AI has moved well beyond search, translation and image recognition to become a core tool in scientific research. It now identifies new drug candidates, predicts climate patterns and disasters, and designs next-generation battery materials. It analyzes vast datasets from large research instruments and surfaces scientific possibilities that human researchers might never have found on their own.
This shift represents a fundamental change in how research is done. Where scientists once formed hypotheses and worked through repeated experiments and calculations, AI trained on data now proposes new hypotheses and supercomputers rapidly validate them. Scientific competitiveness no longer rests on researchers' ideas alone. It requires national research infrastructure capable of computing, training and verifying those ideas at scale.
Major economies have already made this a national strategy. The United States has moved to build an integrated research environment linking supercomputers, research data, experimental facilities and AI models through programs such as the Genesis Mission and ACCESS. Europe pools resources across member states through EuroHPC, while Japan has developed a single-portal system under HPCI that lets researchers apply for and use supercomputing resources — including the Fugaku supercomputer — through one unified channel. What these efforts share is not simply an expansion of hardware. Each country is building a shared-use ecosystem that connects dispersed national resources, lowers barriers to access and pairs computing power with domain-specific expert support.
South Korea has begun mobilizing its national AI capabilities in science through initiatives such as "AI for Science" and "K-Moonshot," and discussions around building the sixth national supercomputer and shared GPU use point in the same direction. Demand for AI-driven research is growing rapidly, yet building large-scale GPU infrastructure from scratch carries steep practical constraints — power supply, cooling, space and operational staffing. What is ultimately needed is not piecemeal scale-up of individual resources but a shared framework that efficiently connects and collectively deploys the nation's entire computing capacity.
A clear bottleneck already exists in domestic research. Demand for AI model training and GPU-based simulation is surging, but resources remain fragmented across institutions. Researchers must navigate separate accounts, application procedures and software environments for each facility they want to use. While demand concentrates on certain resources, others sit idle — an inefficiency that goes beyond inconvenience and represents a structural drag on national R&D productivity.
A shared-use platform for supercomputing resources, including GPUs, is therefore essential. Researchers should be able to access high-performance computing nationwide as easily as borrowing from a library — checking available capacity in real time through a single portal, submitting requests and starting work immediately without complex setup. That platform must be backed by secure large-volume data transfer, efficient allocation policies and specialized support tailored to the distinct needs of fields such as biotech, materials science and space research. Only then can researchers focus on the scientific questions that matter rather than on learning how to operate equipment.
In the AI era, supercomputing is not a massive machine reserved for a handful of specialists — it is public infrastructure for scientific and technological innovation. South Korean science can take a genuine leap forward when a platform that connects dispersed GPU and supercomputing resources and broadens access is in place. What is needed now is not simply more capacity, but a shared-use framework capable of converting that capacity into national research power. How quickly that foundation is built will determine the pace of South Korea's AI-driven scientific revolution.
Jeong Ki-moon is director of the Supercomputing Technology Development Center at the Korea Institute of Science and Technology Information (KISTI).
nbgkoo@heraldcorp.com
