What happened
Reporting published October 5 describes an **October 2 ministerial briefing** on a proposed **4.7-trillion-won public equity investment**, approximately $3.48 billion, for a frontier AI project beginning in 2027. A special-purpose company involving major domestic model developers is one option. The reported split is 3.9 trillion won for 10,000 Nvidia Vera Rubin GPUs and 800 billion won for training data; participating firms would provide matching private funding. Structure, participants and application criteria were not finalized.
The minister also described retaining the existing national foundation-model programme while shifting its emphasis toward industry applications. [Reuters’ October 6 account](https://www.reuters.com/world/asia-pacific/south-korea-plans-develop-35-billion-frontier-ai-model-starting-next-year-2026-10-06/) corroborates the proposal’s broad scale. The official budget and briefing documents were not recovered. Treat this as a planned budget/equity arrangement, **not enacted appropriation, paid-in capital or an operating jointly owned model institution**.
Potential direction of power
Mixed-contested
Equity may allow the public to share returns rather than merely finance private capacity. But a government-backed national champion could also concentrate model development and access in a small consortium. Voting rights, dividend/return terms, licenses and eligibility determine which effect prevails.
Window of contestability
Unclear
Public equity could create a claim on future returns, but voting rights, access conditions and the final investment terms remain unsettled.
