As fast-charging demand grows, ensuring the operational resilience of electric vehicle (EV) infrastructure under congestion becomes a key challenge in large-scale transportation systems. We study a bilevel charging facility location problem in which EV drivers choose paths based on a disutility function incorporating travel time, charging stops, and queuing delays modeled via an \(M/M/c\) system. We develop both arc-flow and path-based bilevel formulations and derive equivalent single-level mixed-integer linear programming (MILP) reformulations using strong duality and piecewise-linear approximations of queuing delays. To address the computational challenges arising from large-scale transportation networks, we propose an exact decomposition algorithm (DA) that iteratively solves a location master problem and an evaluation subproblem capturing path choice and congestion effects. Several acceleration strategies are introduced to improve convergence speed significantly. Computational results show that the proposed DA substantially outperforms the direct solution of the single-level MILP formulations using a commercial solver on small networks. Extensive experiments on large real-world networks from California and a case study in Québec demonstrate strong scalability, with most instances solved to optimality or near-optimality within reasonable time limits. Benchmark comparisons on a closely related problem from the literature further show that the proposed DA consistently outperforms the state-of-the-art algorithm in both solution quality and computational robustness.