FinOps & Cloud Cost Engineering

The practice of bringing financial accountability to the variable spend model of cloud computing through cross-functional collaboration.

Overview

Cloud cost is an engineering problem

FinOps (Financial Operations) is a cloud financial management discipline that brings engineering, finance and product teams together to make data-driven spending decisions. As cloud bills scale from thousands to millions of dollars per month, unmanaged cost becomes an engineering problem, not just a finance problem.

The three phases

The FinOps Foundation defines three phases. Most organizations loop through them iteratively rather than completing them in order.

Inform — Understand what you are spending and why
Optimize — Eliminate waste and right-size resources
Operate — Continuous cost governance embedded in the engineering workflow

Visibility starts with tags

Without consistent tags (team, environment, service, cost-center), cloud bills are opaque.

> [!IMPORTANT] The first engineering action is enforcing tag policies at provisioning time, then building the allocation dashboard.

The three optimization pillars

1. Eliminate waste — Stop idle resources — Forgotten dev instances, orphaned snapshots, zero-traffic load balancers
2. Right-size — Match instance type to actual CPU/memory utilization — Move a consistently 30%-utilized m5.2xlarge to an m5.large
3. Commitment discounts — Reserved Instances and Savings Plans on AWS, Committed Use Discounts on GCP — 30-72% discount for 1-3 year commitments on baseline workload

Unit economics

Unit economics translate cloud cost into business metrics engineers can optimize: cost per API request, cost per user, cost per inference call.

> [!TIP] A 2x inference cost increase shows up immediately in $/query even when low traffic keeps the total bill flat.

Tooling

AWS Cost Explorer — Built-in cost views
Cost and Usage Report (CUR) + Athena — Raw billing analysis
CloudHealth — Cost management platform
Kubecost — Kubernetes-native cost allocation
Infracost — CI integration: cost diff on every Terraform PR
OpenCost — CNCF project for Kubernetes cost monitoring

Key concepts

FinOps Foundation phases: Inform (visibility), Optimize (waste elimination + right-sizing + commitments), Operate (embed cost governance in engineering workflow). Iterative, not sequential.
Reserved Instances (RI): AWS commitment to use a specific instance type in a region for 1 or 3 years, yielding up to 72% savings vs On-Demand. Convertible RIs allow instance family changes. Zonal RIs provide capacity reservation.
Savings Plans: AWS flexible commitment to spend $/hour (compute or EC2) for 1 or 3 years. Applies across instance families, sizes, OS, and regions automatically. Simpler than RIs for mixed workloads.
Spot/Preemptible instances: AWS Spot Instances / GCP Preemptible VMs: spare cloud capacity sold at 70-90% discount. Can be reclaimed with 2-minute notice. Suitable for fault-tolerant batch, CI runners, and ML training with checkpointing.
Unit economics: Cost per unit of business value: cost per API call, per user, per inference, per GB processed. Makes aggregate cloud spend actionable for engineers.
Kubecost / OpenCost: Tools that break Kubernetes costs down to namespace, deployment, and pod level by allocating node cost proportionally to CPU/memory requests. OpenCost is the CNCF-incubating open standard.

When to use it

Engineering interview questions about managing cloud cost at scale
Designing cost allocation and chargeback systems for platform teams
Evaluating which workloads should use Spot/Preemptible vs On-Demand instances
Building cost guardrails into CI/CD pipelines via Infracost
Explaining trade-offs between Reserved Instances vs Savings Plans

Interview questions

How do you reduce Kubernetes cloud costs without sacrificing reliability?
What is the difference between Reserved Instances and Savings Plans on AWS?