This paper presents a transmission expansion planning framework that couples Benders decomposition with an operational layer based on a semidefinite programming (SDP) relaxation of the alternating-current optimal power flow (AC-OPF). The mixed-integer linear programming (MILP) master problem selects line and reactive-power investments across stages and operating conditions, while SDP subproblems evaluate operating cost. To accelerate convergence, we introduce a learned dual-feasible proxy that provides certified dual lower bounds, prioritizes subproblems with greatest impact, and warm-starts the remaining SDP solves. Optimality cuts are always derived from exact dual solutions, preserving convergence guarantees. Numerical experiments on IEEE 24-, 118-, and 300-bus systems demonstrate significant reductions in runtime and SDP evaluations, tight primal–dual gaps, and full AC feasibility without load shedding. The results show that a “Benders-first, SDP-second” strategy, enhanced with dual-feasible proxies, offers a practical and scalable solution for secure multistage planning in realistic power networks.