SK Telecom is teaming up with Nvidia to target the artificial intelligence (AI) infrastructure market. With growth driven by mobile subscriptions and rate plans slowing, the company aims to turn its network and data center operations capabilities into AI computing infrastructure to create a new revenue source.
The partnership also marks an expansion of the relationship between SK Group and Nvidia beyond semiconductor supply into the broader AI infrastructure space. Until now, the two companies had focused their collaboration on memory, including SK hynix's high-bandwidth memory (HBM), but going forward they plan a structure that combines SK Telecom's data center build-out and operations capabilities with Nvidia's graphics processing units (GPU) and software platforms to jointly pursue AI factory design and operations.
◇ Redefining networks as AI computing infrastructure
On the 8th, SK Telecom said it will pursue a "full-stack AI cloud" partnership based on Nvidia's DSX platform, covering everything from chips to data center operations. The two companies plan to scale up "AI factories," data centers specialized for AI workloads, to the gigawatt (GW) level. The first AI factory is slated to go live in Korea in 2027.
An AI factory differs from a conventional data center. While general data centers provide general-purpose cloud services based on servers and storage, an AI factory is closer to dedicated production facilities that handle training and inference for large-scale AI models. It feeds in power and data to produce tokens, the basic unit of Generative AI services.
This is why SK Telecom is stressing the "lowest token cost" and "best performance per watt" in this partnership. AI cloud competitiveness is not determined by GPU volume alone. Semiconductors, memory, networks, cooling, data center operations, and software must be optimized all at once. As AI usage increases, corporations will look for infrastructure that can process more tokens for the same expense.
SK Group Chairman Chey Tae-won and Nvidia CEO Jensen Huang met in Taiwan on the 1st to discuss an AI infrastructure roadmap, aligning with this trend. The two companies plan to set up a consultative body for joint research on next-generation AI factory architecture. They aim to study computing structures that raise GPU and memory performance together from the design stage.
◇ Power and customer acquisition are key to commercialization
For SK Telecom, the partnership could be a test bed for its "beyond telecom" strategy. Carriers have nationwide network operations experience and data center operations capabilities. AI cloud is an area where carriers can convert their networks and operations capabilities into a new business. In effect, telecom networks are being redefined from simple connectivity into foundational infrastructure that runs AI services.
Commercialization, however, faces high hurdles. A GW-level AI factory must procure massive power reliably, and high-priced GPUs must run at high utilization to generate revenue. A person in the IT industry said, "How much demand SK Telecom secures from global Big Tech and large corporations for AI training and inference, and how much it lowers power and cooling expense, will be the key variables that determine the company's AI infrastructure gamble."
Nvidia CEO Jensen Huang said, "Telecommunications networks are evolving into national AI infrastructure," and noted, "SK Telecom, through the Nvidia DSX platform, will be able to build large-scale AI clouds and provide agent AI, enterprise AI, and physical AI to corporations and industries."