AWS cost optimisation for Indian startups: the practical guide
Most Indian startups running on AWS are overpaying by 60-80%. Here's exactly where the waste is and how to cut it, without a disruptive migration or hiring a DevOps team.
Where the money is going
Before cutting anything, you need to know what you're actually paying for. Open AWS Cost Explorer and filter by the last 3 months. In 90% of the Indian startups we audit, the waste comes from the same four places:
- ·Over-provisioned EC2 instances: you spun up a t3.large for a service that uses 8% CPU on average
- ·NAT Gateway data transfer: often ₹15,000-40,000/month that nobody notices until it compounds
- ·Unused or orphaned resources: old snapshots, unattached EBS volumes, idle load balancers
- ·On-Demand pricing for predictable workloads: paying 3x what a Reserved Instance would cost
The quick wins (implement this week)
These changes typically take 2-4 hours and have immediate impact on your next bill:
- ·Right-size your EC2 instances: use AWS Compute Optimizer to get specific recommendations. Most teams can drop to the next size down with zero performance impact.
- ·Delete unused EBS snapshots and volumes: go to EC2 > Volumes, filter for 'available' status, these are unattached and still costing you money.
- ·Review your NAT Gateway: if you're using a NAT Gateway just so Lambda or EC2 can call external APIs, switch to VPC endpoints for AWS services and allow direct internet access for non-sensitive workloads.
- ·Turn off dev/staging environments outside business hours: a Lambda function that stops EC2 at 8pm and starts at 9am saves 55% of compute costs on non-production environments.
The bigger move: Reserved Instances and Savings Plans
If you've been running on AWS for more than 6 months and your usage is stable, this single change will cut your compute bill by 40-72%.
- ·1-year Reserved Instances: 40% discount over On-Demand, no upfront payment required
- ·3-year Reserved Instances: up to 72% discount, makes sense for your database and core API servers
- ·Compute Savings Plans: more flexible than RIs, cover EC2 + Fargate + Lambda across any region
- ·The rule: anything that runs more than 6 hours a day should be on a Savings Plan or Reserved Instance
Moving specific workloads off AWS entirely
Some workloads are genuinely better and cheaper on other platforms. This isn't a full migration, it's carving out the right pieces.
- ·Static frontend: move from S3 + CloudFront to Vercel or Cloudflare Pages. Zero cost, better DX, no configuration.
- ·Postgres database: if you're paying for RDS, evaluate Neon or PlanetScale. For startups under 10 GB, the savings are significant.
- ·Object storage: S3 is fine, but Cloudflare R2 has no egress fees. For Indian apps that serve assets to users, this alone can save ₹10,000-30,000/month.
- ·Redis/caching: ElastiCache has a minimum cost of ~₹3,000/month. Upstash Redis starts at ₹0 with pay-per-use pricing.
The audit checklist
Run through this list to find your specific waste. Most teams find 4-6 items that together account for 60-70% of their bill:
- ·EC2: check CPU utilisation, anything below 20% average is a candidate for downsizing
- ·RDS: check storage auto-scaling, remove unnecessary read replicas in non-production
- ·CloudWatch Logs: set a 14-day retention policy on all log groups (default is never expire)
- ·Load Balancers: any ALB/NLB with zero or low traffic? Delete them.
- ·Elastic IPs: unassociated EIPs cost ₹400/month each, release the ones you don't use
- ·Data Transfer: check your inter-region and internet egress costs, these are often invisible until they're large
Want us to do the audit for you?
We do cloud cost audits for Indian startups. In most cases we find 50-70% savings within the first review. The audit is free, we charge only for implementation.
Book a free cloud auditFrequently asked questions
How much can an Indian startup realistically save on AWS costs?
In our experience auditing 20+ startups, the median saving is 55-65% of the existing bill. The range is wide, some startups have well-optimised setups and save 20-30%, others have years of accumulated waste and cut 80%. The biggest lever is almost always Reserved Instances on compute, followed by NAT Gateway redesign.
Is it safe to downsize EC2 instances without testing?
Not without basic load testing. Before downsizing any production instance, check your CloudWatch metrics for CPU, memory, and network over the past 30 days. If the p95 CPU is below 30% on a busy day, you can almost certainly drop to the next size. Always do it on a staging instance first and monitor for 24 hours.
Should Indian startups use AWS Mumbai or Singapore region?
Mumbai (ap-south-1) for data residency requirements and Indian user latency. Singapore (ap-southeast-1) has historically had slightly broader service availability and marginally lower prices on some services, but the latency difference for Indian users (Mumbai ~10 ms vs Singapore ~60 ms) makes Mumbai the clear choice for user-facing applications.
