Cloud data centers are pivotal for scalable computing but face inefficiencies in resource allocation, causing high energy use. This paper optimizes virtual machine (VM) placement by integrating energy and thermal performance considerations. Using performance-to-power ratio (PPR) and server positions in racks, our method dynamically adjusts load thresholds. A Q-learning-based algorithm minimizes energy use while addressing thermal recirculation. Results show an 18.43% energy reduction compared to Particle Swarm Optimization (PSO) and 20% versus genetic algorithms, with fewer Service Level Agreement (SLA) violations and hotspots. This approach enhances thermal and energy management in cloud data centers.