- Naver Cloud, Upstage, SK Telecom, NC AI, and LG AI Research selected

- All participating teams demonstrated strong capabilities and meaningful ambition, and are expected to help expand Korea’s AI ecosystem across multiple fronts

- The project now moves into full-scale development of AI foundation models with global impact



The Ministry of Science and ICT (MSIT, Minister Bae Kyunghoon) announced that, under the “Sovereign AI Foundation Model” project, five elite teams have been selected following a document review (15 → 10 teams) and a presentation evaluation: Naver Cloud, Upstage, SK Telecom, NC AI, and LG AI Research.



In the presentation evaluation, the review covered technical capabilities and development experience; the quality of development goals; the strength of development strategies and techniques; and expected impact and contribution plans. All ten teams that presented demonstrated excellent AI capabilities and innovative strategies for developing AI models. Through intense competition, the pool was narrowed to five teams; after budget deliberation and adjustments, and other required procedures, the selections will be finalized.



The five selected teams’ lead and partner organizations, their goals for developing and securing foundation models, and their plans for domestic impact and contribution are as follows.





1. Introduction to the Five Elite Teams

【Naver Cloud Team】

“R&D into foundational technologies for an omni foundation model to expand nationwide access to AI”



Lead and partners: Naver Cloud (lead) with Naver, Twelve Labs, Seoul National University R&DB Foundation, KAIST, POSTECH Research and Business Development Foundation, Korea University Research and Business Foundation, and Hanyang University Industry-University Cooperation Foundation.



Project goal: Build a home-grown omni foundation model* for broader public access and industrial diffusion, and secure a national common foundation intelligence that supports real-time interaction by integrating and generating across text, speech, image, and video.

* A single model capable of unified understanding and generation for heterogeneous data such as text, image, audio, and video.



Ecosystem impact and contribution plan: Operate a national-scale AI service platform based on the omni foundation model to provide accessible, practical AI services for the general public, and—as an open innovation platform—run an AI Agent Marketplace where anyone can develop, register, and distribute agents.



【Upstage Team】

“Development of a global-frontier foundation model and creation of a domestic AI innovation ecosystem”



Lead and partners: Upstage (lead) with Nota, Lablup, Flitto, VUNO, MakinaRocks, Law&Company, Okestro, Day1 Company, Allganize Korea, Korea Financial Telecommunications & Clearings Institute, Sogang University Research and Business Development Foundation, and KAIST.



Project goal: Develop “Solar WBL,” a home-grown AI foundation model aimed at reaching the global frontier, to secure ▴ a sustainable domestic AI ecosystem ▴ global technology leadership ▴ genuine technological sovereignty in AI.



Ecosystem impact and contribution plan: Over three years, deliver public AI services to reach 10 million+ users, expand B2B services in cooperation with enterprises across ▴ law ▴ manufacturing ▴ defense ▴ healthcare ▴ finance, and also provide B2G services.



【SK Telecom Team】

“Development of a post-Transformer ultra-large model that fuses language, multimodality, and action—and implementation of K-AI services”



Lead and partners: SK Telecom (lead) with KRAFTON, 42dot, Rebellions, LINER, SelectStar, Seoul National University R&DB Foundation, and KAIST.



Project goal: Implement K-AI services with a next-generation post-Transformer AI model to accelerate Korea’s AI transformation (AX), enhance public access to AI, and secure global AI leadership.



Ecosystem impact and contribution plan: With dual pillars of enhancing public access to AI and boosting industrial competitiveness through AI transformation, the team will identify and scale B2C and B2B AI use cases.* The consortium and affiliates will pool capabilities to jointly promote public and industrial diffusion of B2C and B2B AI services.**

* (B2C) Promote both general-purpose AI agents and domain-specialist AI agents accessible to everyone.

(B2B) Prioritize key domestic industries such as manufacturing, automotive, gaming, and robotics.

(Common) Provide tailored AI guidance for office workers, students, and digitally underserved groups.

** Target outcomes: ① enhance public access to AI, ② accelerate AI transformation (AX), ③ contribute to the AI ecosystem, ④ utilize domestic AI semiconductors, ⑤ secure international competitiveness and enter global markets.







【NC AI Team】

“Development of a scalable multimodal generative foundation model for industrial AI transformation”



Lead and partners: NC AI (lead) with Korea University, Seoul National University, Yonsei University, KAIST, ETRI (Electronics and Telecommunications Research Institute), AIWORKX, POSCO DX, LOTTE Innovate, HL Robotics, InterX, MediaZen, MBC, and NHN.



Project goal: ▴ Develop a 200-billion-parameter, home-grown large-language foundation-model package ▴ Develop an integrated multimodal perception-and-generation foundation-model package built on a home-grown LLM ▴ Build and operate a “DomainOps”* framework to support industrial diffusion of the sovereign AI ▴ Develop industry-specific foundation models for manufacturing, distribution, robotics, content, and public services.

* DomainOps: a framework that enables domain-specific operation, automation, and optimization



Ecosystem impact and contribution plan: ▴Support industrial AI transformation by linking with the DomainOps platform and B2B systems-integration projects*, ▴ promote diffusion of foundation models by publishing models on the Hugging Face Hub and supporting multiple inference frameworks and formats for maximum usability and hardware compatibility, and ▴connect multimodal perception-and-generation technologies to public-facing government services.**

* Through the DomainOps platform, the team will support tailored AI transformation by sector and, in collaboration with systems integrators (SI firms), promote AX deployment for 40 client companies.

** Based on multimodal perception and generation technologies, these capabilities will be integrated into public-facing government services, supporting digitally underserved groups and enabling the development of models for responding to industrial accidents.



【LG AI Research Team】

“Development of K-EXAONE, a world-leading frontier AI model”



Lead and partners: LG AI Research (lead) with LG Uplus, LG CNS, Superb AI, FuriosaAI, FriendliAI, ESTsoft, ESTaid, Hancom, and Wrtn Technologies.



Project goal: Develop a state-of-the-art AI model at the highest global level to lead the ecosystem for AI adoption.



Ecosystem impact and contribution plan: Advance the three pillars of the AI market—AI model development, AI industry ecosystem building, and exemplar services by sector.



 AI models: Develop high-performance AI foundation models that combine specialized expertise with general-purpose capability.

 AI industry ecosystem: Build a full-stack AI industry that anyone can readily access.

 Exemplar services: Pioneer B2C, B2B, and B2G offerings across sectors to accelerate AI transformation (AX) in real-world industrial settings.



2. Key commonalities among the five elite teams

All five teams:



① Sovereign AI — proved strong capabilities to develop high-level AI models domestically, and were assessed as pursuing the essence of sovereign AI by aiming to develop and secure AI foundation models “from scratch.”

② Open-source policy — presented robust open-source policies so that foundation models they develop and release can also be used commercially by other companies, thereby contributing in many ways to expansion of the domestic AI ecosystem, to the acceleration of diverse AI services, and to improved public access to AI.

③ Ambitious scalability — set out concrete, scalable goals and determination to go beyond legacy language models and advance toward multimodal and omni models, aiming for global-level AI models based on innovative technologies.



3. Support details

Data: All five teams applied for data-related support. Data support will include ▲ joint data procurement ▲ broadcast video datasets ▲ team-specific dataset construction and processing. First, taking into account a pool* of high-quality data providers secured through prior calls and consultations, and the common data needs requested by the elite teams, MSIT will, in consultation with the teams, jointly procure (KRW 10 billion) and process the finalized datasets and begin supplying them from September 2025. In addition, MSIT will provide KRW 2.8 billion per team to build and process datasets specialized to each team’s model-development strategy, and will also support high-quality broadcast video datasets for AI training (KRW 20 billion).

*Data held by public bodies such as the National Archives of Korea, National Institute of Korea History, Statistics Korea, Korean Intellectual Property Office, and Korea Culture Information Service Agency, as well as specialized publications, exam questions, etc.



Talent: Among the five teams, the Upstage team requested support in the talent category; the government will provide linked support for costs such as salaries and research funding for overseas researchers (or teams) that the elite team seeks to recruit.

※ Regarding remaining budget under the relevant program (the World‑Class AI Overseas Talent Attraction Program), the government plans to issue a separate call to support firms across Korea in recruiting overseas talent to lead ambitious AI projects.



GPUs: For this project, SK Telecom and Naver Cloud have been selected* as suppliers (for the period H2 2025–H1 2026) to lease GPUs to the government. Teams that include these companies are excluded from GPU support this year (GPU support becomes available again from H2 2026 after the leasing project ends). The Upstage, NC AI, and LG AI Research teams will receive the required GPU support**.

* In the GPU leasing project (1st supplementary budget, KRW 157.6 billion), SK Telecom was selected as supplier of B200 (1,024 units; supporting two elite teams), and Naver Cloud as supplier of H100 (1,024 units; supporting one elite team).

– An additional batch secured from ELICE Group (B200, 512 units) will be allocated via a separate call.

** In consultation with the Upstage, NC AI, and LG AI Research teams, GPU support will be aligned to their plans (in 2025: either B200 × 512 or H100 × 1,024).

4. Next steps

The government will reflect the evaluation findings from this presentation round, and—through the budget review and adjustment stage—finalize each elite team’s project scope and support package. For other participating teams, the government will share evaluation feedback to help maximize future development potential.



Following the remaining procedures, the government plans to sign agreements with the five elite teams as early as the beginning of August, then commence the challenge of developing and securing AI foundation models with global impact, with support for GPUs, data, and talent according to each team’s needs. By late December 2025, the government will conduct a first-stage evaluation, narrowing the pool from five to four teams, based on the foundation models developed or secured this year.

※ The first-stage evaluation will pursue a multi-dimensional approach; the specific method will be finalized in consultation with the elite teams.

※ Public/expert contests are also being prepared; detailed schedules and formats will be announced later.



Separately, an official kick-off ceremony for the five elite teams will be held soon; details will be shared later. Using the ceremony as a springboard, the government plans to assign the titles “K-AI Model” and “K-AI Company” to the five teams, and will actively support the expansion of their foundation models across the AI ecosystem, improvement of public access to AI, AI transformation (AX) in public, economic and social domains, and use cases in national defense and security.



MSIT Minister Bae Kyunghoon said, “I applaud the capabilities of the five selected teams and the passion of every elite team that joined this effort. This bold endeavor marks only the beginning and will be the launchpad for ‘Everyone’s AI.’ The government will actively support the growth of Korea’s AI companies and institutions and foster the expansion of our sovereign AI ecosystem.”



For further information, please contact the Public Relations Division (Phone: +82-44-202-4034, E-mail: msitmedia@korea.kr) of the Ministry of Science and ICT.



Please refer to the attached PDF.