Cloud Cost Optimisation in 2026: An India-First Playbook (with Real INR Numbers)

TL;DR. Across the last 12+ Kalyankari Softnet cloud cost audits, our average client cut 27% off their monthly cloud bill in the first 90 days — without rewriting a single application. This playbook is the exact ordered checklist we use, with INR numbers from real Indian engagements (anonymised). If you'd rather skip the reading, try our free cost calculator for an instant estimate, or book the free 30-min audit to get a precise number for your environment.

Why your cloud bill is probably 20-35% too high

Cloud waste comes from four predictable places. We've debugged each of these on FinTech, EdTech, SaaS and D2C clients across India over the last four years, and the pattern is the same regardless of provider or industry:

• Over-provisioning. The instance type someone picked when the team was 5 people is still running, even though the team is now 50.
• No commitment discounts. 80% of compute is on-demand priced, when 60-70% of it is steady-state and could be on Savings Plans / Reserved Instances / Committed Use Discounts.
• Forgotten resources. Dev environments running 24×7. EBS volumes attached to terminated instances. NAT Gateways nobody can explain. Untagged so nobody noticed.
• Wrong-tier storage. Years-old logs on S3 Standard at ₹2/GB/month when they could sit on Glacier Deep Archive at ₹0.08/GB/month.

If you have never run a structured cost optimisation pass, 22-35% savings is realistic. If you've already optimised once, expect 10-18% on a re-pass. Heavily-optimised environments will yield 4-10% — still worth the effort because the absolute numbers compound.

The ordered checklist (highest ROI first)

Order matters here. Each item is sequenced by the ratio of savings per hour of engineering time required to implement it. Run them top-to-bottom.

1. Reserved Instance / Savings Plan / CUD portfolio review (first 2 weeks, biggest single lever)

Pull the last 90 days of compute usage. Identify everything that has run continuously — not bursty, not "9-to-6", but always-on. That fleet is your commitment-discount candidate. On AWS, a 1-year No-Upfront Compute Savings Plan typically saves 30-35% off on-demand. A 3-year all-upfront convertible RI on a stable production fleet can save 60%+.

Anonymised real example. A Series-A FinTech client in Gurugram with ₹4.2L/month AWS spend. 78% of compute was steady-state. Buying a 1-year Compute Savings Plan covering 80% of that steady fleet dropped the monthly bill by ₹78,000 immediately, with no application changes. ROI: instant.

2. Idle workload sweep (week 3, easy wins)

Run AWS Cost Explorer's "Rightsizing Recommendations" report (or Azure Advisor / GCP Recommender). Cross-reference with CloudWatch CPU/memory metrics over 14 days. Anything below 5% utilisation 24/7 is either oversized or unused.

Common findings:

• Dev RDS instances running 24×7 when devs work 9-7. Stop them on a schedule (CloudFormation StackSets + Lambda, or Instance Scheduler) — saves ~70% of dev DB cost.
• Backup EBS volumes from instances terminated 18 months ago. Snapshot then delete.
• Old Elastic IPs not attached to anything (₹400/month each, easy to lose 10-20 of them in a busy account).
• Load balancers serving zero traffic (someone tested an architecture and forgot).

3. Right-sizing (weeks 4-6, requires care)

For each remaining workload, compare instance size to actual usage. Common patterns we find:

• m5.4xlarge running at 12% CPU — should be m5.xlarge or m6i.large. Saves 50-75%.
• RDS db.r5.2xlarge at 8% CPU and 30% memory — db.r6g.large with Graviton saves 40% and is faster.
• Production EC2 fleets sized for "peak Diwali traffic" running at 15% utilisation 11 months a year. Auto-Scaling Group with a baseline of fewer instances saves 30-50%.

Caution: Right-sizing is the place where teams accidentally cause incidents. Always change one workload at a time, watch for 48 hours, and have a rollback ready. We use AWS Compute Optimizer / Azure Advisor recommendations as a starting point but never apply them blind.

4. Storage tiering (weeks 6-8)

The biggest underrated win. Most accounts have:

• S3 buckets full of old logs / backups on Standard storage at ₹2.20/GB/month. Most of this should be on S3 Standard-IA (₹1.20/GB), then Glacier (₹0.30/GB), then Glacier Deep Archive (₹0.08/GB) for anything older than 365 days. Lifecycle policies do this automatically.
• EBS gp2 volumes that should be gp3 — same performance, ~20% cheaper, and you can dial up IOPS independently. Migration is online, no downtime.
• Snapshots from years ago nobody can attribute. Apply a snapshot lifecycle policy to retain only the last 30/90 days for non-prod, auditable retention for prod.

Anonymised example. An EdTech platform with 50,000+ students had 38TB of old class recordings on S3 Standard, costing ₹68,000/month in storage alone. Lifecycle policy moving anything older than 90 days to S3 Glacier Instant Retrieval, and anything older than 1 year to Glacier Deep Archive, dropped that to ₹16,000/month. Same access patterns, no application changes.

5. Network cost rationalisation (weeks 8-10)

The most invisible cost line on most cloud bills:

• NAT Gateway data processing at ₹0.45/GB. For high-throughput workloads, this can be ₹50K-₹2L/month. Often most of that traffic is to AWS-internal services that should use VPC Endpoints (Gateway Endpoints for S3/DynamoDB are free).
• Cross-AZ data transfer between EC2/RDS in different AZs. ₹0.80/GB. Co-locate by AZ where latency allows.
• Inter-region replication running at full bandwidth when most of it could be batched at off-peak.

6. Architecture-level redesign (months 3+)

Once the operational wins are in, the bigger structural levers come into play:

• Lambda + Fargate Spot for batch / async workloads instead of always-on EC2.
• Aurora Serverless v2 for variable-load Postgres / MySQL workloads.
• S3 Express One Zone for low-latency storage that doesn't need multi-AZ durability.
• Karpenter on EKS for aggressive node right-sizing + Spot integration.

Provider-specific tactics

AWS-specific (the most lever)

• Compute Savings Plans cover EC2, Fargate, and Lambda — broader than Reserved Instances. Almost always the right starting commitment vehicle.
• Graviton (ARM) instances are 20-40% cheaper than Intel/AMD equivalents for most workloads. Test compatibility, then migrate where it works.
• Spot for non-critical workloads (batch processing, CI runners, ML training) — 70-90% savings vs on-demand.
• For workloads in ap-south-1 (Mumbai), capacity for newer instance families (m7g, c7g) is now solid. Don't keep using older families just out of habit.

Azure-specific

• Azure Hybrid Benefit is the biggest lever for any organisation with Microsoft licences — can knock 40%+ off Windows VMs and SQL Server. Often missed.
• Reservations + Savings Plans (yes, Azure has both now). Plans cover compute flexibly; Reservations are for fixed instance types.
• Azure Spot VMs for stateless workloads.

GCP-specific

• Committed Use Discounts (CUDs) are simpler than AWS RIs — buy resource-based commitments without picking instance types. 25-37% savings.
• Sustained Use Discounts apply automatically — but only on Compute Engine VMs, not GKE Autopilot.
• Spot VMs (formerly Preemptible) — 60-91% off on-demand, with predictable 24-hour limits.

The 5 mistakes that cost most teams 10-20%

1. Buying RIs / Savings Plans without baseline data. Commit only to what you can prove is steady. Over-commit is more expensive than no commit.
2. Right-sizing on stale metrics. Look at 14-30 days of actual utilisation, not the day you happen to check.
3. Untagged resources. If you can't attribute a cost line to a team or product, you can't optimise it. Tag policy + Service Control Policy enforcement.
4. Optimising in production first. Always start in dev/staging — same architectural patterns usually apply, fewer consequences when wrong.
5. One-time optimisation. Cloud cost is a moving target — usage grows, AWS launches new services, your team adopts new patterns. Quarterly reviews keep the win compounding.

When to bring in outside help

Honestly: if your monthly cloud bill is under ₹50,000, you can probably do this yourself with the checklist above. AWS Cost Explorer + Trusted Advisor + a free Saturday is enough.

If you're between ₹50,000 and ₹3 lakh/month, an external audit usually pays back in week 4-6. The hardest part is having the time and discipline to actually run the playbook end-to-end while keeping production stable. We do this for clients across Delhi NCR, Bangalore, Mumbai and Pune via our cloud services engagement, with localised expertise documented on our cloud cost audit and FinOps consulting pages.

Above ₹3 lakh/month, the lever often shifts to Enterprise Discount Programs, multi-cloud arbitrage and architecture-level redesign. That's a longer engagement, but the absolute savings get large fast — typical ROI is 8-15× the engagement fee.

Try the calculator first

If you'd rather see a directional savings number for your specific situation in 30 seconds before reading any further, we built a free cloud cost calculator based on the math from the last 12+ KS audits. No signup, no PDF gate. Plug in your spend, your provider and your optimisation level — get a credible savings range immediately.

Or if you want a precise number for your real environment, the free 30-minute cloud cost audit is the fastest path. We read your real bill, send a one-page memo within 48 hours, and tell you in the first 5 minutes if there's nothing here for you. No commitment.

Author: Dharmendra Jha is the founder and principal engineer at Kalyankari Softnet, a Delhi-based cloud and DevOps consultancy. Connect on LinkedIn or WhatsApp +91 8882004118.