SK Telecom and NVIDIA plan gigawatt-scale AI cloud for Korea’s “AI factories,” with first site targeted for 2027

NVIDIA and SK Telecom said on June 7, 2026 that SK Telecom plans to build a gigawatt-scale AI Cloud in South Korea, using NVIDIA’s DSX AI factory platform as the underlying reference architecture. The companies said the first “AI factory” is targeted to come online in 2027, positioning the telecom operator as a provider of purpose-built, GPU-based AI infrastructure rather than general-purpose cloud capacity. The announcement was made through an NVIDIA-sourced press release carried by Yahoo Finance.

In its description of the plan, SK Telecom’s AI Cloud is intended to manufacture “tokens,” which are the basic units of text (and other data) that large language and agent systems process during training and use. The cloud is expected to support training, inference, and “agentic” workloads, and to deliver “sovereign, physical and enterprise” AI services for companies and industries across Korea. NVIDIA framed the broader thesis as telecom networks becoming national AI infrastructure, arguing that network reach can serve as the backbone for new AI cloud offerings. Jensen Huang, NVIDIA’s chief executive, said telecom networks are increasingly becoming national AI infrastructure and that DSX enables SK Telecom to build the AI cloud at scale.

Central to the project is NVIDIA DSX, which NVIDIA describes as a full-stack framework for building AI factories, integrating compute, systems, software, and operations. NVIDIA said DSX is engineered to reduce the cost per token and accelerate “time to first production” for AI factory deployments. In the DSX branding used in the announcement, DSX MaxLPS is positioned as power and efficiency software designed to maximize token performance per megawatt, while DSX OS is positioned as an operations layer for lifecycle management, runtime consistency, health automation, resiliency, and multi-tenant operations.

The new AI cloud plan also builds on SK Telecom’s earlier work around “physical AI,” which links AI models to real-world processes using techniques such as digital twins. NVIDIA said SK Telecom recently discussed applying digital twins to semiconductor fab environments using NVIDIA Omniverse libraries, aimed at optimizing technology for complex, large-scale manufacturing settings. SK Telecom’s newsroom describes a move beyond visualization toward a physical AI platform, and it said it validated NVIDIA Omniverse libraries in real manufacturing conditions and plans commercialization in phases aligned with industrial roadmaps.

The announcement also tied the AI infrastructure effort to SK Telecom’s progress on Korea’s sovereign AI model program. NVIDIA said that in April, SK Telecom adopted open-source NVIDIA Nemotron datasets to train the A.X K1 model as part of the Korea government’s Sovereign AI Foundation Model Project. Nemotron datasets are part of NVIDIA’s model and data ecosystem used for building and deploying agentic AI systems, and the companies’ messaging suggests SK Telecom is trying to connect sovereign model development with sovereign infrastructure deployment.

NVIDIA and SK Telecom further said SKT will become an NVIDIA Cloud Partner, joining a program intended to give participants access to NVIDIA’s latest AI infrastructure, software, and developer ecosystem for delivering AI cloud services. NVIDIA also said it and SK Group are planning joint research aimed at next-generation AI factory architectures, focusing on “silicon-to-grid” innovation across accelerated computing, memory technologies, and data center operations, with additional work on full-stack AI factory optimization to improve efficiency, scalability, and resilience.

The companies did not disclose in the June 7 announcement any specific capital spending figures, the targeted number of GPUs, the exact location(s) of AI factory deployment sites in Korea, pricing for cloud customers, or service-level commitments. They also did not specify how “sovereign” will be implemented in practice in terms of data handling, isolation, or governance beyond general positioning, leaving those details to later updates or separate technical agreements.

Investors and industry observers will likely focus on whether the 2027 “first AI factory” milestone is met, and on what mix of training, inference, and agentic services the AI Cloud can deliver once operational. The market will also watch how quickly SK Telecom can convert infrastructure into industrial adoption, particularly in robotics, manufacturing, and other “physical AI” use cases that the companies highlighted, and whether the joint research agenda around silicon-to-grid architectures produces measurable improvements in efficiency and time to deployment.