My doctoral research project investigates stochastic dynamics in integrated communication–compute system graphs using generative AI and statistical modeling. The objective is to develop scalable, data-driven models that capture time-varying uncertainties across wireless, wired, and distributed compute domains, and to analyze their impact on end-to-end (E2E) latency and reliability. Emerging industrial and mission-critical applications require application-aware performance guarantees across interconnected communication and computing infrastructures. However, these systems are subject to dynamic and domain-specific uncertainties, including channel variability, traffic fluctuations, interference, workload dynamics, and resource contention, that propagate across domains and jointly influence E2E performance. Addressing these challenges requires unified modeling approaches that explicitly capture cross-domain dependencies. The project develops stochastic modeling frameworks that characterize fine-grained temporal dynamics within communication–compute system graphs. High-resolution time-series analysis is integrated with structured statistical modeling and probabilistic generative learning methods to estimate evolving state distributions under dynamic conditions. These models enable scenario synthesis, uncertainty quantification, and robustness analysis in heterogeneous environments. By combining stochastic system modeling with modern generative learning techniques, the project advances data-driven methodologies for analyzing and improving latency and reliability in integrated digital infrastructures. All research is conducted for academic, non-commercial purposes within the doctoral program at KTH Royal Institute of Technology. The doctoral student is supervised by Professor James Gross, KTH Royal Institute of Technology.