Nvidia, Rebellions and SK Telecom Team Up to Build South Korea’s Sovereign AI Infrastructure

Nvidia, the world’s most valuable chipmaker, has joined forces with South Korean AI chip startup Rebellions and telecom giant SK Telecom to build the computing backbone for South Korea’s sovereign AI. The three companies announced the collaboration on Monday, 9 November 2026, at Nvidia AI Day in Seoul, in a deal backed by the South Korean government.

At the heart of the agreement is a simple but important idea: South Korea wants AI systems that run on Korean infrastructure, understand the Korean language, and serve Korean industries — without depending entirely on foreign technology. The new partnership plugs a Korean-designed AI chip into Nvidia’s data-centre platform, giving the country a home-grown piece of the AI supply chain.

The systems built under the deal are expected to be ready for deployment in 2028. No financial terms were disclosed, and the companies announced no firm purchase orders — for now, this is a development plan, not a finished product.

What is sovereign AI?

Sovereign AI is the idea that a country should be able to build and run its own artificial intelligence, on its own infrastructure, under its own rules.

In practice, that means three things. First, the computing hardware — the powerful chips that train and run AI models — should be available inside the country, not only in foreign data centres. Second, the AI models themselves should be trained on local data and speak the local language well. Third, the whole system should follow the country’s own laws on data, privacy and security, rather than depending on another country’s rules.

For South Korea, the stakes are high. The country is one of the world’s biggest makers of memory chips and smartphones, and its government has made AI a national priority under a project called “AI for All”. But almost all of the most advanced AI chips in the world today come from Nvidia, an American company. South Korea wants to keep using the best available technology while also building its own capability, so that it is never fully dependent on a single foreign supplier.

That is where Rebellions comes in. The Seoul-based startup designs AI processors — the specialised chips that run AI models after they have been trained. By connecting its chips to Nvidia’s platform, Rebellions gets access to the world’s most widely used AI software ecosystem, while South Korea gets a domestic chipmaker inside that ecosystem.

What did the three companies agree?

The three companies signed a memorandum of understanding — a formal statement of intent to work together — at Nvidia AI Day Seoul, held at the COEX convention centre in Seoul. The deal is supported by South Korea’s Ministry of Science and ICT, the government department that oversees the country’s technology policy.

Under the agreement, each partner has a clear role:

• Rebellions will adopt Nvidia’s NVLink Fusion technology to connect its next-generation Rebel R200 neural processing units (NPUs) to Nvidia’s open infrastructure platform, and will use Nvidia’s MGX rack architecture to build rack-scale systems.
• SK Telecom will deploy those systems in its AI data centres, and connect them to its own AI models and public-facing services.
• Nvidia provides the platform, the interconnect technology and the software ecosystem that ties everything together.

The companies say the R200 systems should be ready for deployment in 2028. That is still more than a year away, and the R200 chip itself is still under development. The announcement is therefore a roadmap, not a launch: the hard engineering work of joining two different kinds of chips into one system still lies ahead.

How will the chips work together?

The technical heart of the deal is a method called “disaggregated inference”. Inference is what happens when an AI model answers a question or generates text — the everyday work of a chatbot or an AI assistant. It has two stages.

In the first stage, called prefill, the system reads and analyses the user’s question. In the second stage, called decode, it generates the answer, one piece of text at a time.

Under the new plan, Nvidia’s GPUs will handle the prefill stage, while Rebellions’ NPUs will take on the decode stage. Each chip does the job it is best at, and the two work together inside the same rack of servers. The companies say this split should improve both processing speed and power efficiency — two things that matter enormously when AI systems answer millions of questions a day.

The technology that makes this possible is NVLink Fusion. NVLink is Nvidia’s high-speed connection for moving data between its own GPUs. NVLink Fusion extends that connection to chips made by other companies, so a third-party processor like Rebellions’ R200 can share data with Nvidia’s GPUs at very high speed and with very little delay. Without it, mixing chips from different makers inside one system would be slow and impractical.

The systems will be built using Nvidia’s MGX rack architecture — a standard design for the metal racks that hold servers in a data centre. Using a shared standard means the Korean chips can slot into the same racks, cooling systems and power supplies that data centres already use for Nvidia hardware.

Why this deal, and why now?

The partnership did not come out of nowhere. It builds on a relationship that has been growing all year.

Rebellions’ current chip, the ATOM-Max, is already running inside several SK Telecom services. As of September 2026, those chips were handling around 14 million requests and more than 4 billion tokens a day — a token being a small piece of text that an AI model processes. That real-world track record gave the three companies confidence to plan a bigger, deeper integration.

In April 2026, SK Telecom, Rebellions and British chip designer Arm signed a separate agreement to build a joint AI inference server. In June 2026, Nvidia and SK Telecom announced a plan for a gigawatt-scale AI cloud in South Korea using Nvidia’s DSX architecture, with an operational target of 2027. In July 2026, Nvidia and SK Group widened their relationship further, with talk of a potential investment commitment of more than $500 billion — a figure the companies described as a possibility, not a signed deal.

Monday’s announcement is the next step in that sequence: the point where the talk turns into a concrete engineering project, with a named chip, a named technology and a target date.

The timing also fits a wider pattern. Governments around the world are racing to secure their own AI capacity, worried that access to advanced chips could be restricted by export controls, trade disputes or supply shortages. South Korea, which sits between the United States and China in the global technology rivalry, has more reason than most to want its own options. The deal gives it a Korean chip inside the world’s dominant AI platform — a middle path between total dependence and going it alone.

What does SK Telecom get out of it?

For SK Telecom, South Korea’s largest mobile carrier, the deal is part of a much bigger transformation. The company is trying to reinvent itself from a telecom operator into an AI company, and it needs data-centre capacity and AI models of its own to do it.

Under the agreement, SK Telecom will deploy the new GPU-NPU systems in its AI data centres, test their performance and stability, and assess whether they are ready for live services and commercial use. The company also plans to connect the systems to its own AI model, A.X K2, and to the mass-market AI services it is building under the government’s “AI for All” project.

SK Telecom’s chief technology officer, Jeong Seok-geun, said the company would use the partnership to strengthen its full-stack sovereign AI — meaning it wants to control the whole chain, from the chips and data centres at the bottom to the models and consumer services at the top.

The wider Nvidia announcement in Korea

The Rebellions deal was not the only announcement Nvidia made in Seoul. The company also unveiled a broader collaboration with South Korea’s Ministry of Science and ICT covering AI chips, data centres, models, skills and agent security.

On the software side, Nvidia released deployment recipes — ready-made instructions for running AI models efficiently — for four Korean foundation models: SK Telecom’s A.X K2, LG AI Research’s K-EXAONE 2.0, Upstage’s Solar Open 2 and Motif Technologies’ Motif-3. That matters because a chip is only useful if developers can easily run their models on it, and these recipes lower that barrier for Korean developers.

On security, Hyundai Motor Group is exploring Nvidia’s NemoClaw blueprints and the OpenShell runtime for building AI agents — software that can carry out tasks on its own. Korean security company AIM Intelligence is integrating its Starfort agent-control platform with OpenShell, aiming to keep such agents under control in workplaces.

Taken together, the announcements show Nvidia’s strategy in South Korea: not just selling chips, but embedding itself across the country’s AI ecosystem — from the hardware in the data centre to the models, the developers and the security tools around them.

What could go wrong?

For all the ambition, there are real risks, and the companies have been careful not to overpromise.

First, the R200 chip is still being developed. Rebellions is designing it for low-latency, energy-efficient inference on very large models, including models with trillions of parameters. Those are design goals, not measured results. Until the chip exists and has been tested, nobody knows whether it will hit its targets.

Second, joining two different kinds of processors into one system is genuinely hard engineering. The prefill-decode split sounds neat on paper, but making GPUs and NPUs hand work to each other thousands of times a second, without losing speed, is the kind of problem that takes time to solve. The 2028 deployment target gives the engineers room, but it also shows how far away a working system is.

Third, the agreement is a memorandum of understanding, not a binding contract. It states intent rather than obligation. No transaction value, no order quantity and no firm purchase commitment was disclosed. If the technology does not work as hoped, or if market conditions change, the deal could shrink or stall.

And fourth, the broader numbers floating around the partnership should be read carefully. The figure of more than $500 billion, mentioned in connection with Nvidia and SK Group in July 2026, was described as a potential investment commitment, not a signed deal. The gigawatt-scale AI factory is a target for 2027, not a finished building.

Why it matters for the rest of the world

South Korea’s approach is being watched closely by other countries facing the same dilemma: how to stay plugged into the world’s best AI technology while building enough of their own to stay independent.

If the Rebellions-Nvidia model works — a domestic chip, plugged into the dominant platform, running national AI services — it could become a template for other mid-sized technology powers, from Europe to Southeast Asia. Countries that cannot build a whole AI industry from scratch might instead build one critical piece of it and integrate the rest.

For Pakistan and other developing countries, the lesson is different but just as relevant. Sovereign AI is not only about chips. It is also about models trained in local languages, data centres inside the country, and services built for local needs. The Korean example shows that even a country with world-class manufacturing cannot do it alone — it needs partnerships, standards and a plan that runs over years, not months.

The deal also underlines a shift in the AI industry itself. For years, Nvidia’s platform was a closed world: only Nvidia chips could use its fastest connections. NVLink Fusion opens that world to other chipmakers. If more companies follow Rebellions into the Nvidia ecosystem, the AI hardware market could become more varied and more competitive — good news for buyers.

What happens next?

The next milestone is engineering: Rebellions must finish the R200, connect it to Nvidia’s platform through NVLink Fusion, and prove that the combined system works at scale. SK Telecom will then test it in its data centres, measuring speed, stability and power use against the systems it runs today.

If those tests go well, the first deployments should arrive in 2028, with Korean AI services — from SK Telecom’s own products to government-backed “AI for All” services — running partly on Korean silicon for the first time.

Until then, the announcement is best read as a statement of intent from three companies and a government that all want the same thing: AI that South Korea can call its own, built on technology the world already uses.

Frequently asked questions

What is sovereign AI?

Sovereign AI means a country can build and run its own artificial intelligence on its own infrastructure, under its own laws. It covers the hardware, the AI models, the data and the services — so the country is not fully dependent on foreign technology.

What is NVLink Fusion?

NVLink Fusion is Nvidia technology that lets AI chips made by other companies connect to Nvidia’s platform using its high-speed NVLink connection. It allows specialised processors, like Rebellions’ R200, to work alongside Nvidia GPUs inside the same system.

What is the Rebel R200?

The Rebel R200 is Rebellions’ next-generation neural processing unit (NPU), a chip designed for running AI models efficiently. It is still under development, and systems using it are expected to be ready for deployment in 2028.

What is disaggregated inference?

Disaggregated inference splits the work of answering an AI question between different chips. In this deal, Nvidia GPUs handle the first stage (reading the question) and Rebellions NPUs handle the second stage (generating the answer), aiming for faster responses and lower power use.

When will the new systems be ready?

The companies expect the R200 systems to be ready for deployment in 2028. The announcement is a development plan, not a finished product.

Conclusion

Nvidia, Rebellions and SK Telecom have set out a plan to give South Korea a piece of its own AI future. A Korean-designed chip, connected to the world’s most widely used AI platform through NVLink Fusion, will run in SK Telecom’s data centres and power Korean AI models and services.

The deal is backed by the South Korean government, builds on a year of growing cooperation, and fits a global trend of countries seeking more control over the technology that increasingly runs their economies. But it is still a plan, not a product: the chip is unfinished, the engineering is hard, and the 2028 target is more than a year away.

If it works, South Korea will have shown how a country can stay inside the world’s dominant AI ecosystem while building a genuine piece of its own. Other countries will be watching — and taking notes.

Sources

• Nvidia, Rebellions and SK Telecom Announce Collaboration to Advance Sovereign AI Infrastructure With NVIDIA NVLink Fusion (GlobeNewswire, 9 November 2026): https://www.tradingview.com/news/globenewswire:a02cf44a6094b:0-nvidia-rebellions-and-sk-telecom-announce-collaboration-to-advance-sovereign-ai-infrastructure-with-nvidia-nvlink-fusion/
• SK Telecom, Nvidia, Rebellions join forces to build Korea-specific AI infrastructure (The Korea Herald, 9 November 2026): https://biz.heraldcorp.com/article/10930847
• NVIDIA Deepens Korean Sovereign AI Work on Chips and Rack Systems (Unite.AI, 9 November 2026): https://www.unite.ai/nvidia-deepens-korean-sovereign-ai-work-on-chips-and-rack-systems/
• Nvidia, Rebellions and SK Telecom Team Up on Korea’s Sovereign AI With NVLink Fusion (PakTechStan, 9 November 2026): https://paktechstan.com/article/nvidia-rebellions-and-sk-telecom-team-up-on-korea-s-sovereign-ai-with-nvlink-fusion

This article is for general information and education only. It is not legal or financial advice.