SK Telecom Launches Korea's First Hyperscale AI Model

SK Telecom, which is also South Korea’s leading mobile operator, announced the release of A.X K1 on December 28. It is a 519-billion-parameter AI model trained by a consortium of South Korean companies as part of the government's "Sovereign AI Foundation Model" project. 

The model represents Korea's ambition to become a top-three global AI power alongside the United States and China.

The advancement is more geopolitical than technical. At 519 billion parameters, A.X K1 is comparable in scale to frontier models like GPT or Claude Opus, but the emphasis is not on benchmarks. It is on whether Korea can build domestic AI capability without depending on American or Chinese platforms.

The consortium includes Liner for retrieval technology, SelectStar for data construction, Krafton for multimodal expertise, 42dot for on-device AI, and Rebellions for custom NPU hardware. This is a full-stack approach. Rather than licensing American chips and training models on imported infrastructure, the consortium built each component domestically.

SK Telecom plans to open-source A.X K1 across South Korea's AI ecosystem, providing APIs through developer communities and direct platform access. This is different from how American labs operate. OpenAI and Anthropic control their models tightly. Google distributes some models but retains strategic control. SK Telecom is distributing the capability broadly domestically.

AI capability is becoming a strategic asset at the nation-state level. The United States has been dominant partly because OpenAI, Google, and Anthropic operate within American jurisdiction with access to American compute and talent. China has DeepSeek and other models but faces international regulatory pressure. Korea, Canada, and the EU are now building their own alternatives.