Accurate energy demand forecasting is essential for optimizing power generation, reducing operational costs, and ensuring grid stability in modern smart grid systems. Recent advancements in Artificial Intelligence (AI) have introduced deep learning architectures capable of capturing complex temporal dependencies in time series data. This study proposes an AI enabled forecasting framework utilizing Long Short-Term Memory (LSTM) neural networks to predict short- and medium-term electricity demand based on historical smart grid data. The LSTM model is trained and evaluated on real world datasets comprising energy consumption patterns, weather conditions, and temporal variables. Performance is benchmarked against traditional statistical models such as ARIMA and machine learning algorithms including Random Forest and Support Vector Regression. Experimental results demonstrate that the LSTM based approach significantly outperforms baseline models in terms of prediction accuracy and robustness, effectively capturing nonlinear trends and seasonal variations. The findings highlight the potential of AI driven forecasting in enhancing demand side management and facilitating sustainable energy planning within intelligent power systems.