Subscribe

How to cut your cloud bill: egress, rightsizing and storage costs

Find the hidden spend in egress, operations and idle resources

Data reviewed on Sep 8, 2026 · LLM Stats indexes and official prices

The data, today

The context

Cloud bills rarely grow because of an instance's list price. They grow because of what nobody looked at during design: traffic leaving for other regions or the internet, millions of storage operations, instances sized for a peak that happened once, orphaned volumes and logs nobody reads. So the first FinOps job is visibility: knowing which service, which team and which feature is behind each line item on the invoice.

Storage shows clearly why price per GB is misleading. According to our storage guide, the same terabyte can cost up to 21× more depending on provider and traffic pattern, because egress and operations outweigh the space itself. This page ranks the usual savings levers and tells you which metrics to check for each one before you change anything in production.

What to decide

  1. 1

    Where do I start reviewing the bill?

    Enforce tags for team, environment and service, and group spend by them. Sort line items from largest to smallest and study the top ten, which usually hold the problem. Separate usage-based from fixed cost and hunt for untagged resources: they are almost always forgotten ones.

  2. 2

    How do I reduce egress?

    Measure outbound GB by destination: internet, another region or another availability zone. Put a CDN in front of static content, keep chatty services in the same zone and compress responses. If your workload serves heavy traffic, compare storage providers with low or included egress.

  3. 3

    How do I rightsize instances?

    Look at p95 CPU and memory over at least two weeks, not one-day averages. If p95 sits well below capacity, drop a size and watch latency. For steady workloads, evaluate usage commitments with the AWS savings calculator; for variable ones, autoscaling with sensible bounds.

  4. 4

    What does storage really cost?

    Add four components: GB stored per class, read and write operations, egress, and retrieval from cold tiers. Estimate them from your real traffic and compare providers in our storage guide. Set lifecycle rules to move or delete objects nobody reads anymore.

  5. 5

    How do I rein in the observability bill?

    Measure ingested log GB, active metric series and traces per service. Shorten retention for debug logs, sample traces and drop high-cardinality labels. Decide which data is actually queried and archive the rest to object storage, which costs far less per GB.

Common mistakes

  • Comparing providers on compute or per-GB price alone while ignoring egress and operations.
  • Buying long-term usage commitments before rightsizing what you already run.
  • Leaving test environments running overnight and on weekends with no automatic shutdown.

Tools and comparators

AWS savings calculatorFinOps guideCloud options comparatorObject storage comparatorDatabases comparator

Guides to go deeper

Want a recommendation for your case?

Tell us your volume and the options you are weighing. We reply in writing with the numbers of your real usage; no commitment.

Request advice

No provider pays for its position. Indexes come from LLM Stats; prices from each provider's standard API. How we measure

Reduce AI costsMeasure tokens per request and pay for big models only when neededAI customer supportWork out cost per conversation without dropping below your quality barDocument search and RAGFind the right passage before asking the model for an answerCoding with AIMeasure cost per coding task and review what the agent writesHandle traffic spikesAbsorb traffic bursts without dropping requests or the databaseShip to productionDeploy from Git with separate environments and clean configuration
The Codifly brief

A clearer perspective.
In your inbox.

Analysis, guides and new technology comparisons.

You can unsubscribe whenever you like.