South Korea's K-Moonshot: A National Plan to Scale AI-Driven Science
South Korea is moving to centralize its AI capabilities to accelerate scientific discovery and solve high-impact national challenges. The National AI Strategy Committee approved the K-Moonshot implementation plan, with two headline targets: reach the world's fifth-highest research productivity by 2030 and complete 12 high-stakes science and technology missions by 2035.
The approach is simple and ambitious-pool compute, data, and models; put accountable leaders in charge; and let AI automate the scientific loop from hypothesis to analysis. For researchers, this signals a shift from fragmented efforts to coordinated, GPU-rich, model-first programs designed for measurable outcomes.
What's Being Built
- National Science AI Research Center: A hub to integrate research data, GPUs, and AI models across government-funded institutes and science/tech universities.
- Shared compute: Over 8,000 GPUs-sourced from Supercomputer No. 6 and government purchases-dedicated to scientific workloads.
- Domain foundation models: Independent base models for biotechnology, materials, and semiconductors, with specialized models layered on top.
- Autonomous AI scientist workflows: Gradual deployment of systems that iterate from hypothesis generation to experiment design and result analysis.
Targets and Timelines
- By 2030: Double Korea's share of the world's top 1% highly cited papers from 4.1% (2023) to 8.2% and rank fifth globally.
- By 2035: Deliver 12 national missions across strategic fields such as advanced biotechnology, future energy, AI humanoids, quantum technology, space and aviation, semiconductors, displays, secondary batteries, and materials/nano.
How Missions Will Run
The government will finalize 12 missions and appoint Program Directors (PDs) by March after reviewing public proposals from a nationwide contest. PDs will hold strong authority and responsibility across the full stack-from R&D to commercialization-reducing handoffs and speeding translation.
Example Missions
- Biotechnology: Discover and advance 10 innovative new drugs using AI-driven design and screening.
- Future energy: Complete end-to-end design and begin construction of small modular reactor (SMR) ships. For context on SMR technology, see the IAEA's overview.
Budget and Compute Allocation
The plan prioritizes up to 1 trillion KRW by redirecting existing resources, including strategic projects at government institutes and ministry R&D funds. Additional funding will be proposed in next year's budget cycle.
From 2027 to 2031, about 464 billion KRW will fund AI model development in six core fields, with early emphasis on biotechnology and materials. Centralized GPU access and shared datasets aim to reduce duplication and increase model reuse across programs.
What This Means for Labs and R&D Teams
- Data readiness first: Standardize metadata, ontologies, and data quality pipelines so your datasets are portable to national models.
- Compute strategy: Plan job scheduling and model training around the shared 8,000+ GPU pool; prioritize fine-tuning and retrieval over training from scratch where possible.
- Model ops: Establish evaluation suites (task, safety, and bias checks) and versioning for foundation and specialized models.
- Autonomous workflows: Pilot closed-loop "AI scientist" runs on low-risk experiments; scale as reliability metrics stabilize. For a primer on self-driving labs, see this Nature explainer.
- Talent and training: Upskill staff in prompt engineering, agentic workflows, lab automation, and data/ML operations.
- Commercialization path: Engage early with PDs on IP, regulatory, and tech-transfer plans to compress time-to-market.
Timeline to Watch
- Now-March: Mission selection and PD appointments via the nationwide proposal process.
- 2027-2031: Major investment window for domain AI models; expect shared benchmarks and model releases.
- 2030: Citation impact milestone (top 1% share to 8.2% and fifth globally).
- By 2035: Completion of 12 national missions.
Resources for Research Teams
- AI for Science & Research - practices for lab automation, model selection, dataset strategy, and evaluation.
- AI Learning Path for Research Scientists - tools and workflows for hypothesis generation, experiment design, and autonomous-agent loops.
The signal is clear: centralized compute, domain models, and accountable leadership are now the default for high-impact science programs. If your team prepares its data, workflows, and talent around that reality, you'll be ready to plug into K-Moonshot and ship results that matter.
Your membership also unlocks:
