<p>The exponential growth of data centers has led to increased challenges in transaction processing, particularly in the context of nested transactions where resource contention, deadlocks, and race conditions can cause significant performance degradation. These issues are especially pronounced in optical data centers, where the complexity of transaction management and the rapid pace of data transmission can exacerbate existing challenges. This paper presents a novel approach that integrates Fiber Delay Lines (FDLs) for effective conflict resolution with Deep Neural Networks (DNNs) for dynamic load balancing, specifically designed for optical data center environments. We present the mathematical foundations behind the integration of these technologies, highlighting how FDLs can help mitigate nested transaction conflicts by introducing controlled delays to resolve resource contention. Simultaneously, DNNs are employed to predict and balance the load dynamically, optimizing resource allocation and ensuring that the system operates at peak efficiency. The proposed approach not only addresses transaction conflicts but also adapts in real-time to fluctuating workloads, ultimately enhancing performance and throughput across the data center. This work presents both theoretical analysis and simulation implementation strategies for combining FDLs and DNNs, demonstrating their potential to transform resource management in high-demand, high-performance computing environments. Considering the proposed mechanism, the load balancing factor exceeds 0.2 even at a higher load of 0.85. At a load of 0.8, the transaction loss probability is approximately 10⁻³, with a delay of nearly 1.6 slots.</p>