The increasing reliance on cloud platforms for data-intensive applications has amplified the importance of accurate time-series prediction in domains such as cloud resource management, network traffic forecasting, and large-scale system monitoring. Modern cloud environments generate massive volumes of temporal data from virtual machines, containers, microservices, and distributed sensors. Predicting future system behavior from these time-series streams is essential for optimizing resource allocation, preventing service degradation, and improving system reliability. However, conventional forecasting models such as ARIMA and exponential smoothing often fail to capture nonlinear temporal dependencies and long-term contextual patterns present in complex cloud-generated datasets. Deep learning models provide advanced capabilities for modeling sequential data through architectures designed to learn hierarchical and long-range temporal relationships. This study evaluates the predictive performance of multiple deep learning models including Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Temporal Convolutional Networks (TCN), and Transformer-based architectures within cloud computing platforms. Experiments are conducted using cloud-based distributed training environments to assess scalability, computational efficiency, and predictive accuracy across large-scale time-series datasets. Model performance is analyzed using metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and inference latency under varying workloads. The results provide insights into the suitability of different deep learning architectures for cloud-native time-series forecasting, highlighting trade-offs between prediction accuracy, training cost, and deployment scalability in distributed cloud infrastructures.