SK Telecom announced its A.X K1 foundation model on December 28, 2025, describing it as Korea’s first 500B‑scale hyperscale AI with 519 billion parameters. The model will anchor a government-backed “sovereign AI” consortium and is intended to be opened to Korean companies and researchers via APIs and open-source access.
This article aggregates reporting from 5 news sources. The TL;DR is AI-generated from original reporting. Race to AGI's analysis provides editorial context on implications for AGI development.
A.X K1 is one of the most aggressive national-scale bets we’ve seen outside the U.S. and China: a 519B‑parameter “teacher model” explicitly positioned as sovereign infrastructure rather than just another commercial LLM. SK Telecom and its consortium partners are trying to compress an entire AI stack—from custom NPUs and data centers to agents and consumer apps—into a single vertically integrated program, with A.X K1 as the brain. That’s notable because it treats foundation models as shared digital public capital, not just a product line. ([news.sktelecom.com](https://news.sktelecom.com/en/2533?utm_source=openai))
Strategically, this pushes Korea into the top tier of AI powers by parameter scale and ambition. The consortium is already wired into real distribution through SKT’s A. (A‑DoT) assistant and Liner’s 11 million‑plus users, and it plans open-source releases and APIs to catalyze a broader local ecosystem. If they execute, Korea can become a regional counterweight to U.S. and Chinese clouds, and A.X K1 could become a default base model for Korean-language and on‑device use cases. ([stocktitan.net](https://www.stocktitan.net/news/SKM/sk-telecom-unveils-a-x-k1-korea-s-first-500b-scale-hyperscale-ai-hk99nyiwi882.html?utm_source=openai))
For the global AI race, A.X K1 is another data point that 500B‑scale reasoning models are becoming table stakes for countries that want autonomy at the frontier. It also subtly raises the bar for “sovereign AI”: it’s no longer enough to have a 70B domestic model—you’re now competing with national projects that bundle hardware, infra, and deployment channels around trillion‑token‑class systems.

