Cloud Cost Optimization / FinOps

Understand, track and reduce the cloud bill without hurting performance, with engineering, finance and business working together.

The cloud is 'pay as you go'. One forgotten big server or NAT/egress traffic can eat lakhs of rupees a month.
You scaled up for sale day but forgot to scale down later: the bill for idle resources keeps running.
Without tagging and cost allocation you do not know how much each team/service spent.
For startups the cloud bill can be the biggest expense. A 30-40% saving is straight profit.
Unit economics: tracking 'cost per order' or 'cost per stream hour' is essential for business decisions.

Monthly bill Rs 1 crore: where is it going? (AWS Cost Explorer, tag-wise)

First see tag-wise where the money goes, then apply the lever for each slice

Compute: which load goes on which pricing

1.24x7 baseline (40 servers): Savings Plan / Reserved Instance, 1-3 year commit, up to ~72% discount
2.Normal-day up-and-down load: On-demand + autoscaling
3.Sale spikes, batch, CI: Spot (Karpenter brings a new node), up to 90% cheaper, 2-min notice

Mistakes: running a DB/critical server on Spot, buying RIs for the peak, ignoring NAT/cross-AZ cost

The other levers

1.Right-size: 30% of servers at only 10% CPU -> a smaller instance (Compute Optimizer)
2.Schedule: dev/test off from 10 PM to 8 AM
3.Data transfer 20%: images/video through the CDN (less origin egress + load), chatty services in the same AZ (avoid cross-AZ charges), VPC endpoints, watch NAT
4.Storage 15%: S3 Standard (new data) -> Standard-IA (after 30 days) -> Glacier (logs older than 90 days)
5.Visibility / showback: team / service / env tags on every resource, a namespace-wise team bill with Kubecost / OpenCost, Budgets + anomaly alerts
6.Result: the bill is ~40% lower, Rs 1 crore -> ~Rs 60 lakh / month, same performance

Right-sizing: if CPU use is 10%, take a smaller instance; size by looking at metrics (CloudWatch/Compute Optimizer).
Autoscaling: add/remove servers according to traffic; shut down dev/test environments at night (schedule).
Spot Instances: unused capacity is very cheap (up to 90% less than on-demand), but can be taken back on 2 minutes' notice. For stateless/batch work.
Reserved Instances & Savings Plans: a 1 or 3 year commitment, a big discount on steady baseline load (up to ~72%).
Storage tiers: S3 Standard -> Standard-IA -> Glacier lifecycle rules; old logs/backups go to cheaper storage.
CDN offload: serve images/videos from the CDN, reducing origin load and data transfer cost.
Egress cost: data transfer towards the internet and cross-AZ/cross-region is paid. Keep chatty services in the same AZ and use VPC endpoints.
Tagging & showback: team/service/env tags on every resource, then show each team its own bill; budgets and anomaly alerts.

Tool	Type	How it is used
AWS Cost Explorer	Cost analysis (AWS)	Service/tag wise bill, forecasts, RI/Savings Plan recommendations; works with AWS Budgets and Cost Anomaly Detection.
Azure Cost Management + Billing	Cost analysis (Azure)	Subscription/resource group/tag wise cost, budgets, alerts and Azure Advisor's right-sizing suggestions.
Kubecost / OpenCost	Kubernetes cost allocation	Shows cost at namespace/deployment/pod level: how much each team is consuming in a shared cluster.
AWS Compute Optimizer	Right-sizing	Looks at EC2, EBS and Lambda usage metrics and recommends a smaller/right size.
Karpenter / Cluster Autoscaler	K8s node autoscaling	Adds/removes nodes based on pending pods, with a spot + on-demand mix.
S3 Lifecycle / Intelligent-Tiering	Storage tiering	Automatically moves objects to IA/Glacier tiers according to access pattern.

Point	AWS Cost Explorer	Azure Cost Management	Kubecost
Scope	AWS account/org only	Azure (and limited AWS)	A Kubernetes cluster (on any cloud)
Granularity	Service, account, tag, resource	Subscription, RG, tag, resource	Namespace, pod, label, container
Recommendations	RI/Savings Plans, rightsizing	Advisor: rightsizing, reservations	Pod request right-sizing, idle nodes
Install	Built-in, nothing to install	Built-in	Installed in the cluster with a Helm chart
Cost	Console free, API calls paid	Free for Azure usage	OpenCost free, Kubecost enterprise paid
Best for	An overview of the AWS bill and commitments	The Azure bill and budgets	Team-wise cost in a shared K8s cluster

AWS Cost Explorer + Budgets: the first step if you are on AWS. See where the money goes.
Azure Cost Management: built-in for Azure workloads, use it along with Advisor.
Kubecost/OpenCost: when all services run in one shared Kubernetes cluster and the cloud bill shows only 'EC2'.
Savings Plans/RI: for the steady 24x7 baseline; spot: for batch, CI, stateless workers; on-demand: for spikes.
Storage lifecycle: for logs/backups that are accessed less after 30 days.

Scenario: cutting an e-commerce company's cloud bill by 40%.

The monthly bill is Rs 1 crore; Cost Explorer tag-wise showed 50% compute, 20% data transfer, 15% storage.
A 1-year Compute Savings Plan was taken on the baseline 24x7 load (40 servers). A big discount on the compute bill.
Sale-day spikes and image processing/batch jobs went on spot instances; Karpenter brings a new node on a spot interruption.
Compute Optimizer showed 30% of servers at 10% CPU: they were moved to a smaller instance type.
Product images are served from the CDN: both origin egress and load dropped; chatty cross-AZ calls were brought into the same AZ.
Logs older than 90 days go to S3 Glacier through a lifecycle rule; dev/test environments are off from 10 PM to 8 AM.
Monthly showback to every team with Kubecost; budget alerts were set. The bill dropped by about 40% with the same performance.

Running a stateful DB or a single critical server on spot: the instance will be gone in 2 minutes.
Buying Reserved Instances for the peak load: the paid capacity sits idle the rest of the time.
Ignoring data transfer cost: NAT Gateway, cross-AZ and egress quietly become a big part of the bill.
Interview tip: say these FinOps pillars: 'right-size, autoscale, spot + savings plans mix, storage tiering, CDN, tagging'. Quote the numbers: spot up to 90%, Savings Plan up to ~72%, a 40% bill reduction in the example.