemyung told a parliamentary budget committee that 10,000 GPUs funded under a 2025 supplementary budget had already arrived and were in use, and that GPUs planned for 2026 were under contract and expected to enter service this year.
Why It Matters
The deal matters first because it shows how NVIDIA's strategy is evolving. For years, the company's dominance in AI rested on selling its own chips. NVLink Fusion opens a different path: letting other companies' processors plug into NVIDIA's platform and ecosystem. For Korea, that opening is an opportunity to build a domestic AI chip industry without having to replicate NVIDIA's entire infrastructure stack from scratch. Rebellions gets a route into data centres through a standard rack design and a proven supply chain; NVIDIA gets a deeper foothold in a country that is investing heavily in AI; and South Korea gets more of its AI stack built at home.
It matters, too, because inference — running AI models for real users — is becoming the largest and fastest-growing part of the AI computing market. Training a model is a one-off effort; serving it to millions of people is continuous, and the cost of that serving, in energy as well as money, shapes whether AI services are affordable. The planned split between NVIDIA GPUs for prefill and Rebellions NPUs for decode matches each stage of the work to the chip that does it most efficiently, and if it works as intended it could lower the cost and energy use of running large models in SK Telecom's data centres.
There is also a strategic dimension. South Korea is home to some of the world's most important semiconductor companies, including Samsung and SK Hynix, yet much of the AI hardware running in the country today is designed abroad. A Korean-designed NPU, built into standard racks and deployed by the country's largest carrier, would give Seoul a tangible stake in the hardware layer of AI, not just in memory chips and manufacturing. The government's support through MSIT underlines that this is seen in Seoul as industrial policy as much as a commercial deal.
Illustration: An AI-generated illustration of the Seoul skyline at dusk with digital network overlays symbolising sovereign AI infrastructure. This is an artist's impression, not a photograph of the event.
For the wider industry, the collaboration is a test case. If Rebellions' R200 can be integrated successfully into NVIDIA's ecosystem and deployed at a major carrier, other specialised chipmakers may follow the same route. If integration proves difficult, or if the 2028 timeline slips, it will be a reminder that announcements are easier than systems. Either way, the outcome will be watched closely by chip startups, cloud operators and governments weighing their own sovereign AI plans.
What Happens Next
The most important date in the announcement is 2028. Rebellions' R200 systems are expected to be ready for deployment that year, which gives the partners roughly two years of development, integration and testing before SK Telecom can put the systems into service. Between now and then, the work will focus on connecting the R200 to NVIDIA's platform through NVLink Fusion, building rack-scale systems on the MGX architecture, and developing the disaggregated inference approach that splits work between GPUs and NPUs.
MSIT's support is likely to remain visible. The ministry said it is supporting collaboration between NVIDIA and Korean NPU companies, beginning with Rebellions and SK Telecom, and the government has framed the project as a step towards strengthening Korea's sovereign AI infrastructure.
Several questions remain open. The announcement did not disclose financial terms, the number of systems SK Telecom will deploy, or performance benchmarks for the R200 — none of which would be expected at this early stage, since the chip is still under development. What the companies have promised, for now, is a working combination of Korean silicon, NVIDIA infrastructure and a carrier-scale deployment. The next concrete milestone will be evidence that the integration works: R200 chips running inside MGX racks, serving real models at the efficiency the partners have described.
Sponsored: explore today's featured sponsored offers
FAQ
What is sovereign AI, and why does South Korea want it?
Sovereign AI means a country can build, run and govern AI systems using its own computing capacity, data, models and expertise. South Korea wants AI services tailored to the Korean language, its industries and its priorities, running on infrastructure the country can operate and control, rather than depending entirely on foreign providers.
What is the Rebel R200, and when will it be ready?
The Rebel R200 is Rebellions' next-generation neural processing unit, designed for low-latency, energy-efficient inference on frontier models, including multitrillion-parameter mixture-of-experts models. According to the announcement, R200 systems are expected to be ready for deployment in 2028.
What is NVLink Fusion, and what does it do in this collaboration?
NVLink Fusion extends NVIDIA's platform to custom processors designed by other companies. It lets chipmakers combine their specialised silicon with NVIDIA's high-bandwidth, low-latency scale-up networking, so the Rebel R200 can work inside the same rack as NVIDIA processors instead of running on a separate, incompatible system.
What role will SK Telecom play?
SK Telecom plans to deploy the combined systems in its AI data centre, connecting them to its proprietary foundation models, including A.X K2, and to public-facing services under the government's "AI for All" project, as part of its full-stack sovereign AI strategy.
How is this story different from other recent NVIDIA news?
This article covers only the collaboration announced in Seoul on 9 November 2026 between NVIDIA, Rebellions and SK Telecom. It is a separate development from other recent NVIDIA-related stories, including earlier reports about the company's board and about the economics of AI infrastructure, which concern different matters.
Sources
- NVIDIA, Rebellions and SK Telecom, "NVIDIA, Rebellions and SK Telecom Announce Collaboration to Advance Sovereign AI Infrastructure With NVIDIA NVLink Fusion," GlobeNewswire, 9 November 2026 — TradingView mirror of the release
- Reuters, "South Korea says Nvidia issues won't affect contracted GPU supplies," 9 November 2026 — reuters.com
- NVIDIA Newsroom — announcement coverage, 9 November 2026
- Rebellions — company background, rebellions.ai
- SK Telecom Newsroom — company background, news.sktelecom.com
- Unite.AI, "NVIDIA Deepens Korean Sovereign AI Work on Chips and Rack Systems," 9 November 2026 — unite.ai
Comments
Post a Comment