Why it matters
The strategic read
Turns SK Telecom from a reported Ulsan data-center monetization watch item into a primary-source NVIDIA AI-factory platform announcement. The receipt frames telecom networks, memory, data-center operations and national AI sovereignty as one Korean AI Cloud stack, and it puts another gigawatt-scale DSX platform beside IREN, OpenAI and other AI-factory lanes competing for power, GPUs, memory, networking and data-center operating talent.
Source-supported terms
What was disclosed
- 01
NVIDIA and SK Telecom announced that SK Telecom plans to build a gigawatt-scale AI Cloud in Korea using the NVIDIA DSX platform.
- 02
The first AI factory is planned to come online in 2027.
- 03
The AI Cloud is designed for sovereign, physical, enterprise and agentic AI workloads across Korea, with a vision to expand to greater Asia regions.
- 04
SK Telecom will become an NVIDIA Cloud Partner, using NVIDIA infrastructure, software and developer ecosystem to deliver AI cloud services.
- 05
NVIDIA DSX is described as the full-stack reference architecture covering software, hardware and operations for AI factories.
- 06
NVIDIA DSX MaxLPS software is intended to improve token performance per megawatt, while NVIDIA DSX OS supports lifecycle management, health automation, resiliency and multi-tenant AI factory operations.
- 07
NVIDIA and SK Group also announced plans to pursue joint research on next-generation AI factory architectures.
- 08
The joint-research scope includes silicon-to-grid innovation across accelerated computing, memory technologies and data-center operations.
- 09
SK Telecom connected the announcement to its work applying NVIDIA Omniverse digital twins to SK hynix semiconductor fabs and to Korea sovereign AI model efforts using NVIDIA Nemotron datasets.
Disclosure boundary
What is known—and what is not
Primary NVIDIA / GlobeNewswire announcement. Public details disclose the parties, gigawatt-scale AI Cloud intent, DSX architecture path, first-AI-factory 2027 target, NVIDIA Cloud Partner framing and joint-research scope. Public materials do not disclose capex, binding purchase commitments, GPU counts, exact sites, utility service agreements, PPAs, financing, EPC awards, memory-supply commitments, tenant/customer contracts, data-center permits or final phase schedule.