Our digital infrastructure is continually impacted by societal events and cultural activities, leading to significant fluctuations in usage patterns. However, current forecasting methods are constrained by technical limitations and therefore neglect these external drivers, resulting in inefficiencies, overprovisioning, and frequent disruptions. This project explores cutting-edge AI-driven methodologies to predict internet usage patterns by interpreting broad societal signals. By integrating diverse data modalities and sophisticated machine learning techniques, we are developing an advanced forecasting system that aligns infrastructure resources dynamically with evolving real-world and online activities. Our approach aims to effectively anticipate anomalous activity, enabling preemptive adjustments to avoid critical infrastructure disruptions. Ultimately, our goal is to create digital infrastructures capable of proactively adapting to complex societal dynamics, thereby enhancing reliability, efficiency, and overall service robustness.