Hydropower-Driven AI Data Center Infrastructure in Nepal: A Screening-Level Techno-Economic Assessment Under Run-of-River Seasonality

Abstract

Nepal has large hydropower potential, but its run-of-river generation profile creates a reliability challenge for always-on digital infrastructure. This paper develops a hydropower to-data-center (H2DC) screening framework for Nepal. The framework combines site screening, electrical topology, seismic allowance, cooling suitability, and an All-In Energy Cost (AIEC) equation. Its main contribution is methodological and case-applied: it translates national hydropower abundance into a practical pre-feasibility screen for AI data center siting under seasonal adequacy constraints. Official sector reporting cites 83,000 MW of theoretical hydropower potential and 42,000 MW of economically feasible potential, while installed hydropower reached 2,538.273 MW by FY 2023/24. For data centers, however, the decisive issue is not annual resource size but dry-season supply. Public evidence indicates minimum-to-maximum hydropower output ratios of about 0.33-0.34 for Nepal’s run-of-river fleet. Three pathways are assessed: a flexible captive micro-grid (1-5 MW), a grid-backed campus (10-50 MW), and a larger hybrid campus with short-duration storage (50-200 MW). A one-way sensitivity screen shows that procurement tariff and financing cost dominate AIEC, while winter balancing determines whether a workload must be flexible or grid-backed. Under baseline assumptions, modeled AIEC ranges from USD 0.075 to USD 0.115/kWh. This range can support inference, sovereign cloud, and flexible batch computing, but it is not enough to support uninterrupted AI training without firm grid access and stronger market validation