NVIDIA and SK Group signed letters of intent for what the companies described as a $500-billion-plus AI infrastructure and memory initiative. The plan calls for SK Telecom to develop up to 2 gigawatts of NVIDIA Vera Rubin-based AI cloud capacity in South Korea, while NVIDIA and SK hynix pursue a long-term supply and codevelopment partnership for next-generation AI memory. The announcement does not break down the headline figure among capital spending, equipment purchases, memory supply, research or other commitments.
SK Telecom plans to build the cloud on NVIDIA’s DSX full-stack AI factory architecture, deploying Vera Rubin accelerated computing with SK hynix HBM4. DSX provides a common architecture across compute, networking, storage, infrastructure software, power, cooling and facility operations, with the objective of optimizing token output within a fixed power envelope. The first AI factory is planned to enter service in 2027, and the planned capacity will support sovereign, enterprise, agentic and physical AI services in South Korea and across the wider Asia-Pacific region.
Under the planned memory partnership, NVIDIA would secure a longer-term supply path for future AI memory, while NVIDIA and SK hynix would codevelop and optimize HBM for accelerator platforms and workloads ranging from large-language-model training to AI agents and robotics. The release did not disclose memory volumes, pricing, individual factory sites, GPU counts, capacity phasing, power interconnection plans or network topology. Those details, along with definitive agreements, will determine how the letters of intent translate into deployed infrastructure.
• Agreement status: Letters of intent signed by SK Group and NVIDIA
• Stated scope: More than $500 billion across AI factories and next-generation memory; allocation not disclosed
• Planned capacity: Up to 2 gigawatts of SK Telecom AI cloud infrastructure in South Korea
• Compute platform: NVIDIA Vera Rubin accelerated computing
• Memory: SK hynix HBM4, followed by joint work on future HBM generations
• Factory architecture: NVIDIA DSX spanning compute, networking, storage, software, power, cooling and operations
• Initial timing: First AI factory planned to enter service in 2027
• Target workloads: Sovereign AI, enterprise AI, LLM training, agentic AI and physical AI
“Together with SK Telecom and SK hynix, we are building a new generation of AI factories that will power Korea’s next wave of growth,” said Jensen Huang, founder and CEO of NVIDIA.
🌐 Analysis
The July agreement expands the scale and specificity of the AI cloud plan that SK Telecom and NVIDIA disclosed on June 7. The earlier announcement described a gigawatt-scale DSX deployment; the new letters of intent set an upper target of 2 gigawatts and identify Vera Rubin and SK hynix HBM4 as core platform components. This links the facility roadmap directly to NVIDIA’s next-generation compute architecture and SK hynix’s memory production.
The initiative also connects three layers that operators normally procure separately: carrier-operated cloud capacity, accelerator systems and HBM supply. The companies say the arrangement is intended to support joint optimization across memory, compute, networking and facility power, but the announcement remains an early framework. Site selection, utility capacity, network design, equipment volumes, supply commitments and phased construction milestones were not disclosed.
