The plan splits in 2027 into two tracks: a frontier model project and a separate track to deploy Korean models across industries.

South Korea plans a 4.7 trillion won frontier AI development programme starting with the new year, funded by the government to build domestic frontier models, AI accelerators, and compute infrastructure.

The initiative aims to combine state equity investment with private funding to focus on computing chips, data, and talent to reach leading model performance.

Participation is open to single companies, consortia, or special-purpose vehicles, with an emphasis on enabling startups, firms, universities, and researchers to take leading roles.

Finalists from current competitions will be reevaluated, with two winners chosen after the third-stage review, while the frontier competition proceeds independently.

Officials acknowledge that the goal is not to match OpenAI or DeepMind head-to-head but to develop a domestic system capable of competing with leading open-weight architectures and aiming to stand alongside China’s best models.

Most of the budget—about 3.9 trillion won—will fund 10,000 Nvidia GPUs, with roughly 800 billion won reserved for data acquisition and overhead.

Strategic implications include boosting domestic AI accelerator demand, giving cloud operators government-backed capacity to lease to smaller firms, and enabling cheaper training budgets for developers.

A lead developer will be chosen through a competitive tender after Parliament approves the 2027 budget, with a possible announcement as early as February.

A new frontier AI competition will open to more participants, rather than extending support only to the current two finalists.

The frontier effort will be distinct from the existing home-grown foundation model programme, which will continue and select two final teams in February 2027.