Telco to “AI-co”—a Look at SK Telecom’s Aggressive Approach to Capitalize on AI Demand

In January 2025, SK Telecom officially launched its SKT GPU-as-a-Service (SKT GPUaaS), an on-demand Artificial Intelligence (AI) cloud service it has been preparing to launch as part of its AI Data Center (AIDC) business division at a data center in Seoul, South Korea. This service allows businesses to choose the number and duration of Graphics Processing Units (GPUs) depending on the scale or purposes of AI services and can configure customized packages with flexible prices determined based on the contract period, number of GPUs, and form of prepaid billing. In order to manage the relevant infrastructure, SK Telecom is leveraging its AI Cloud Manager solution to optimize resource utilization. This GPUaaS service is being launched based on NVIDIA’s H100 Tensor GPUs, with plans to introduce one of its latest GPUs—the H200 Tensor—sometime in 1Q 2025. Until the end of January, it is offering a 20% launch discount to help accelerate onboarding of subscribers and has reportedly already fielded inquiries from over 100 companies, ranging from large corporations to research institutions, prior to the launch.

Throughout 2024, especially toward the latter half of the year, SK Telecom began accelerating its efforts to establish its AI infrastructure business. The AIDC business division was only formally launched in November of last year, at SK Telecom’s AI Summit, and the AI Cloud Manager solution was unveiled in October. It has also been preparing for the launch of this solution by investing in Lambda, a global GPU cloud company, in order to secure a stable GPU supply and expertise.

This announcement is in line with a wider company objective for SK Telecom—to morph from being a traditional telecommunications company (telco) into an AI company (or “AI-co”) that is a global leader in AI infrastructure. In December 2024, it unveiled a new corporate structure consisting of seven business divisions, with all of these divisions supported by a shared infrastructure group and a shared human resources group. Only three of those are related to its original telecommunications business:

1. Mobile Network Operator
2. Wireline/Media (this division includes its broadband subsidiary)
3. Enterprise

The other four divisions are completely focused on AI:

1. Adot (A.): This is related to SK Telecom’s domestic personal AI agent.
2. Global Personal AI Agent (GPAA): This division focuses on providing a non-domestic version of its Adot solution. SK Telecom began working with Perplexity, a U.S.-based Generative Artificial Intelligence (Gen AI) conversational search engine developer, to launch this service in the United States, with a beta version scheduled to become available from March 2025 dubbed Aster.
3. AIX: The AI transformation division will be in charge of creating meaningful AI use cases and promoting expansion in domestic and global markets.
4. AIDC

SK Telecom’s AI strategy is not solely focused on hyperscale data centers. It is also working on introducing edge AI by using its existing telco infrastructure—it is one of a few telcos that are founder members of the AI-RAN Alliance. It is currently conducting Proof of Concept (PoC) projects for edge AI in six areas, such as AI robots and healthcare, to understand specialized AI services. What is currently unclear about the change in its corporate structure is just how much of its workforce has been shifted to its new divisions and diverted away from its legacy telco operations. What is clearer though is that SK Telecom is very aligned with NVIDIA’s vision for the industry and it is working very closely with NVIDIA in developing its services.

SK Telecom’s aggressive approach in moving away from its telco focus and reinvesting itself as an AI company is both a reflection of the opportunity that AI presents and the extremely challenging nature of the telco industry, especially since the release of 5G networks. The revenue from its wireless business has only grown ~1-2% Year-over-Year (YoY) for the last 3 reported fiscal years. However, there is a large Capital Expenditure (CAPEX) associated with building not only AI data centers, but then also upgrading its Radio Access Network (RAN) infrastructure so that it can be used for third-party workloads; therefore, SK Telecom’s approach also involves significant risk in that it is relying on the demand for AI services to continue to grow so that it can generate sufficient revenue and recoup the significant investment it has made, especially in NVIDIA’s GPUs.

The transition to become technology companies (techcos) is not a brand new concept within the telco sphere. The opportunity and promise of AI inference revenue is sure to lure more telcos into adopting a similar, though much less aggressive, approach as SK Telecom’s, as they may not be in the fiscal position to completely change their organizational structure and business model, and immediately dedicate a portion of their workforces to focus on AI services over their existing telco business. The outcome of SK Telecom’s strategy, if successful, will ultimately change its market perception away from being considered a telco to more of a conglomerate, which offers AI, telecommunications, and broadband services for domestic and global markets.

The strategy adopted by SK Telecom is a good example for other telcos for how they can attempt to capture the growing AI opportunity. ABI Research has identified a couple of short- and medium-term strategies that telcos can adopt to help them capitalize:

1. Accelerate Internal Data and AI Transformation: Before expanding their efforts beyond their current sphere, telcos must first ensure that they improve their internal utilization of data to improve the efficiency of their operations and, in turn, their networks. AI still relies on clean, relevant, and valuable data in order to provide the maximum benefit in return, so simply putting AI technology everywhere does not automatically solve all problems. Proper utilization of their data will enable AI to help telcos have an easier path toward achieving highly autonomous networks (such as Level 4 as per the TM Forum definition). Improving the efficiency of their telco operations and their networks can help reduce the Operational Expenditure (OPEX) strain and ensure that their existing business becomes less fiscally challenging as telcos begin expanding their horizons toward the AI vertical.
2. Dedicate Some Workforce to Solely Focus on the AI Vertical: Whilst the promise of AI inference revenue is indeed intriguing, the market is still too small for the entire industry to shift immediately. It is important for telcos to shift some of their workforce into dedicated divisions, or even business units, focusing on addressing the growing AI opportunity so that they can be perceived not as “telcos offering AI services” but fully as “AI-cos.” Researching the potential use cases that telcos are better positioned to provide to customers, especially in the edge domain, is crucial for formulating a strong Go-to-Market (GTM) strategy and ensuring a strong Return on Investment (ROI) once they begin to offer services. In the short term, turning existing real estate into AI data centers is a safer, slightly less CAPEX-intensive approach in order to begin offering GPUaaS, rather than purchasing new real estate and building AI data centers. Partnering with NVIDIA more closely allows for the opportunity of a shared cost and revenue model and can make it easier to get into the market than going it alone, given NVIDIA’s brand recognition.