Applying Monolithic to Microservices Strategy for Elastic Container Deployment for AI Applications

As applications adopt Artificial Intelligence (AI), cloud deployment of AI models becomes a challenge due to size, cost and latency. This paper presents a study on adopting a monolithic AI application to a microservices-based AI architecture for cloud. The study involves containerizing services using Docker, deploying them via AWS ECS with both Fargate and EC2 backends and evaluating the performance of the application based on latency and cost. Experimental results demonstrate that, while serverless containers via Fargate offer cost benefits and ease of deployment, GPU-intensive services significantly benefit from ECS with EC2 instances. The microservices approach, combined with strategic deployment choices, improves scalability, maintainability, and system reliability for AI workloads in cloud-native environments.