The cloud-based application component placement problem is complex and often tackled with heuristics to find near-optimal solutions maintaining the application’s performance and avoiding resource over-provisioning. Mapping application components into virtual machines in potentially unpredictable cloud conditions is challenging and lacks performance guarantees. To address this challenge, we present an Artificial Intelligence (AI)-based resource and workload-aware mechanism, formulated as a dynamic decision-making problem solver based on Markov Decision Process (MDP). By leveraging Deep Reinforcement Learning (DRL) models, Depth-Search-First, and Dancing Links algorithms, our approach provides workload-dependent solutions ensuring keeping a balance between the application’s overall performance and the virtual machines’ resource allocation. In our study, we conducted experiments utilizing simulated multi-component-cloud-web-based applications, employing both the Deep-Q-Network (DQN) and Proximal-Policy-Optimization (PPO) architectures. Our evaluations show that the PPO outperforms the DQN, by predicting near-optimal sets of application components within virtual machines in less than 10-episode steps(on average).