The government is launching a support program providing up to 1.5 billion won annually per individual to encourage Korean AI (artificial intelligence) core talents active overseas to return to Korea. Through the new 'Generative AI Frontier Talent' program set to launch next year, the government aims to secure 20 individuals and boost efforts to catch up with global technology in developing independent AI models.
According to the Ministry of Science and ICT's 'Generative AI Frontier Talent' program description document obtained through the office of Lee Jeong-heon, a lawmaker from the Democratic Party of Korea, on the 7th, the government plans to invest 25 billion won in the program next year. While the 25 billion won budget was disclosed during the announcement of next year's budget proposal, this is the first time that specific details regarding the number of supported individuals and methods have been confirmed.
The primary target initially considered by the government is top-tier AI researchers and developers of Korean nationality active overseas. The plan involves attracting talent who have gained research and development experience abroad to domestic AI companies, thereby accumulating not only human resources but also research capabilities and technology within Korea.
The government calculated the 25 billion won as '20 global talents × 1.5 billion won per person × 10/12 months'. This reflects a 10-month period for next year's program duration based on an annual support of 1.5 billion won per individual. The 1.5 billion won is not merely a salary concept but a budgetary support unit that can include costs necessary for the talent's research activities, such as labor expenses and research funds. Specific support items and methods will be determined during the program execution process, and both the unit amount and number of individuals may vary depending on conditions.
Support will be provided through capital injection. The Ministry of Science and ICT will inject 25 billion won into Korea Development Bank, which will then re-inject the funds into domestic AI companies and others. Companies will utilize the invested capital to secure global talent and develop and enhance independent AI models.
The program is also linked to the development of frontier-level independent AI models. Currently, the competition for 'Independent AI Foundation Models' (referred to as 'Dokpa-mo') has been narrowed down to three elite teams: LG AI Research, SK Telecom, and Upstage. The Generative AI Frontier Talent program explicitly designates companies developing independent AI models as support targets, drawing attention to potential collaboration with these firms. Specific supported companies and allocation of personnel are expected to be determined in the future.
The government is moving to secure 'AI brains' following its efforts in computing infrastructure such as GPUs (graphics processing units) due to intensifying global talent competition. While major U.S. tech giants are offering substantial compensation packages to top AI researchers, talent costs are also rising in China. It is reported that Xiaomi offered an annual salary of around 10 million yuan (approximately 2 billion won) when recruiting key researcher Luo Fuli, who participated in the development of DeepSeek-V2.
The program will be implemented from 2027 to 2030. The government's medium-term fiscal plan includes an annual investment of 25 billion won. The success or failure of the program is expected to hinge on how effectively companies can combine their own compensation, research environments, and computing resources with government support to actually bring top-tier Korean AI talents from overseas back to Korea.
Lee Jeong-heon (Rep.) said, "We actively welcome the government's decision to recruit top-level talents who have left for abroad and to build an independent AI foundation model," adding that "the annual support of 1.5 billion won per person, including labor costs and research funds, is a core strategy to protect South Korea's future growth engines and AI sovereignty."
in TECH
