Cloud Cost Management Tools: AWS vs Azure vs Google Cloud
Cost analysis, budgets, price estimates and commitment discounts: how cost management works on AWS, Azure and Google Cloud, and the FinOps habits that matter.
The cloud bill is the one monitoring signal everyone in the company eventually looks at, usually right after it spikes. A forgotten GPU instance, a NAT gateway shipping terabytes, logs nobody reads at premium ingestion rates: none of these show up on a dashboard until finance asks. Each provider ships a decent set of native cost tools, and they’re free. This post covers what each kind of tool is for, how AWS, Azure and Google Cloud compare, and the FinOps habits that make the tools worth opening.
What cost management covers
Cost management, or FinOps when it becomes a practice, is about making cloud spend visible, predictable and efficient, and about making engineers accountable for what they run. The native tools cover four jobs:
| Job | Question it answers |
|---|---|
| Analysis and reporting | Where is the money going, and who owns it? |
| Budgets and alerts | Will we overspend this month, and who needs to know? |
| Estimation | What will this new architecture cost before we build it? |
| Commitment discounts | How do we pay less for the baseline we know we’ll run? |
None of it works without allocation: tags or labels (owner, application, environment, cost centre) applied consistently, plus an account, subscription or project structure that maps to teams. Untagged spend is unowned spend, and unowned spend never gets optimised. Enforce tagging with policy. The identity and governance post covers how.
Cost analysis and reporting
Every provider has an interactive cost explorer that breaks spend down by service, account, region and tag, with forecasts and anomaly detection. For anything beyond basic charts, export the detailed billing data (line-item usage with tags) to your own storage and analyse it with SQL or a BI tool. That’s where the real insight lives: which team’s log volume doubled, which environment runs 24/7 for no reason.
Look into the FOCUS specification (FinOps Open Cost and Usage Specification). The providers increasingly support it, and it makes multi-cloud cost reporting much less painful.
AWS Cost Explorer Soon
Interactive charts of cost and usage by service, account, tag and cost category, with forecasts, rightsizing and Savings Plans recommendations. Cost and Usage Reports (CUR 2.0) and Data Exports deliver line-item detail to S3 for analysis in Athena or QuickSight.
Cost analysis views by subscription, resource group, tag and meter, with forecasts, anomaly alerts, scheduled exports to storage and the ability to analyse AWS costs too. Supports FOCUS-format exports.
Cloud Billing Soon
Billing reports by project, service, SKU and label, with cost breakdown and cost table views. Export detailed billing data to BigQuery for custom analysis and Looker Studio dashboards.
Budgets and alerts
Budgets turn cost data into early warnings. Set them per team, application or environment, with alerts at forecast thresholds (say 80% and 100% of the monthly budget) so someone hears about a problem in week one, not after the invoice. Add anomaly detection, which catches sudden spikes that a monthly budget won’t notice until it’s too late.
One misconception to clear up: budgets don’t stop spending on their own. They alert. If you want a hard stop for a sandbox, wire the budget alert to automation that shuts things down or removes permissions.
AWS Budgets Soon
Cost, usage, reservation and Savings Plans budgets with email or SNS alerts on actual or forecast spend, and budget actions that can apply IAM or SCP policies or stop EC2 and RDS instances. Cost Anomaly Detection flags unusual spend with ML.
Budgets at management group, subscription or resource group scope, alerting at percentage thresholds of actual or forecast cost. Action groups can trigger automation such as shutting down VMs.
Budgets per billing account, project or label with threshold alerts on actual or forecast spend. Programmatic notifications go to Pub/Sub, so a function can react, for example by capping usage. Budgets alert but don’t cap spend by themselves.
Give every sandbox and personal subscription or project a small budget with an automated action. It’s the cheapest insurance against a leaked key spinning up crypto-miners, or a test cluster left running over a holiday.
Price estimation
Pricing calculators let you estimate an architecture before you build it, compare regions and tiers, and see the effect of commitments. They’re only as good as your usage assumptions, though. The usual misses are data transfer, request charges, log ingestion and managed NAT. Estimate those explicitly.
AWS Pricing Calculator
Build and share estimates for architectures across services, now also available in the console with your own discounts and commitments applied. The AWS Price List API exposes prices programmatically.
Estimate costs per product, region and tier, apply reservations and Hybrid Benefit, and export or share estimates. The Azure Retail Prices API gives programmatic access to list prices.
Google Cloud Pricing Calculator
Estimate costs for Google Cloud products with configuration-level detail and committed-use options; the Cloud Billing Catalog API provides SKU prices programmatically.
Commitment discounts
Once you know your steady baseline, commitments are the biggest single lever: typically 30–70% off on-demand prices in exchange for a 1- or 3-year commitment. The models differ:
- Spend-based, flexible (AWS Compute Savings Plans, Azure savings plan for compute, Google flexible CUDs): commit to an hourly spend, apply it across instance families and regions. Less discount, less risk.
- Resource-based (Reserved Instances, Azure Reservations, Google resource-based CUDs): commit to a specific resource type in a region. Deeper discount, less flexibility.
Google is unusual in also giving sustained-use discounts automatically for VMs that run most of the month, with no commitment at all.
Savings Plans and Reserved Instances
Compute Savings Plans discount EC2, Fargate and Lambda for a 1- or 3-year spend commitment, flexible across instance families and regions. Reserved Instances still apply to RDS, ElastiCache, Redshift and OpenSearch.
Azure Reservations and savings plan for compute
Reservations give the deepest discounts for a specific VM size, database or service in a region. The savings plan for compute trades some discount for flexibility across VM sizes and regions. Both combine with Azure Hybrid Benefit.
Committed use discounts (CUDs)
Resource-based CUDs commit to vCPU and memory in a region; spend-based (flexible) CUDs commit to an hourly spend across services. Sustained-use discounts apply automatically to eligible VMs without any commitment.
Don’t buy commitments before you right-size. Committing three years to oversized VMs locks in waste. Right-size and clean up first, then commit to the baseline that’s left, and review coverage quarterly.
Where the savings usually are
Tools show you the bill. These are the places I’d look first on any cloud:
- Idle and orphaned resources. Unattached disks, old snapshots, idle load balancers, unused public IPs, stopped-but-billed resources. The advisors (Trusted Advisor, Azure Advisor, Active Assist) list them for free.
- Non-production schedules. Dev and test running nights and weekends is roughly 65% waste. Schedule shutdowns.
- Right-sizing. VMs and databases sized for a peak that never comes. Use the recommendations, then verify with a week of metrics.
- Storage tiers and lifecycle. Move old data to cheaper tiers automatically (see the storage post).
- Data transfer and NAT. Cross-zone chatter, internet egress and NAT processing charges. Private endpoints and a CDN often pay for themselves.
- Logging volume. Debug logs in production and duplicate log pipelines (see the monitoring post).
- Spot capacity. Batch, CI and stateless workers on Spot at a fraction of the price.
Comparing the providers
The native tools are broadly equivalent, and all free. The differences are in the details:
- AWS has the most granular data (CUR 2.0) and the most options for budget actions, and Cost Anomaly Detection is good. The flip side is a lot of moving parts.
- Azure Cost Management has the friendliest built-in analysis, including AWS cost ingestion, and Advisor ties cost to Well-Architected guidance.
- Google Cloud makes BigQuery billing export the centre of analysis, which is powerful if you’re comfortable with SQL. Sustained-use discounts give savings without commitments.
Common pitfalls
- No tagging strategy. Without consistent owner and environment tags, nobody can be held accountable. Enforce tags at creation.
- Finance-only ownership. Engineers make the decisions that drive cost. Give each team its own cost view and budget.
- Alert fatigue. Budget alerts sent to a shared mailbox no one reads don’t help. Route them to the owning team’s channel.
- Over-committing. Commitment coverage above your stable baseline turns flexibility into sunk cost.
- Optimising once. Cost drifts every month. Make a short monthly review a habit, not a project.
Conclusion
Cost management works when spend is visible per team, budgets warn early, commitments cover a right-sized baseline, and someone reviews the bill every month. The native tools on each cloud are good enough for that. The hard part is tagging, ownership and habit. A concrete next step: export last month’s detailed billing data, group it by owner tag, and look at how much is untagged. Getting that number down is the first FinOps milestone. Every tool side by side is in the Cost Management comparison, and the cloud interview questions include a FinOps section.