[E-Daily Reporter Yoon Jung-hoon] Omdia, a market research firm, recently analyzed SKT's AI data center strategy in an in-depth report titled 'Sovereign AI Infrastructure: SK Telecom Thinks cloud tech stack to address GPU monetization' and evaluated on the 24th that it has the most advanced monetization model among global telecommunications companies.
SKT became an independent AI company in September last year. It also aims to invest 5 trillion won (about 3.6 billion U.S. dollars) in AICIC over the next five years and achieve annual sales of more than 5 trillion won by 2030. Among them, the AI data center division alone aims to achieve 1 trillion won (about 700 million dollars) in sales by 2030.
The results are already being shown in figures. The report said that the investment effect of its own virtualization platform "Petasus AI Cloud" and the acquisition of Pangyo Data Center combined to bring AIDC sales to $354 million in fiscal 2025, a 35 percent increase year-on-year.
Omdia pointed out its own "Petasus AI Cloud" as the technical foundation of SKT's monetization strategy. Previously, in order to split a large physical cluster into multiple customers, the InfiniBand-based network had to be reorganized directly, and this work alone took several days.
According to the report, Petacus first created GPUs in units of virtual machines (VMs) instead of physical servers and then grouped them into clusters, shortening the cluster setup based on 1,000 GPUs to less than 40 minutes.
SKT's release of the Korean sovereign AI foundation model 'A.X K1' as an open source was also analyzed as a key monetization strategy.
By opening it up so that companies can customize it to their environment, they designed a structure in which demand for learning and inference flows into SKT's GPU infrastructure as the number of cases of A.X K1-based B2B increases. Omdia added that many affiliates, including SK Hynix, SK Innovation, SK AX, and SK Broadband, have already expressed their intention to participate in the demonstration and are supporting initial demand.
The profit model also evolved...From 'GPU per hour' to 'rate per token, based on SLA'
Omdia predicted that SKT's monetization path will evolve from the current hourly GPU charging to inference-oriented charging per token.
SKT officially announced the launch of the "Sovereign AI Service Package" last month at MWC Barcelona 2026, which combines AI infrastructure, its own model, and industry-specific services, and plans to switch to an SLA and performance-based billing system that guarantees predefined quality in exchange for customers using the entire stack in the long run, the report said.
According to global carrier trends, SKT is clearly different in that major overseas carriers focus on securing GPUs and expanding data center capacity, while SKT uses a vertical integration strategy that encompasses infrastructure, models, and services.
German Deutsche Telekom collaborated with Nvidia to build an AI factory with 10,000 GPUs, and Italy's Iliad invested 3 billion euros to expand the multi-gigawatt data center in Europe, and European telecommunications companies are focusing on securing physical infrastructure under the pretext of 'data sovereignty'.
Japan's Softbank is also entering capacity competition by opening two AI data centers with up to 550MW in 2026.
Omdia evaluated that SKT has a three-step structure that increases supply efficiency with GPU virtualization technology, internalizes demand with its own foundation model, and raises margins with bundled services and performance-based billing.
Omdia pointed out that although the current Haein cluster is optimized for LLM learning, the infrastructure transition remains a key task as inference workloads are expected to overwhelm learning over the next five years. SKT plans to complete a nationwide distributed inference network by building large concentrated clusters and distributed edge inference nodes in parallel, along with the transition to a high-throughput, low-latency, and high-availability structure.
"SKT's strategy is a challenging and forward-looking strategy that presents a clear path to monetization in the long run by addressing AI infrastructure optimization challenges while meeting South Korea's Sovereign AI goals," said Inderpreet Omdia Kaur, senior analyst.
윤정훈(yunright@edaily.co.kr)
Companies in Korea are also focusing their capabilities on AI.
We missed the first three years, but now fully supported
It's nice to see you moving forward.