The hardware procurement alone is projected to cost 3.5 trillion Korean won (approximately $2.5 billion). Bae emphasized that this figure does not include the substantial additional costs required for data centers, high-speed networks, power infrastructure, and cooling facilities. Currently, Korean companies involved in the state-supported sovereign AI project receive between 500 and 735 NVIDIA B200 GPUs each, a level Bae described as insufficient for reaching the global frontier.
The government is reviewing a “two-track strategy” to bridge this gap. The first track continues the existing sovereign AI project, which focuses on securing domestic models for public institutions and businesses to accelerate industry-wide adoption. The second track involves a new, large-scale project dedicated to developing a world-class AI model with superior general-purpose performance.
Bae framed the development of frontier AI as a matter of national sovereignty rather than mere technological competition. He compared high-level AI models to strategic assets like nuclear weapons, warning that nations may soon seek to control these technologies. “We cannot rule out scenarios where foreign AI services suddenly halt or prices surge by 20 or 100 times,” Bae stated, noting that even open-source models from rivals like China could eventually become closed systems as their influence grows.
While the budget for the frontier project remains undecided, the Ministry of Science and ICT plans to move forward with security-specialized AI models this year. These models will be developed by fine-tuning existing domestic technology with data on software vulnerabilities and cyberattacks. Public sector deployment is expected within 2026, with private-sector adoption following shortly after. To fund these multi-trillion won initiatives, the government is considering “investment-type R&D,” where the state takes equity stakes in special purpose companies or funds rather than providing traditional grants.