Vendor lock-in is a common fear, but the financial realities of scaling can push startups to re-evaluate their cloud allegiances. In this real-world case study, a promising Indian FinTech startup analyzing credit risk models saw their AWS infrastructure costs spiraling out of control. By strategically lifting-and-shifting to Google Cloud Platform (GCP) and adopting a serverless-first approach, they achieved a breathtaking 52% reduction in their cloud spend—totaling over ₹32 Lakhs ($40,000) in annual savings.
The startup was originally built natively on Amazon Web Services. Their architecture was robust but expensive:
The Issue: The FinTech app processed 85% of its load between 9 AM and 6 PM. Their AWS Fargate containers handled the API traffic smoothly, but because Fargate bills based on provisioned vCPU and memory rather than strict request execution time, they were continuously billed for baseline API containers that sat mostly idle at night.
Additionally, their heavy EC2 data processing servers were notoriously difficult to auto-scale perfectly without dropping long-running analysis jobs, so the team simply left them running 24/7 on On-Demand pricing to avoid disruptions.
Attracted by a generous startup credit program and aggressive pricing for fully-managed serverless products, the engineering team executed a 3-month migration to Google Cloud Platform.
They containerized their core API and moved it from AWS Fargate to Google Cloud Run. Unlike typical container services, Cloud Run has a "scale-to-zero" capability. During the 14 hours a day when traffic was minimal, their container count dropped to near-zero, and they were only billed per 100 milliseconds of actual request execution. This single change slashed their web-tier compute costs by 35%.
For the long-running credit analysis jobs, they moved from raw EC2 instances to Google Kubernetes Engine (GKE) Autopilot. They also leveraged Google's Custom Machine Types. Instead of buying a pre-packaged 8 vCPU / 16GB RAM machine on AWS and wasting 6GB of RAM, they provisioned exactly 8 vCPU and 10GB RAM nodes on GCP, optimizing compute to exactly what the code required.
For the nodes that did have to run 24/7 (like their primary Cloud SQL databases), GCP automatically applied Sustained Use Discounts. Unbeknownst to the team originally, GCP automatically discounts machines that run for more than 25% of the month, natively reducing database compute costs by up to 30% without requiring an upfront 1-year RI commitment like AWS.
The operational overhaul directly targeted the primary flaw in their old architecture: paying for idle resources.
Total Cloud Cost Reduction
By moving to a combination of scale-to-zero Cloud Run, tightly fit Custom Machine types, and automatic Sustained Use Discounts, the team dropped their monthly infrastructure bill by over 50%. The resulting ₹32 Lakhs in annual savings fundamentally extended their runway and allowed them to hire an additional full-time engineer.
← Back to Home