Overview

After the AWS migration I ran a focused cost program that took monthly infrastructure spend from ~$4,512 to ~$3,544 — roughly $11,600/year (~₹11L) — without touching reliability or SLAs. Every number was reconciled against authoritative vendor billing exports, and the whole thing was written up as a brief for the CEO.

The problem

The bill looked "fine" until I modelled it line by line. Once I had a consolidated DigitalOcean + MongoDB Atlas + AWS cost model, the waste was obvious: cross-region and cross-internet data transfer, a database tier sized for fear rather than load, and a parallel cloud we no longer needed.

What I did

• Killed NAT egress for the database. Set up Atlas ↔ AWS VPC peering and routed ~10,300 GB/month of database traffic over private peering instead of the NAT Gateway — eliminating ~$390/month with zero code change or redeploy.
• Co-located compute with the data. Moved compute into the database's region to cut cross-region transfer.
• Right-sized the database a full tier (M50 → M40, ~$634/month) by shrinking the working set first — slow-query forensics, index cleanup, and pipeline rewrites. (full write-up)
• Decommissioned DigitalOcean (~$287/month) after a 14-day rollback window, removing the dual-stack risk entirely.

Impact

• $968/month (₹93k) saved — $11,600/year (₹11L).
• A repeatable, evidence-based cost model the business can keep using.
• Savings that came from engineering (peering, right-sizing, decommission), not from cutting capacity.