![]() |
Five consortiums selected for the "K-AI Project" begin developing next-generation foundation models, aiming to build Korea’s sovereign AI model, in Seoul on August 4. / Source: Asia Today |
The five consortiums chosen for the “K-AI Project” have officially begun work on next-generation foundation models, each with distinct strategies. The initiative is viewed as a national R&D experiment to establish a “Korean AI sovereignty model,” aiming not only to secure technology but also to strengthen industrial competitiveness and create a self-reliant ecosystem. Evaluations will be held every six months, with one team to be eliminated this December and two finalists selected by 2027.
The five selected groups include Naver Cloud, LG AI Research Institute, SK Telecom, NC AI, and Upstage, ranging from tech giants and telecom firms to game companies and startups, signaling a balanced approach across diverse sectors.
Naver Cloud will focus on “core technology R&D for an omni-foundation model accessible to all.” Leveraging HyperCLOVA X, data centers, cloud infrastructure, and AI applications, it is partnering with video multimodal startup Twelve Labs to develop an exportable AI model for both public and industrial use, with plans for marketplace-based AI service distribution.
LG’s consortium, centered on LG AI Research, will develop “K-EXAONE,” an open-source frontier model, while creating a vertically integrated ecosystem encompassing data collection, AI semiconductors, APIs, and industry-specific services. Partners include LG Uplus, LG CNS, Superb AI, Furiosa AI, Hancom, and Riiid Technologies.
SK Telecom’s consortium aims to commercialize AI through “next-gen transformer-based large models integrating language, multimodal data, and behavior.” It will use its A.X series and develop “omnimodal” technology combining images, audio, and video, partnering with Krafton, 42dot, Rebellions, Liner, and SelectStar. SKT will also optimize Rebellions’ NPU and source GPUs for efficient training.
NC AI will develop an “expandable multimodal foundation model” integrating text, speech, video, and 3D motion using its VARCO LLM and VARCO Vision. With proof-of-concept experience in fashion, smart cities, and robotics, it will build domain-specific models and platforms across manufacturing, distribution, and public sectors, with 54 partner institutions.
As the only startup, Upstage will target industrial AI adoption in medical, legal, public, and education fields. It will design new model architectures and training algorithms to build a frontier model, then expand size, language support, and specialized features step by step. Tech partners include Rablab, Nota AI, and Flitto.
Kim Myung-joo, professor at Seoul Women’s University, noted, “Government support for startups is crucial, while avoiding duplication in GPU allocation. The key is which model proves more practical and sustainable in real-world industries.”
The first evaluation is set for December, marking the start of a high-stakes competition to define Korea’s sovereign AI future.
1
2
3
4
5
6
7