The Cost Impact of AWS Lambda's INIT Phase Billing on Serverless vs Container Deployment Decisions: An Empirical Analysis

Serverless computing platforms such as AWS Lambda are widely adopted for their pay-per-use billing model, which eliminates idle resource costs. On August 1, 2025, AWS introducedbilling for the Lambda INIT phase as a separate cost component, fundamentally changingthe economics of serverless deployments. No prior empirical study has measured thefinancial impact of this change across variable workload intensities. This paper presentsa controlled empirical comparison of AWS Lambda and a Dockerized Flask applicationon an EC2t3.micro instance across 600 experimental runs spanning three workload levels:light (500 primes), medium (5,000 primes), and heavy (50,000 primes). Results showLambda meanresponse times of 1,327 ms, 1,432 ms, and 2,375 ms respectively, comparedto Docker’s stable 498 ms, 527 ms, and 545 ms. The INIT phase accounts for over 99% of Lambda response time under light and medium workloads and remains a significant cost contributor under heavy workloads. Critically, the cost crossover point at which Docker on EC2 becomes more economical is not fixed: it shifts from 82,000 requests per day under light workloads to 50,000 requests per day under heavy workloads, a 39% reduction driven entirely by increased INIT phase billing under heavier computation. These findings provide the first empirical, workload-specific guidance for deployment decisions under AWSLambda’s revised pricing structure