South Korean telecommunications giant SK Telecom (SKT) has unveiled ambitious plans to construct a gigawatt-scale artificial intelligence (AI) cloud within its home country, partnering with Nvidia to build dedicated infrastructure aimed at supporting the nation's rapidly growing AI demands. The initiative marks a significant shift from traditional telecom operations toward becoming a central player in the AI infrastructure space.
The core of the project involves SKT deploying Nvidia's DSX platform to build a large-scale AI factory dedicated to manufacturing the compute tokens required for advanced AI workloads. The first of these facilities is expected to become operational in 2027, targeting both training and inference tasks as well as emerging agentic AI applications. Unlike general-purpose cloud environments, SKT's AI cloud will be customised for graphics processing unit (GPU)-accelerated computing, designed to provide sovereign, physical, and enterprise AI services to industries across Korea. The company plans to eventually expand these services to the broader Asian market.
Nvidia's DSX Platform: The Foundation
Nvidia's DSX platform is a full-stack reference architecture comprising software, hardware, and operations to generate AI tokens efficiently. Key components include the DSX MaxLPS software, which drives the highest possible token performance per megawatt, and the DSX OS operating layer that manages lifecycle, health automation, and multi-tenant capabilities of an AI factory. This end-to-end solution is specifically built for running GPU-accelerated workloads at massive scale, making it ideal for telecom operators looking to pivot into AI cloud services.
Nvidia CEO Jensen Huang emphasised the strategic importance of telecom networks in the AI era: 'Telecom networks are becoming national AI infrastructure. They connect people, companies, devices and machines – and now they can become the backbone of new AI clouds. With Nvidia DSX, SK Telecom can build Korea's AI cloud at scale and bring agents, enterprise and physical AI to the companies and industries that power Korea and the world.'
The partnership extends beyond mere infrastructure deployment. SK Group and Nvidia also announced joint research into next-generation AI factory architectures, spanning what they term 'silicon-to-grid' innovation. This will focus on accelerated computing, memory technologies, and datacentre operations to drive more resilient and scalable AI services.
Strategic Context: SKT's AI Transformation
SKT chairman Chey Tae-won noted that the partnership secures full-stack AI infrastructure capabilities for the company, covering everything from chip-level hardware to broader datacentre operations. 'We will work with Nvidia to tackle GPU, memory and energy challenges, and become a leading AI cloud company shaping Asia's AI ecosystem,' he added. SKT will also become an official Nvidia Cloud Partner, joining the tech giant's global ecosystem of AI infrastructure and software providers.
This agreement represents the latest in a series of strategic moves by SKT to transform itself from a traditional connectivity provider into a full-stack global AI company. Earlier this year, a consortium led by SKT built the A.X K1 foundation model, which not only excels in Korean-based tasks but also demonstrates strong performance in mathematics and coding. The 519-billion-parameter model is currently being used in the South Korean government's sovereign AI project, highlighting the company's growing influence in national AI initiatives.
Beyond Nvidia, SKT has been actively building its own custom AI infrastructure stack to improve power efficiency. In April 2026, the company partnered with semiconductor firm Arm and AI accelerator startup Rebellions to co-develop custom datacentre hardware and software specifically optimised for inference workloads. These moves indicate a deliberate strategy to reduce dependency on generic cloud solutions and create differentiated capabilities.
Furthermore, SKT has been making substantial investments in the broader AI ecosystem. Besides establishing a dedicated AI investment unit to back emerging startups, the operator has poured significant capital into generative AI, most notably making a $100m investment into Claude developer Anthropic in 2023. This investment positions SKT at the forefront of large language model development while securing early access to cutting-edge AI technologies.
Industrial AI and Digital Twins
The telco's deepening ties with Nvidia were evident at the latter's GTC Taipei event, where SKT showcased how it is using Nvidia Omniverse libraries to apply digital twins to SK Hynix semiconductor fabs. This application optimises complex manufacturing environments and lays the groundwork for the industrial AI workloads that the new gigawatt-scale cloud aims to support. Digital twins allow real-time simulation and optimisation of production lines, reducing downtime and improving yield—critical factors for semiconductor manufacturing.
The gigawatt-scale AI cloud will also address the growing demand for sovereign AI infrastructure in Asia. Many governments are increasingly concerned about data sovereignty and national security, preferring locally operated cloud services that keep sensitive data within borders. By building this dedicated infrastructure within Korea, SKT provides a sovereign alternative to global hyperscalers, while still offering access to cutting-edge Nvidia technology.
Implications for the Asian AI Market
The partnership between SKT and Nvidia has broader implications for the Asian AI market. As telecom operators across the region watch this development, many may consider similar transformations. The ability to deliver AI compute tokens—essentially the currency of AI processing—could become a new revenue stream for telcos that have traditionally relied on connectivity and spectrum assets. Moreover, by partnering with Nvidia directly, SKT gains access to the latest GPU architectures and software optimisations before many competitors.
The timeline is ambitious: first facility online by 2027. Given the scale—gigawatt-level power consumption—this will require massive investments in energy infrastructure, cooling systems, and networking. SKT's experience in running large-scale datacentres for its telecom operations provides a foundation, but the AI factory concept is fundamentally different. AI factories are designed for continuous high-density GPU workloads, requiring advanced liquid cooling, high-bandwidth interconnects, and specialised power management. Nvidia's DSX platform addresses many of these challenges through integrated software and hardware design.
Energy efficiency will be a critical factor. With gigawatt-scale power draw, even small improvements in token performance per megawatt translate into significant operational savings. The DSX MaxLPS software is precisely aimed at maximising that metric, ensuring that every watt is used as efficiently as possible. Additionally, SK Group's involvement means access to SK Hynix's advanced memory technologies, which are crucial for memory-bound AI workloads.
The joint research on 'silicon-to-grid' innovation suggests that the partnership will explore new ways to integrate AI accelerators with energy delivery, potentially developing custom power architectures that reduce transmission losses and improve overall efficiency. Such research could have spinoff benefits for the entire AI industry.
From a competitive standpoint, SKT's move positions it against existing AI cloud providers such as AWS, Google Cloud, and Microsoft Azure, as well as specialist GPU cloud providers like CoreWeave. However, by focusing on sovereign and enterprise AI services, SKT carves a niche that global hyperscalers may struggle to serve due to data residency concerns. Korean industries—especially semiconductors, automotive, and finance—can leverage this infrastructure for their AI workloads while maintaining full control over data.
Finally, the partnership underscores Nvidia's strategy of enabling dedicated AI factories around the world, rather than relying solely on hyperscalers to deploy its hardware. By partnering directly with telecom operators, Nvidia expands its ecosystem and ensures that its technology reaches a wider range of end users. As Huang noted, telecom networks are becoming national AI infrastructure; partnerships like this one validate that vision and set a precedent for other countries to follow.
Source: ComputerWeekly.com News