The Ministry of Science and ICT is mobilizing efforts to develop South Korean foundation models in artificial intelligence (AI) technology, closely following the advancements of the United States and China.
On September 3, the ministry selected the Naver Cloud consortium as the contractor for a security-focused AI foundation model. A total of 33 industry, academic, research, and public institutions are participating in the initiative. The goal is to build an AI model optimized for the South Korean environment to protect critical national infrastructure and key industrial sites.
Before the official start of the project, Naver Cloud had already begun preliminary training using 4,000 graphics processing units (GPUs) it possessed. With the selection of the contractor, the government will also provide an additional 256 B200 GPUs and 256 H200 GPUs from LG, enabling further model enhancement and real-world testing.
The consortium plans to develop a super-large security-focused model with a final expert mixture (MoE) structure of 700 billion parameters, creating two distinct models: one enhanced for defense capabilities and another for attack capabilities. This approach aims to ensure readiness for all phases of cyber security, including both offensive and defensive operations.
Meanwhile, the development of a medical and bio-specialized model is nearing completion, with the support for the consortium led by Lunit and the Korea Advanced Institute of Science and Technology (KAIST) set to end on September 9. Lunit is working on a comprehensive medical science AI foundation model that connects basic scientific research, drug development, and clinical knowledge, while KAIST is developing the 'Next-Generation Bio Foundation Model K-Fold' for protein structure prediction.
Prior to this, the universal foundation model development project was undergoing a second-stage evaluation, with only the final third verification remaining. The competition has narrowed to three contenders: SK Telecom (SKT), LG AI Research, and Upstage. These elite teams plan to enhance model performance while validating usability and agent performance in real-world industrial, service, and environmental contexts.
The Ministry of Science and ICT requires that all three projects achieve performance levels competitive with major models from the US and China in the global market. There are concerns that key national resources, such as GPUs and data, are concentrated among a few large corporations.
The government emphasizes that the goal of the foundation model development project is to catch up with the frontier models of the US and China. It believes that all resources should be focused on the companies and institutions that can achieve this goal most quickly.
On August 31, Ha Jung-woo, the executive vice chairman of the National AI Strategy Committee, stated, "While it is important to effectively utilize AI, our ability to create powerful AI ourselves remains crucial. We need to establish a system where capable groups can collaborate, and the government should significantly support research and development with GPUs and data."
Looking ahead, the Ministry of Science and ICT plans to begin developing science and technology specialized foundation models in fields such as bio, mathematics, earth science, and secondary batteries next year. A budget of 13 billion won has been allocated for this initiative. A ministry official noted, "An announcement regarding this will be made early next year, and while we are not excluding corporate participation, the focus will be on universities and government-funded research institutions to develop the models."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.