- South Korea will launch a state-backed frontier AI project in 2027, aiming to develop models capable of competing with leading open-source models from China.
- The government plans to invest 4.7 trillion won, equivalent to $3.48 billion, in equity rather than simply providing corporate subsidies under the traditional approach.
- Around 3.9 trillion won will be allocated to 10,000 Nvidia Vera Rubin GPUs, while 800 billion won will be used to purchase training data.
- Participating companies will be required to match the government’s investment with equivalent private funding, mobilizing additional resources beyond the public budget.
- One option under consideration is to establish a special-purpose company bringing together leading AI model developers, allowing the state to share in the financial returns from its investment.
- The government believes global AI competition has entered a phase requiring far greater resources, making the previous approach of distributing GPUs among multiple competing companies less suitable for frontier ambitions.
- The National Foundation Model Program launched in 2025 will continue, but its focus will shift toward specialized models and practical applications.
- The two teams reaching the final at the end of 2026 will receive 1,000 GPUs in 2027 and may choose to join the frontier project or continue developing specialized models.
📌 The most notable point is South Korea’s shift from “dividing resources among multiple competing companies” to “concentrating resources to compete at the frontier.” The government plans to invest 4.7 trillion won, including 10,000 Vera Rubin GPUs and a substantial data budget, while requiring matching private capital. This is not merely an AI support policy, but a model in which the state co-invests, shares the risks, and potentially shares the returns from strategic AI capabilities.
