• cloud ops
• foundation models
• time series forecasting
• transformer
• virtual machine workload forecasting

Cloud computing enables efficient resource sharing across physical machines. Predicting virtual machine (VM) workloads is crucial for optimizing resource allocation but remains challenging due to complex temporal patterns and long-term dependencies. Traditional models like Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) struggle with scalability. This thesis explores foundation models, particularly transformer-based architectures, for VM workload forecasting. Results show that these models are comparable to LSTMs in performance while being more efficient and adaptable. Their scalability makes them a promising alternative for dynamic cloud environments by simplifying resource management and improving resource exploitation.