Lunit Consortium Gains Full NVIDIA Support — "AI Model in 10 Months"

A coalition of 23 Korean organizations across industry, academia, research, and healthcare—collectively known as the Lunit Consortium—has embarked on a national project to develop a medical science–specialized AI foundation model spanning all stages of drug discovery and development.

On October 31, the Ministry of Science and ICT (MSIT) announced the selection of the Lunit Consortium as the lead implementer for the AI Specialized Foundation Model initiative. Government support began November 1, including access to NVIDIA’s latest high-performance graphics processing units (GPUs) for model development.

As part of its broader plan to strengthen Korea’s AI infrastructure and startup ecosystem, the government recently secured over 260,000 units of NVIDIA’s next-generation B200 GPUs. Of these, 256 units have been allocated to the Lunit Consortium—part of a total support package valued at $12.6 million.

The project will run from November 2025 to September 2026, divided into two five-month phases. Exceptional performers may receive additional GPU allocations based on progress. MSIT emphasized that the program’s goal is to establish a world-class, open-source medical AI foundation model by 2026 for immediate adoption across the industry.

Through this initiative, Lunit plans to create a large language model (LLM) integrating diverse layers of biomedical knowledge—from molecular and protein data to omics, drug information, clinical guidelines, and patient datasets—into one unified framework. The aim is to establish an AI system representing a continuous “chain of evidence” across the entire medical-science lifecycle.

The resulting agentic platform will connect specialized sub-models for clinical and research applications, functioning both as a Clinical Decision Support System (CDSS) for physicians and a Biomedical Co-Scientist (BMCS) for researchers. Final deliverables will include large-scale foundation and domain-specific models for medical science, proteins, omics, and clinical data—each with up to 32 billion parameters—along with the CDSS, BMCS, and a public-health chatbot (CareChat) to be integrated into Kakao Healthcare’s platform.

These systems will allow healthcare professionals to query clinical cases and receive evidence-based recommendations with cited sources from medical guidelines and research papers, while researchers can interactively generate and test hypotheses. A pilot service for the public will be demonstrated through Kakao Healthcare’s Health Care Connection Hub.

According to MSIT, the project is expected to enhance clinical accuracy and safety, boost R&D productivity, and expand public access to healthcare benefits.

The models under development are massive in scale—up to 32 billion parameters—and require high-performance GPUs to process the vast molecular and genomic datasets inherent to medical and omics research.

Industry experts note that NVIDIA’s newest B200 GPU is key to the project’s success. While earlier LLMs have relied on A100, H100, or H200 chips, the B200 offers 5–10 times the performance of the A100 and 2–3 times that of the H100/H200. With 256 B200 GPUs allocated, the consortium’s computing capacity is equivalent to roughly 1,280–2,560 A100s or 500–800 H100s—underscoring the immense scale of the infrastructure.

Lunit serves as both the lead institution and the project’s chief technology coordinator, having spearheaded consortium formation and the overall technical roadmap from the outset.

The company brings proven AI expertise in oncology diagnostics and analysis, exemplified by Lunit SCOPE, its AI-driven pathology software for tissue slide interpretation.

A Lunit spokesperson stated, “Our objective is to interconnect every stage—from clinical decision-making to research and drug development—into a single chain of medical-scientific evidence, thereby dramatically improving the success rate of both clinical and R&D outcomes.” The representative added, “We plan to release all models as open-source tools to help establish Korea’s AI sovereignty in medical science. Together with our consortium partners, we aim to secure additional investment, advance commercialization, and bring these technologies to market within the next two years.”