This paper presents a dynamic resource provisioning framework for virtualized computing environments using limited lookahead control (LLC). The framework addresses the challenge of managing power consumption and maintaining quality of service (QoS) in data centers hosting online services. The approach models the cost of control, including switching costs, and explicitly encodes risk in the optimization problem. Experiments using the Trade6 enterprise application show that a server cluster managed by the controller conserves 22% of the power required by a system without dynamic control while maintaining QoS goals. The controller also reduces SLA violations and host switching activity, with a risk-aware controller reducing SLA violations by 35% compared to a risk-neutral controller. The framework is validated on a small server cluster and extended to larger systems using trace-based simulations. The controller is designed to handle dynamic workloads and adapt to changes in workload intensity. The LLC approach is effective in reducing power consumption and improving system efficiency, and it is scalable through the use of neural networks for larger systems. The paper concludes that a risk-aware controller with β = 2 provides superior performance in reducing SLA violations and host switching activity, and that a neural network-based approach can achieve energy savings comparable to the baseline controller while incurring significantly less computational overhead.This paper presents a dynamic resource provisioning framework for virtualized computing environments using limited lookahead control (LLC). The framework addresses the challenge of managing power consumption and maintaining quality of service (QoS) in data centers hosting online services. The approach models the cost of control, including switching costs, and explicitly encodes risk in the optimization problem. Experiments using the Trade6 enterprise application show that a server cluster managed by the controller conserves 22% of the power required by a system without dynamic control while maintaining QoS goals. The controller also reduces SLA violations and host switching activity, with a risk-aware controller reducing SLA violations by 35% compared to a risk-neutral controller. The framework is validated on a small server cluster and extended to larger systems using trace-based simulations. The controller is designed to handle dynamic workloads and adapt to changes in workload intensity. The LLC approach is effective in reducing power consumption and improving system efficiency, and it is scalable through the use of neural networks for larger systems. The paper concludes that a risk-aware controller with β = 2 provides superior performance in reducing SLA violations and host switching activity, and that a neural network-based approach can achieve energy savings comparable to the baseline controller while incurring significantly less computational overhead.