Here’s a scenario that’ll feel familiar. Your team pushed hard for cloud migration. Leadership signed off. Engineers got access to AWS or Azure, and within months, you had a modern, scalable infrastructure. Job done, right?
Then the bills started coming in, and they kept climbing. Month after month, the spend goes up, but nobody can quite explain where it’s all going. Your CFO wants answers. Your board wants a number. And your engineers are too busy shipping features to run cost audits.
This is not a rare situation. A 2025 study by Harness found that roughly 21% of enterprise cloud infrastructure spend, that’s $44.5 billion globally, gets burned on resources that are idle or barely used. Think about that for a second. Nearly one dollar in five, just gone.
Cloud cost optimization is how you fix that. Not by cutting corners or slowing your teams down, but by being deliberate about what you’re running and what you’re paying for. This guide walks through what it actually means, where waste typically hides, and what you can realistically do about it.
What Is Cloud Cost Optimization?
At its core, cloud cost optimization is the practice of making sure you’re not paying for cloud resources you don’t need, and not overpaying for the ones you do.
That sounds obvious. But in practice, it’s harder than it looks. Cloud environments grow fast. Teams spin up resources for projects, experiments, or demos — and don’t always clean up after themselves. Infrastructure gets provisioned for worst-case scenarios and never scaled back down. And because cloud billing is complex (AWS alone has over 630 pricing dimensions), most companies don’t have a clear picture of where their money is actually going.
Cloud cost optimization sits inside the broader world of FinOps, a practice that brings together your finance and engineering teams to make smarter, shared decisions about cloud spending. The FinOps Foundation describes it as a cultural and operational framework that maximizes business value from cloud while creating real financial accountability. In plain terms: it gets the people spending the money and the people tracking the money into the same room.
| Important distinction: Cloud cost optimization is not about spending less. It’s about spending smarter. The goal is to make sure every dollar you invest in cloud is delivering actual value to the business. |
How Bad Is the Waste Problem?
Worse than most CTOs think, honestly. Here’s what the data shows:
| 32% of cloud budgets are wasted on overprovisioned or idle resources — Flexera / N2W, 2025 |
And it doesn’t stop there:
-
- Only 30% of companies can accurately say which team or application is responsible for which cloud spend.
-
- 54% of cloud waste comes down to one thing: lack of visibility.
-
- 48% of IT organizations say rising cloud costs are their single biggest challenge right now.
-
- Nearly 62% of FinOps programs are still at what the FinOps Foundation calls the “crawl” phase — meaning real optimization maturity is still a long way off for most companies.
The pattern we see repeatedly is this: cloud adoption grows fast, governance doesn’t keep pace, and waste quietly accumulates in the background while everyone’s focused on delivery. Nobody set out to waste money. But nobody was watching closely enough to stop it either.
The good news? Most of it is fixable. Companies that run a structured optimization program typically cut monthly cloud spend by 25-30% within the first year. That’s real money.
Where Does Cloud Waste Actually Come From?
Before you can fix anything, you need to know what you’re fixing. In our experience working with mid-market and enterprise cloud environments, waste almost always shows up in the same places:
1. Resources Nobody’s Using
This is the biggest one. Dev environments that run all weekend with no one logged in. EC2 instances that were provisioned for a project six months ago and never terminated. Staging environments that get spun up for a release and forgotten. Every single one of those is charging you by the hour.
| Real example: An EC2 instance running at 8% average CPU — provisioned for peak load, never right-sized after go-live. Small on its own. Multiply by 40 instances across a mid-size team and you’re looking at thousands of dollars a month in pure waste. |
2. Overprovisioning — The “Just In Case” Tax
No engineer wants to be the person whose under-provisioned server caused an outage. So, they size up. Then size up again. Reasonable instinct, expensive habit. Overprovisioned compute alone accounts for roughly 10-12% of total cloud waste across enterprises. Your production database does not need 64 GB of RAM if it’s consistently using 11.
3. Orphaned Resources Piling Up
Every cloud environment accumulates junk over time. Unattached storage volumes. Elastic IPs nobody’s using. Load balancers pointing at nothing. Old snapshots from servers that don’t exist anymore. Individually, they’re tiny. Collectively, they can add 3-6% to your monthly bill — for absolutely nothing.
4. Paying On-Demand Prices for Predictable Work
On demand is the most expensive way to run the cloud. It makes sense for unpredictable or short-lived workloads. For your production database that’s been running 24/7 for two years? You’re leaving serious money on the table. Reserved Instances or Savings Plans on AWS can cut costs by up to 72% on those workloads. Most companies either haven’t committed or made the wrong commitments without looking at actual usage first.
5. Data Transfer Costs Flying Under the Radar
Egress charges are sneaky. Moving data out of the cloud, between regions, between availability zones — it all costs money, and it rarely shows up as a line item anyone’s watching. In multi-cloud setups, it gets even messier. We’ve seen organizations get blindsided by egress bills that were bigger than their compute costs.
6. No Tagging, No Visibility, No Accountability
If your resources aren’t tagged — by team, environment, application, cost center — you can’t see who’s spending what. And if you can’t see it, you can’t fix it. Worse, teams have no reason to be cost-conscious if there’s no visibility into the impact of their choices. Tagging sounds boring. It’s foundational to everything else.
7 Ways to Actually Cut Cloud Waste
These aren’t theoretical. Each one is a lever you can pull right now, and most will show results within 30-90 days.
1. Get Visibility Before You Do Anything Else
Seriously — don’t touch a single resource until you can see your full cloud spend broken down by team, environment, and application. This requires two things: a consistent tagging strategy and a cost management dashboard you check.
-
- Tag everything. Owner, environment (prod/staging/dev), application, cost center. Make it mandatory for new provisioning.
-
- Use AWS Cost Explorer, Azure Cost Management, or Google Cloud Billing as your starting point. For multi-cloud, look at tools like CloudZero, CloudHealth, or Flexera One.
-
- Set up budget alerts. You should know when spend spikes before the monthly bill arrives — not after.
This step alone often surfaces surprises. Most teams find resources they didn’t know existed.
2. Rightsize Your Compute — And Keep Doing It
Rightsizing means matching your instance size to what a workload uses, not what it might theoretically need. AWS recommends targeting 40-60% average CPU utilization. If you’ve got instances sitting at 8-15% utilization consistently, you’re overprovisioned.
-
- AWS Compute Optimizer is free and gives you ML-based rightsizing recommendations across EC2, Lambda, and EBS.
-
- Azure Advisor does the same for Azure VMs.
-
- Don’t do this once. Workloads change. Something sized correctly 6 months ago may be wrong today.
| Rightsizing overprovisioned compute typically delivers 30-50% cost reduction on affected resources, with zero performance impact when done correctly. |
3. Shut Down What Isn’t Running 24/7 (Because It Shouldn’t Be)
Your dev and test environments don’t need to run overnight. Or on weekends. An environment that’s scheduled off from 7 PM to 7 AM on weekdays and fully offline Saturday-Sunday runs for roughly 45 hours a week instead of 168. That’s a 73% reduction in running hours for those resources.
-
- AWS Instance Scheduler, Azure DevTest Labs, and Infrastructure as Code tooling (Terraform, Pulumi) can all automate this.
-
- While you’re at it, run a sweep for orphaned resources. Unattached volumes, unused IPs, forgotten snapshots. Just delete them.
4. Move Predictable Workloads Off On-Demand
If a workload has been running consistently for months, you already know its usage pattern. Stop paying on-demand prices for it. The savings are significant:
| Commitment Type | Discount vs. On-Demand | Flexibility | Best Fit |
| AWS Savings Plans (1yr) | Up to 66% | High — works across instance families | Dynamic fleets, Kubernetes, Lambda |
| AWS Reserved Instances (1yr) | Up to 40% | Medium — tied to instance type/region | Stable production workloads |
| AWS Reserved Instances (3yr) | Up to 72% | Lower — longer commitment required | Long-running critical infrastructure |
| Azure Reservations (1yr) | Up to 72% | Medium | Consistent Azure VM workloads |
| GCP Committed Use Discounts | Up to 70% | Medium | GKE, Compute Engine steady workloads |
| Spot / Preemptible Instances | Up to 90% | Low — can be interrupted | Batch jobs, fault-tolerant workloads |
Sources: AWS Savings Plans pricing, Azure Reservations documentation, Google Cloud Committed Use Discounts
One important caveat: base your commitment decisions on actual usage data, not estimates. Committing to unused capacity trades one form of waste for another.
5. Sort Out Your Storage Tiers
Not all data needs to live on the most expensive storage tier. A lot of it doesn’t need to be instantly accessible at all. Lifecycle policies can automatically move aging data to cheaper tiers without anyone having to think about it.
-
- S3 Intelligent-Tiering, Azure Cool Blob Storage, and Google Nearline are all significantly cheaper than standard storage for infrequently accessed data.
-
- For archival data, S3 Glacier and Azure Archive Storage are a fraction of the cost.
-
- Also, look at data transfer patterns. Unnecessary movement between regions or availability zones adds up fast, especially in multi-cloud environments.
6. Build a FinOps Practice, Not Just a Toolset
Tools can surface waste. They can’t fix the cultural problem of nobody owning cloud costs. That requires practice.
What does that look like in practice? Someone, a person or a team, needs to own cloud cost performance as a KPI. Engineering and finance need to be looking at the same dashboards. There needs to be regular (monthly at a minimum) cost reviews where both sides are in the room. And cost needs to be treated as a non-functional requirement, like performance and security, not an afterthought you check at month-end.
-
- When engineers can see the cost impact of their architectural decisions, behavior changes on its own.
-
- FinOps adoption grew 46% in 2025. About 70% of large enterprises now have a dedicated FinOps function or team.
-
- Companies that embed cost accountability into engineering culture consistently outperform those that treat it as a finance problem.
7. Let AI Do More of the Heavy Lifting
The tooling has gotten genuinely good in the last couple of years. You don’t have to do all of this manually.
-
- AWS Compute Optimizer uses ML to recommend rightsizing across EC2, Lambda, and EBS continuously.
-
- Amazon Q (launched late 2025) lets you ask natural language questions like “Why did my bill go up 40% last month?” and get useful answers.
-
- Azure Advisor now includes ML-based cost forecasting built into the interface.
-
- Third-party tools like Turbonomic, Cast AI, and CloudZero go further — they don’t just recommend, they act, making optimization decisions automatically within parameters you define.
If you’re managing a large, complex multi-cloud estate, manual optimization just doesn’t scale. This is where automation earns its keep.
Cloud Waste Benchmarks by Industry
Waste rates vary a lot depending on your sector. Here’s where different industries typically land — useful context for setting realistic targets:
| Industry | Typical Waste Range | What’s Driving It |
| Financial Services | 22–28% | Heavy DR footprints for compliance, slow decommissioning cycles, over-built redundancy |
| Healthcare & Life Sciences | 24–30% | PHI data retention rules, imaging workload over-provisioning, over-tiered archival storage |
| Retail & eCommerce | 27–33% | Peak-season capacity that never gets scaled back, log and analytics storage bloat |
| Technology & SaaS | 25–31% | Kubernetes headroom, debug environments left running, feature-flag sandboxes that outlive the feature |
| Manufacturing & Logistics | 26–32% | IoT pipeline over-provisioning, high inter-region data movement costs |
| Media & Entertainment | 28–35% | Transcoding farms sized for burst, content archive lifecycle gaps |
The range matters less than the direction. Mature FinOps programs consistently get to 15-20% waste regardless of starting point. Getting from 32% down to 18% on a $5M annual cloud spend is a $700K saving. That’s not a rounding error.
Why Optimization Programs Stall And What To Do About It
Most organizations start a cost optimization initiative with good intentions and stall out within 90 days. Here’s why, and how to get past it:
“We don’t know who owns what.” This comes down to tagging. The fix isn’t glamorous; it’s implementing a tagging policy, enforcing it for new resources through infrastructure as code, and working backwards through legacy environments, starting with your highest-spend services. It takes time. Start anyway.
“Engineering and finance don’t talk.” This is the most common FinOps failure mode. Finance sees a bill. Engineering sees tickets. Neither side has the full picture. Fix it with shared dashboards, joint cost reviews, and, critically, giving engineers visibility into cost data alongside performance metrics.
“We bought Reserved Instances, but they’re not aligned to actual usage.” Commitment management is its own discipline. Regularly check your RI and Savings Plan utilization. AWS allows convertible RI exchanges when commitments drift from actual usage; use them.
“The environment is too complex to manage manually.” At scale, it is. This is the inflection point where automation tools stop being optional. If you’re running hundreds of accounts across multiple cloud providers, you need tooling that monitors, flags, and acts continuously, not a quarterly spreadsheet review.
Frequently Asked Questions
What’s the difference between cloud cost optimization and FinOps?
Cloud cost optimization is what you do: rightsizing, removing waste, shifting to commitment pricing, and fixing storage tiers. FinOps is the framework around how you do it sustainably, the governance, accountability structures, team collaboration, and cultural practices that make optimization stick rather than fade after the first audit.
How much can realistically be saved?
Companies running a structured program typically report 25-30% reductions in monthly cloud spend within the first year. The starting point matters: organisations with no existing optimization and high baseline waste see the biggest early gains. Non-production scheduling alone can cut costs on those environments by 60-70%. The question isn’t really “how much”, it’s “how fast can you start?”
Where should we start?
Visibility first, always. Get your tagging in order and set up a cost management dashboard that gives you real-time spend by team and environment. From there, the fastest wins are eliminating idle and orphaned resources (low risk, immediate impact), rightsizing obvious over-provisioned compute, and scheduling dev/test environments to shut down outside business hours.
Do we need a dedicated FinOps team?
For large enterprises running millions a month across multiple cloud accounts — yes, increasingly so. Around 70% of large enterprises now have one. For smaller footprints, you can get a long way by embedding cost ownership into existing engineering teams with shared tooling and a regular review cadence. The key is that someone owns it. “Everyone is responsible” almost always means nobody is.
What’s the role of AI in cloud cost optimization in 2026?
It’s becoming central to how mature programs operate. The days of generating a recommendation report and hoping someone acts on it are fading. Tools like AWS Compute Optimizer and Amazon Q, Azure Advisor, Turbonomic, and Cast AI are moving toward autonomous optimization — making decisions continuously within guardrails you set. For complex multi-cloud environments especially, this is where the biggest efficiency gains are coming from.
So, What Now?
Cloud waste isn’t a sign that your team did something wrong. It’s almost inevitable when you move fast, give engineers self-service access, and don’t have the governance structures to keep pace with growth. The cloud providers don’t exactly make it easy to stay lean; their default configurations tend to favor provisioning more, not less.
But it is fixable. And it’s worth fixing. The organizations that treat cloud cost optimization as an ongoing operational discipline, not a one-time project someone runs when the CFO complains, consistently get more out of their cloud investment. More budget freed up for actual innovation. Less pressure from leadership about runaway infrastructure costs.
Start with visibility. Fix the easy wins fast. Build the FinOps practice around them. And don’t try to do it all at once; sequenced progress beats ambitious plans that stall.
| At Sthenos, our Cloud Consulting team has worked with mid-market and enterprise organizations to cut cloud waste and build environments that are efficient by design — not just cheaper by accident. If your cloud bills are growing faster than your business is, let’s talk. |
Want to know where your cloud budget is going? Book a free Cloud Cost Assessment with Sthenos. We’ll audit your current environment, identify your top waste sources, and give you a clear action plan — no commitment required.
