Share
Nvidia, SK Telecom and Korean chip startup Rebellions will jointly develop sovereign AI infrastructure. The technology tying the project together is Nvidia NVLink Fusion.
Rebellions makes its own AI chips, and NVLink Fusion is built to bring third-party accelerators like those into Nvidia’s ecosystem, so they work alongside it instead of outside it.
For South Korea, the goal is to depend less on external technology. The plan leans on the two industries where the country already has real muscle: semiconductors and telecommunications.
What each company brings
NVLink Fusion is Nvidia’s way of integrating accelerators from other companies into one connected system. The stated aims are better energy efficiency and stronger performance on AI workloads.
SK Telecom plans to launch a gigawatt-scale AI factory in Korea built on Nvidia’s DSX AI Factory architecture, with an operational target of 2027. The buildout is meant to start on Nvidia’s Blackwell platform and would then move to the upcoming Vera Rubin platforms.
Rebellions’ ATOM-Max NPUs are already running inside several SK Telecom services. NPUs, or neural processing units, are chips designed specifically for AI tasks.
As of September 2026, those chips were handling approximately 14 million requests and more than 4 billion tokens every day. The workloads include call summarization, text-to-speech, and AI-driven contact centers.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
We respect your privacy. Unsubscribe anytime.
How the partnership got here
In April 2026, SK Telecom, Rebellions and Arm signed an MOU. It covers a joint AI inference server that pairs Arm’s CPU infrastructure with Rebellions’ proprietary technology.
In July 2026, Nvidia and SK Group widened their relationship. That expanded scope could potentially lead to an investment commitment of more than $500 billion. It spans AI factories and next-generation memory work through SK hynix.
In August 2026, Nvidia and Rebellions held early-stage discussions pointing toward possible technical and investment collaboration.
The NVLink Fusion arrangement links Nvidia’s interconnect, SK Telecom’s network and data center plans, and Rebellions’ inference chips into one stated initiative.
What this means
For Nvidia, NVLink Fusion lets it take part in sovereign AI projects even when the accelerator isn’t its own. Its architecture, platforms and interconnect stay at the center of the system.
For Rebellions, a daily load of more than 4 billion tokens inside a major carrier is a strong credential. Plugging into Nvidia’s ecosystem through NVLink Fusion could also give Rebellions access to deployments it would struggle to win alone.
For SK Group, the pieces fit together neatly. SK Telecom runs the services and plans the AI factory, while SK hynix works on memory. Rebellions supplies domestic inference chips, with Nvidia and Arm filling in the rest.
The more than $500 billion figure is described as a potential commitment, not a signed check. The April agreement is an MOU, which states intent rather than binding terms. The Nvidia and Rebellions talks were explicitly early-stage, and the gigawatt-scale factory is a 2027 target, not a finished building.
The plan’s move from Blackwell to Vera Rubin ties SK Telecom’s schedule to Nvidia’s product cadence.
