Accurately estimating the energy production of a wind farm, for both planning and operation, is still an ongoing challenge for the industry and remains a complex task. Improving wake modeling accuracy is essential for optimizing turbine spacing, minimizing wake losses, and enhancing overall wind-farm efficiency. Of the many available methods, analytical wake models are still widely considered and implemented in the industry. The Jensen model is most commonly used because it is simple to implement and computationally inexpensive, especially when compared to higher fidelity simulations. However, the linear expansion of the wake can compromise the accuracy of the model, especially in complex or hilly terrain. This paper introduces a hybrid method that combines the Jensen formulation with a Long Short-Term Memory (LSTM) neural network. The purpose is to take advantage of the temporal information contained in SCADA records and to capture nonlinear wake effects that the analytical model cannot represent. The approach was tested on the Hill of Towie wind farm in Scotland, which includes 21 Siemens SWT-2.3-VS-82 turbines, using one year of operational SCADA data (2023). The proposed hybrid Jensen–LSTM model achieved an R2 of 0.92 compared to 0.82 for the classical Jensen model with lower RMSE and MAE values, confirming its superior predictive capability. This hybrid physics–AI framework provides a practical pathway for more accurate and computationally efficient wake predictions and can be generalized to other wind farm configurations.