--- id: "2026-10-05-south-korea-3-5b-frontier-ai-model-program" url: "https://postcutoff.com/e/2026-10-05-south-korea-3-5b-frontier-ai-model-program/" as_of: "2026-11-22T13:32:00+01:00" date: "2026-10-05" date_precision: day category: policy-safety importance: 3 confidence: high status: [Confirmed] sources: 5 editor: human editor human_review: null version: "2026-10-18" --- As of: 2026-11-22 13:32 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/e/2026-10-05-south-korea-3-5b-frontier-ai-model-program/ # South Korea plans a 4.7 trillion won ($3.5B) state-backed frontier AI model program from 2027, with 10,000 Nvidia Vera Rubin GPUs On Oct 5, 2026 South Korea's Ministry of Science and ICT said it will launch a 4.7 trillion won (about $3.5B) program in 2027 to build a national frontier AI model, combining state equity with matching private money: 3.9 trillion won for 10,000 Nvidia Vera Rubin GPUs and 800 billion won for training data. A lead developer is to be picked by open tender once parliament passes the 2027 budget. ## Key facts - Budget: 4.7 trillion won (~$3.5B) of government equity investment; 3.9 trillion won for 10,000 Nvidia Vera Rubin GPUs, 800 billion won for training data (Korea Times) - Private participants must match the government investment (Korea Times); open to single companies, consortiums or special-purpose vehicles, including startups and universities (Reuters) - Timeline (Reuters): tender after parliament approves the 2027 budget in December; winner possibly chosen as early as February; program starts from March 2027. Application criteria to be announced later in October 2026 - Goal: a model with leading performance able to compete with leading Chinese open models, plus national-security uses (Korea Times) - Separate from the existing national (sovereign) foundation-model program, which continues with a third-stage review selecting two final teams in February 2027 and shifts toward industry-specific models; finalists may join the frontier project - Announced by Second Vice Minister of Science and ICT Ryu Je-myung - Korea Herald: the existing 'Dokpamo' (independent AI foundation model) project splits into a frontier track and an industrial-deployment track; current teams LG AI Research, SK Telecom and Upstage get roughly 1,000 high-end GPUs each, which the ministry considers insufficient; a plan to distribute about 29,000 GPUs is expected by year-end - Ryu (Korea Herald): 'The goal of securing our own foundation model technology has not changed, and it will not be discontinued'; on finalists: 'We will select two. That is certain' - Application timing (Korea JoongAng Daily, Oct 5): the Ministry of Science and ICT will open applications 'as soon as its budget is finalized at the end of the year', after National Assembly review; the Sovereign AI Foundation Model Project is being redesigned into 'a frontier model development track and an industrial deployment track', and development 'could begin as early as February' (when the sovereign project's final teams are chosen) ## What happened South Korea's science ministry unveiled a plan for a single, heavily funded national frontier-model effort. Instead of spreading money across several teams, it will concentrate chips, data and talent on one lead developer (a company, consortium or special-purpose vehicle) picked by tender, with the state taking equity and private partners matching it. Most of the money buys compute: 10,000 Nvidia Vera Rubin GPUs. ## Why it matters It is one of the largest government-funded frontier-model programs outside the US and China, and an explicit shift from "sovereign foundation models" toward trying to reach the frontier. 10,000 Vera Rubin GPUs is far below the US labs' clusters, so whether it can compete is open. Figures come from the ministry's briefing as reported by Korea Times and Reuters; the formal application criteria were not yet published on Oct 6. ## Your AI and this story - GPT-6.2 Astra (training cutoff May 2026): 127 days after its cutoff - Claude Opus 5.5 (training cutoff June 2026): 97 days after its cutoff - Gemini 4 Hydrogen (training cutoff March 2026): 188 days after its cutoff - Grok 4.7 (training cutoff May 2026): 127 days after its cutoff ## Sources 1. [Korea JoongAng Daily: Korea splits AI push into two tracks amid confusion over sovereign model project (Oct 5)](https://www.koreajoongangdaily.com/business/korea-splits-ai-push-into-two-tracks-amid-confusion-over-sovereign-model-project/12904222) (koreajoongangdaily.com, press) 2. [Korea Times: Korea to launch $3.5 bil. frontier AI project next year](https://www.koreatimes.co.kr/business/tech-science/20261005/korea-to-launch-35-bil-frontier-ai-project-next-year) (koreatimes.co.kr, press) 3. [Reuters (via Yahoo): South Korea plans to develop $3.5 billion frontier AI model starting next year](https://www.yahoo.com/news/articles/south-korea-plans-develop-3-022153172.html) (yahoo.com, press) 4. [Korea Herald (Jie Ye-eun, Oct 5): report on the program's frontier and deployment tracks](https://www.koreaherald.com/article/10893306) (koreaherald.com, press) 5. [The Standard (HK): South Korea plans to develop US$3.5 billion frontier AI model](https://www.thestandard.com.hk/innovation/article/344691/South-Korea-plans-to-develop-US35-billion-frontier-AI-model-starting-next-year) (thestandard.com.hk, press) ## Changes - 2026-10-18: Application timing (after the year-end budget) and two-track redesign, JoongAng Daily - 2026-10-14: Added Korea Herald details (Dokpamo tracks, current GPU allocation, Ryu quotes) from lead - 2026-10-06 (filed): Created ## Related - 2026-10-05: [Cohere launches North 2, its enterprise agent platform, with cross-session memory, a rebuilt agent harness and org-wide token spending caps](https://postcutoff.com/e/2026-10-05-cohere-north-2/index.md)