SK Telecom and NVIDIA to build gigawatt-scale AI Cloud in Korea
SK Telecom has announced plans to build a gigawatt-scale AI Cloud in South Korea, using NVIDIA's DSX full-stack reference architecture as the engineering blueprint. The first AI factory — purpose-built infrastructure optimised for GPU-based AI workloads — is scheduled to come online in 2027, with the ambition to eventually serve markets across wider Asia.
The initiative marks a significant expansion of the existing relationship between NVIDIA and SK Group. Under the agreement, SK Telecom will become an NVIDIA Cloud Partner, gaining access to the latest accelerated computing hardware, software and developer ecosystem. Separately, NVIDIA and SK Group have announced plans for joint research into next-generation AI factory architectures, with a stated focus on silicon-to-grid optimisation spanning accelerated computing, memory technologies and data centre operations.
What the DSX platform brings
NVIDIA DSX is engineered as an AI-factory reference architecture, combining hardware, software and operational tooling into a single integrated stack. Two software components are highlighted: DSX MaxLPS, which the company says maximises token performance per megawatt, and DSX OS, a lifecycle management and multi-tenancy layer designed to help operators improve margins. The platform is positioned by NVIDIA as a way to reduce time-to-production while lowering the cost per token — the core unit of output for large language models and agentic AI systems.
SK Telecom's cloud is intended to support training, inference and agentic workloads, as well as sovereign AI and physical AI services. The sovereign AI element is particularly relevant: in April 2026 SK Telecom disclosed it had used NVIDIA's open-source Nemotron datasets to train the A.X K1 model as part of a Korean government Sovereign AI Foundation Model Project. At GTC Taipei in late May, SK Telecom also demonstrated work applying digital twins to SK Hynix semiconductor fabs using NVIDIA Omniverse libraries.
Jensen Huang, founder and chief executive of NVIDIA, said telecom networks are "becoming national AI infrastructure" and that the partnership would allow SK Telecom to "bring agents, enterprise and physical AI to the companies and industries that power Korea and the world." Chey Tae-won, chairman of SK Group, said the collaboration had secured "full-stack AI infrastructure capabilities, from chips to data centre operations."
Market and competitive context
The announcement sits within a broader global race to build sovereign, national-scale AI infrastructure. Governments and major telcos across the EU, the Gulf and Southeast Asia have all announced comparable programmes over the past 18 months, often anchored to NVIDIA hardware given its near-monopoly on high-performance AI accelerators. The NVIDIA DSX programme — which competes conceptually with hyperscaler managed AI infrastructure from AWS, Microsoft Azure and Google Cloud — is designed to let carriers and regional operators build differentiated AI cloud businesses rather than reselling hyperscaler capacity.
South Korea's position is notable: SK Hynix is among the world's leading producers of high-bandwidth memory (HBM), a critical component in NVIDIA's H100 and Blackwell GPU families. A deeper SK Group–NVIDIA research partnership on memory and compute co-design therefore carries strategic weight beyond a typical infrastructure deal. The stated joint-research focus on "silicon-to-grid innovation" suggests both parties are treating this as a long-term technology relationship, not a procurement arrangement.
On the regulatory side, South Korea's government has been actively promoting domestic AI capability through the Sovereign AI Foundation Model Project, and a gigawatt-scale data centre build will face scrutiny around grid capacity and carbon commitments as Korea works towards its 2050 net-zero target. Neither company disclosed power-purchase arrangements, cooling strategy, or the specific data-centre sites involved.
The deal is structured as a platform commitment with a 2027 delivery milestone for the first factory; investors and enterprise buyers will be watching for capacity announcements, named anchor customers, and published performance benchmarks as the build-out progresses.