Water-Supply Disaster Driven by AI Demand Concentration: Assessing and Building Regional Resilience

The spatial concentration of water demand from artificial intelligence (AI) and semiconductor industries is emerging as a new disaster driver, outpacing the slow and difficult-to-reverse expansion of water supply in climate-stressed basins. South Korea's first government-led dam program in 14 years has revived debate over the value of new dams; framing the issue in terms of aggregate sufficiency obscures the central challenge. By synthesizing official statistics, national water security is recast as a problem of regional and sectoral allocation under deep uncertainty, with three contributions. First, a previously overlooked structural inversion: was quantified the integrated water-supply plan for the Yongin semiconductor cluster alone already exceeds the entire supply gain of the 2024 fourteen-site climate-response dam package by a factor of approximately 1.6. A Monte-Carlo analysis yields a supply-demand inversion ratio of about 1.5 for the semiconductor anchor alone, rising to 1.7-5.7 when capital-region data centers are included — above parity regardless of the contested data-center estimates. Second, building on the reliability-resilience-vulnerability framework, a two-axis resilience typology was introduced to classify the five major basins by dominant failure mode, with classification stability confirmed by sensitivity analysis. This typology explains a counterintuitive result that the Yeongsan-Seomjin basin, despite holding the highest per-capita water endowment in the country, proved the most crisis-prone, and falling to 24.5% storage in 2022-2023. Resilience is governed by carry-over storage, source diversity, and connectivity rather than by average endowment. Third, the value of dams is argued to lie not in mean firm yield to resilience under uncertainty, requiring a reframing of evaluation metrics. This study provides a screening-level diagnosis based on official statistics rather than model-based prediction, because dominant failure modes differ by basin; siting and allocation must be reconciled before new demand overtakes new supply.