Loading
Cloud Economics · Pillar2026-07-27· 15 min read
15 ways to cut a cloud bill
that actually work.
Grouped by category, backed by a worked example, and honest about which tactics move the needle and which are theater.
Estimated Cloud Waste
0%
Of IaaS/PaaS spend, 2026 (Flexera)
Commit Discount
0%
Max off on-demand, 3yr reserved
Tactics Covered
0
Compute, storage, egress, governance
Worked Example Cut
0%
Illustrative before/after, this post
← Back to Global Tech SearchNobody budgets for waste.
The line item with
The invisible line item:
The category most
The tactics that make
Applying this to one
The State Of Cloud Waste
Nobody budgets for waste.
Everybody has it.
Flexera's 2026 State of the Cloud Report — a survey of hundreds of cloud decision-makers — puts estimated wasted IaaS/PaaS spend at 29%this year, actually ticking up from the prior year's trend as AI workloads and new PaaS services add billing complexity faster than teams can track it. That number isn't a failure of any one team; it's what happens by default when compute is easy to provision and hard to decommission. Nobody wakes up and decides to waste 29% of their bill. It accumulates, one forgotten dev instance and one oversized database at a time.
This guide is the pillar piece for the cost-focused posts on this site. I already broke down the raw compute-and-egress math across AWS, Azure, and GCP in my pricing comparison with a live calculator. This piece is what you do with that math once you know it — fifteen concrete tactics, grouped by where the waste actually lives, in the order I'd personally tackle them.
Where Cloud Waste Typically Concentrates
Illustrative breakdown of commonly reported waste categories — not a single-sourced statistic
Oversized / idle computeOrphaned storage & IPsUnoptimized egressNo commitment discount usedNon-prod running 24/7
Compute — Tactics 1-4
The line item with
the biggest single lever.
Compute is usually where the largest single chunk of recoverable spend sits, because it's provisioned for peak and then left running at average. These four tactics compound: rightsizing first, then applying commitment discounts to what's left, then moving anything tolerant of interruption to spot capacity.
1
Rightsize to actual utilization
Pull 2-4 weeks of CPU/memory metrics and downsize any instance running under ~40% average utilization. This is consistently the single largest recoverable category.
2
Buy committed-use for steady-state load
1- and 3-year reserved/committed pricing runs 30-55% below on-demand on all three major clouds for workloads that don’t change shape month to month.
3
Use spot/preemptible for fault-tolerant batch work
CI runners, render jobs, and retryable batch pipelines can run on spot capacity at 60-90% off, provided the job tolerates interruption.
4
Autoscale non-production to zero after hours
Dev and staging environments running 24/7 for an 8-hour workday are paying for roughly 3x the compute-hours they need.
Storage — Tactics 5-7
The invisible line item:
what you forgot to delete.
Storage waste is quieter than compute waste — nothing crashes, nothing pages anyone — which is exactly why it survives audits for years. It's also some of the easiest recoverable spend to find, because none of it requires a judgment call about whether a workload still needs the resource.
5
Delete orphaned volumes, snapshots, and unattached IPs
Every major cloud leaves detached EBS/managed-disk volumes and reserved-but-unused IPs behind after instance teardown — they keep billing indefinitely.
6
Set object storage lifecycle tiers
Move infrequently accessed objects from hot to cool/archive tiers automatically. Logs and backups older than 30-90 days rarely need hot-tier retrieval speed.
7
Consolidate small workloads onto fewer instances
A fleet of a dozen 5%-utilized micro-instances carries more fixed per-instance overhead than two properly-sized ones running the same aggregate load.
Egress & Delivery — Tactics 8-11
The category most
cost guides skip entirely.
This is the category I care most about, because it's the one I see skipped in almost every generic FinOps checklist. Egress pricing is where hyperscalers and independent infrastructure diverge hardest — I measured a 9-12x gap in per-gigabyte overage rates in my pricing breakdown — and yet most cost-cutting content jumps straight from “rightsize your instances” to “buy reserved capacity” without ever pricing the bytes actually leaving the network.
8
Put a CDN in front of anything public
Caching static assets and API responses at the edge cuts origin egress directly — bytes served from cache never hit the hyperscaler egress meter at all.
9
Compress and right-size media at origin
Serving AVIF/WebP at the correct dimensions instead of a single oversized master file routinely cuts image egress by more than half before any CDN is involved.
10
Route bandwidth-heavy delivery off the hyperscaler entirely
For workloads that are mostly “serve bytes to the internet,” a bundled-bandwidth VPS is frequently cheaper end-to-end than hyperscaler compute plus metered egress — see the egress-rate gap in my AWS/Azure/GCP breakdown.
11
Keep cross-service traffic in-region
Cross-region and cross-AZ transfer is billed even between your own services; co-locating chatty services in the same region/zone removes that line item entirely.
Governance — Tactics 12-15
The tactics that make
the other eleven stick.
None of the previous eleven tactics survive a single quarter without governance behind them. Rightsizing an instance once is a project; keeping it rightsized as traffic changes is a process — and process only happens if someone owns the number and something alerts them when it moves.
12
Set budgets and anomaly alerts
AWS Budgets, Azure Cost Management alerts, and GCP Budgets are free and catch a runaway resource within hours instead of at month-end invoice review.
13
Tag every resource by team and project
Untagged spend is spend nobody owns, which means nobody is incentivized to reduce it. Tagging is the precondition for every other tactic on this list actually sticking.
14
Review commitment coverage quarterly, not once
Workloads change shape. A reserved-instance purchase that fit last year’s traffic can quietly become over- or under-committed; revisit the mix every quarter.
15
Default every new non-prod environment to a shutdown schedule
The cheapest waste to eliminate is the waste you never create — make auto-shutdown the default for new environments rather than an opt-in cleanup task.
A Worked Example
Applying this to one
realistic workload.
To make this concrete rather than abstract, here's an illustrative model — not a claimed real customer bill — built on the real published rates from my AWS/Azure/GCP breakdown: a small SaaS backend running four always-on general-purpose instances (three production, one that turns out to be an idle staging box someone forgot about) plus 2 TB/month of outbound transfer served directly from origin.
Rightsizing the oversized instances and reserving the steady-state ones cuts the compute line first. Fronting the delivery with a CDN and moving the bulk of that 2 TB behind cache removes most of the metered egress. Killing the orphaned staging box and its two forgotten snapshots removes the rest. None of these steps individually look dramatic — that's the point. They compound.
Illustrative Monthly Bill, Applying Tactics in Sequence
Modeled example using real published AWS/Azure/GCP rates — not an actual customer invoice
If you only take one thing from this guide, take the ordering: fix compute first because it's usually the biggest number, fix egress second because it's the most commonly skipped, and put governance in place last so the first two don't silently regress six months from now. I keep this site's own infrastructure small enough that most of these tactics are overkill for it today — but the ordering is the same one I'd use the day that stops being true.
Written by Abhishek Kushwaha
Founder and writer at Global Tech Search, based in Kathmandu, Nepal. Covers AI, infrastructure, markets, and climate with sourced data and original analysis. More about the author →
Recommended Tools
Affiliate disclosure — we may earn a commission at no cost to you.
Sources
- Flexera, 2026 State of the Cloud Report — estimated wasted IaaS/PaaS spend at 29%
- AWS, Azure, and GCP on-demand/reserved compute rates and egress tiers as compiled in this site's AWS vs Azure vs GCP pricing breakdown (July 2026)
- AWS Budgets, Azure Cost Management + Billing, Google Cloud Budgets & alerts — official documentation
The worked example in this post is an illustrative model built on real published pricing, not a disclosed real customer invoice. Waste-category percentages are a commonly cited illustrative pattern, not a single-sourced statistic — the one number in this post backed by a named external study is the 29% Flexera waste estimate.
Global Tech Search
Fix the line item that's actually leaking.
More independent, sourced cost breakdowns of real infrastructure decisions.
Back to Dashboard →