Open radio access networks (ORANs), characterized by their emphasis on the disaggregation of hardware and software components of radio access networks (RANs) and the utilization of open standards, present fertile ground for flexibility and innovations. However, they also introduce complexities in managing bandwidth, user equipment (UE) power allocations, power consumption (PC), and virtualization. To maximize energy efficiency (EE), PC must be effectively managed, and transmitted power and bandwidth to UEs must be optimized. The task in ORANs is more complex due to new components, designs, and protocols. This paper is concerned with maximizing EE in the context of ORANs. The primary focus lies in the realm of virtual machines (VMs) allocation within a multi-tiered cloud infrastructure. To achieve such optimization, a thorough PC model is developed, incorporating various network parameters, including the number of resource blocks (RBs), the number of connected UEs of each VM, the total number of UEs, the initial and maximum power of the server, the number of VMs, the transmitted power, signal-to-interference-noise ratio (SINR), and bandwidth. This work also optimizes the EE through resource allocation models, focusing on open interfaces such as enhanced common public radio interface (eCPRI) and 7.2x fronthaul. It also explores virtualization and network protocols’ role in managing channel capacity and latency, highlighting the trade-off between throughput and protocol overhead. A complex optimization methodology, namely the Lagrange multiplier has been used along with three numerical algorithms: sequential quadratic programming (SQP), active set, and interior point to solve the objective of maximizing EE. The outcomes of this research show the influence of network parameters on PC and EE under some constraints, including the number of UEs, the number of VMs, the number of RBs, and channel capacity. The problem has been recognized as a non-linear problem characterized by inequality restrictions with mixed convexity and concavity behavior. However, different convergence criteria can impact the EE, as more stringent convergence criteria may result in less accurate solutions, and vice versa. For the three algorithms, the interior point and SQP algorithms outperform the active set in most of the cases. Under optimized configurations, SQP enhanced EE by approximately 15% compared to traditional architecture. In addition, 50% increment in the number of UEs, concurrent with 50% less VMs’ allocation when using SQP were achieved compared with interior point. Finally, 15% decrease in latency was accomplished when using interior point compared to SQP.