- SK Telecom is building open AI models to support sovereign AI, putting model ownership at the center of its telco AI strategy
- The operator’s A.X model family is designed for Korean-language performance and industry-specific AI use cases
- SK Telecom’s approach shows how telcos can use models to gain more control over AI deployments
AI models are becoming a strategic control point for telcos, but not every operator is approaching them the same way. Some telcos experimenting with open models, like AT&T, are looking for cheaper tokens. SK Telecom is driving towards something else: sovereignty.
“SK Telecom’s decision to develop its own models reflects both AI sovereignty and cost efficiency considerations,” Taeyoon Kim, SK Telecom’s Head of Foundation Model, told Fierce. “A key objective is to build the capability to design and develop AI models optimized for the Korean language as well as the specific needs and requirements of domestic industries.”
Though it does leverage external commercial and open models for some tasks, SK Telecom has been building its in-house AI model capabilities since 2018. As part of Korea’s Sovereign AI Foundation Model Project, Kim said the operator has been expanding these capabilities into hyperscale foundation models and industry-specific AI as well.
SK Telecom’s proprietary A.X model family includes the 519 billion-parameter A.X K1 as well as the multi-modal 688 billion-parameter A.X K2. Both have been made publicly available through open-source model platform Hugging Face (which was just acquired by Nvidia).
What is SK Telecom doing with AI models?
In short, SK Telecom isn’t just using AI models; it’s trying to become a national-scale AI model builder. And the operator is not pretending this work has been easy.
Kim told Fierce “One of the biggest challenges in developing our own models has been achieving both training stability and high performance for large-scale models within limited time and compute resources, while bringing them to a level where they can be practically deployed.”
But its efforts are starting to pay off. Kim said SK Telecom has signed an MOU with Korea’s Ministry of National Defense to develop and validate specialized AI models based on A.X that are optimized for defense environments. It is also engaged in validation projects to prove out AI applications in real-world manufacturing environments.
“By optimizing model size and architecture for specific service requirements, we aim to improve operational efficiency and scalability in real-world deployments,” Kim said.
What advice can SK Telecom offer to other telcos?
Asked what advice SK Telecom can offer to other operators looking to develop their own models, Kim said the goal doesn’t always have to be building from scratch.
“It is important to clearly define the core AI capabilities that are essential to their own services and businesses,” Kim said. “Operators can leverage open models where appropriate, while continuing to build core technical capabilities by gaining hands-on experience across the full model development and operations lifecycle — including training, fine-tuning and serving.”
Read more about SK Telecom and its AI efforts
Here’s why SK Telecom is building its own AI stack
SK Telecom’s AI data center bet gains traction with 92.5% revenue growth
Korean operators ramp up AI investments in AI superpower push
SK Telecom spins up SK Hyper for 15GW AI data center buildout