South Korea plans to launch a state-backed project next year to build a frontier AI model, backed by 4.7 trillion won (about $3.5 billion) in government equity investment, The Korea Times reported Sunday. Second Vice Minister of Science and ICT Ryu Je-myung outlined the plan at a press briefing Friday.

What Happened

According to The Korea Times, the budget would be split between 3.9 trillion won for graphics processors and 800 billion won for training data. The hardware would amount to about 10,000 Nvidia Vera Rubin GPUs made available to participating developers. Setting up a special purpose company to manage the investment is one option under consideration.

Participants will be chosen through an open application and evaluation process, and they will have to match the government's contribution with private funding. The ministry said it would release an outline of the selection criteria later this month.

RelatedAnthropic Asks Appeals Court to Rehear Ruling Upholding Pentagon's Supply-Chain Risk Label

The stated goal is a model that can compete with the leading open-source models from China. The Korea Herald reported that officials consider going head-to-head with the largest U.S. AI companies unrealistic. Ryu tied the timing to a shift in the field, saying, according to The Korea Times, that competition among advanced models entered a new phase after the preview of Anthropic's Claude Mythos in April.

Why It Matters

The plan marks a change in how Seoul supports domestic AI. Its existing national foundation model program, launched in 2025, has been run as an elimination contest among several teams, each receiving a share of government computing resources. Ryu told reporters that support on the order of 1,000 GPUs per team is not enough to catch up with frontier-class systems and that the scale needs to be completely different, according to The Korea Herald.

Concentrating 10,000 top-end Nvidia chips behind a single effort would be a large step up for a government program, though it would still be smaller than the clusters used by the biggest American and Chinese labs. Aiming at Chinese open-source models, rather than at closed U.S. systems, sets a target that officials appear to view as reachable within a national budget.

The matching requirement means the outcome depends on private companies being willing to commit significant capital of their own alongside the state.

Context

The current contest pits LG AI Research, SK Telecom and Upstage against each other in its third stage. The Korea Herald reported that the evaluation runs through January, with two finalists to be named around mid-February. That program will now be steered toward industry-specific models, and finalists can choose to join the frontier project or focus on deploying models in industry. The new frontier competition will also be open to other entrants, including startups.

Separately, the ministry is preparing to distribute about 29,000 government-secured GPUs across the domestic AI ecosystem, the Herald reported.

What to Watch

The funding is part of the 2027 budget proposal and needs approval from the National Assembly, which is expected in December. If it passes, the frontier program could begin as early as March, according to The Korea Herald.

Before then, the ministry's selection criteria, due later in October, will show who is eligible and how much private money is required. Which companies step forward as anchors, and whether the government opts for a special purpose company, remain open questions.