SK Telecom has announced a major partnership with NVIDIA to build large-scale "AI factories" in Korea, with the first one scheduled to launch in 2027.
This isn't just about building another data center. An 'AI factory' is a full-stack system designed from the ground up to produce artificial intelligence efficiently. The plan is to use NVIDIA's DSX reference architecture, a blueprint that standardizes and accelerates the construction of these massive facilities. This move will position SKT as a key provider of sovereign AI in Asia, meaning an AI infrastructure that operates within a nation's own borders, giving it more control over its data and AI development.
So, why is this happening now? The timing is driven by a convergence of critical factors. First, the technological foundation is now in place. NVIDIA's DSX platform provides partners like SKT with a playbook to build and operate these factories, optimizing for the lowest cost per token. The roadmap includes using current Blackwell chips and transitioning to the next-generation Rubin platform, ensuring the facilities remain state-of-the-art.
Second, major supply chain risks are being mitigated. At the recent Computex conference, NVIDIA's CEO signaled that the company has secured supply for robust growth, easing concerns about chip availability. Furthermore, reports indicate that key memory suppliers, including SKT's affiliate SK hynix, have been certified for HBM4, the next-generation memory essential for the Rubin platform. This synergy within the SK group—combining SK hynix's HBM with SKT's operations—creates a powerful competitive advantage.
Third, strong government backing provides a guaranteed demand anchor. The South Korean government's 'AI G3' initiative aims to pour approximately 10 trillion won into making the country a top-3 global AI power. This policy push to localize computing power and reduce dependence on overseas hyperscalers creates a ready-made domestic market for the services SKT's AI factory will offer.
However, the path forward isn't without challenges. Gigawatt-class data centers consume enormous amounts of electricity, and the grid around the Greater Seoul area is already strained. Successfully navigating these power constraints and securing suitable locations will be critical for SKT and NVIDIA to meet their ambitious 2027 timeline. Execution on these physical infrastructure fronts will be just as important as the technology itself.
- Glossary -
- Sovereign AI: An AI ecosystem, including infrastructure and models, that is developed and controlled within a country's borders to ensure data privacy, security, and national competitiveness.
- DSX: NVIDIA's reference architecture and platform for building, simulating, and operating AI factories at scale, designed to optimize performance and energy efficiency.
- HBM (High Bandwidth Memory): A type of high-performance computer memory used in high-end GPUs, crucial for training large AI models due to its ability to transfer data very quickly.