The fast spread of cloud computing (CC) has greatly raised the energy consumption of data centers, which emphasizes the need of sustainable operational plans. Accurate workload forecasting plays a crucial role in optimizing resource allocation, reducing energy consumption, and minimizing carbon emissions. This study introduces a CNN-Transformer model, which integrates Convolutional Neural Networks (CNNs) for spatial feature extraction and Transformer architectures for capturing temporal dependencies, providing an effective framework for cloud workload prediction. Empirical evaluations on real-world workload data demonstrate that the proposed model achieves a Mean Squared Error (MSE) of 0.0059, Mean Absolute Error (MAE) of 0.0381, and an \( R^2 \) score of 0.994, outperforming existing state-of-the-art (SOTA) models. These findings demonstrate the model’s capacity to improve forecast accuracy while also enabling energy-efficient cloud operations via real-time carbon emission monitoring and adaptive resource allocation. Furthermore, an analysis of the model’s computational complexity reveals that it maintains an efficient balance between accuracy and processing speed, making it suitable for large-scale cloud environments. This work represents a significant advancement in sustainable cloud infrastructure and offers promising applications in green data center management.