The recent rise of electric vehicle sharing is driven by its environmental and social advantages, as well as its potential to encourage the adoption of electric vehicles (EVs). Nonetheless, the effectiveness and sustainability of an electric vehicle sharing system (EVSS) strongly depend on the quality of its design and operational planning. In this study, we focus on the strategic sizing of a one-way station-based EVSS. We introduce a bi-objective two-stage stochastic programming model that simultaneously considers the system's economic performance and service quality under uncertain customer demand. To manage the risk of low service levels in adverse scenarios, our model incorporates a conditional value-at-risk (CVaR) measure. Furthermore, to ensure the practical applicability of the model, we explicitly integrate tactical decisions on fleet deployment, as well as operational decisions related to vehicle relocation and battery recharging, into the EVSS design problem. This leads to a large-scale mixed-integer linear program. We propose to solve it by an original approximate Benders decomposition method, in which aggregated information from scenario sub-problems is embedded directly into the master problem through second-stage variables and constraints. Computational experiments on randomly generated instances demonstrate the effectiveness of the proposed approach compared with a state-of-the-art MILP solver. Finally, using a case study, we examine the trade-off between economic profitability and service quality.