Infrastructure As Code In AI Engineering: Towards Reproducible And Efficient Model Deployment

Abstract

The increasing complexity of Artificial Intelligence (AI) systems has amplified the need for reproducible, scalable, and efficient model deployment infrastructures. This study investigates the role of Infrastructure as Code (IaC) in enhancing reproducibility and operational efficiency within AI engineering environments. Using a mixed-method experimental design, three prominent IaC tools; Terraform, AWS CloudFormation, and Ansible were evaluated across on-premises, hybrid, and cloud-native setups. Key performance metrics, including deployment efficiency, reproducibility rate, resource utilization, and scalability index, were analyzed through ANOVA, regression, and correlation modeling. Results revealed that Terraform achieved the highest performance across all parameters, with a reproducibility index of 0.98 and a scalability index of 0.84, demonstrating statistically significant improvements in automation-driven deployments (p < 0.05). Cluster and correlation analyses further highlighted strong associations between reproducibility, scalability, and deployment efficiency, confirming that automated IaC environments yield consistent and high-performing AI deployments. The findings underscore IaC’s pivotal role in bridging the gap between model development and production, offering a systematic, code-driven approach for managing complex AI infrastructures. By enabling reproducibility, reducing configuration drift, and optimizing computational resource allocation, IaC establishes a robust foundation for sustainable, transparent, and efficient AI engineering practices in modern data-driven enterprises.