On this page · 7 sections
For five years, the story was improvement: organizations got better at finding and removing waste. That ended this year. Flexera's 2026 State of the Cloud report finds that estimated wasted spend on IaaS and PaaS is 29%, reversing a five-year downward trend. It was 27% in 2025. The same report, which surveyed 753 cloud decision-makers and users, has 85% naming cost management a top challenge, ahead of security.
The State of FinOps 2026 data tells the same story from the practitioner side: self-estimated waste is 29%, and 63% of organizations now have a dedicated FinOps team. More teams, and still more waste. This article explains why and gives you a four-week plan to start recovering it.
One caution before the plan. The 29% figure is a self-reported estimate from survey respondents. It shows how common waste is, not what your own bill contains. The only way to know your number is to measure it.
Why the Number Went Up
Flexera's reading is that AI workloads and newer PaaS and SaaS services have made cost harder to predict: dynamic usage, harder rightsizing decisions and new pricing metrics. In our experience, three patterns drive it:
- AI capacity is expensive to leave idle. GPU instances and hosted model endpoints cost far more per hour than the resources teams were used to forgetting. See why Kubernetes GPU utilization averages 5%.
- Services bill on unfamiliar meters. Tokens, provisioned throughput and per-request pricing do not map onto the instance-hour mental model.
- Teams ship faster than ownership catches up. New environments appear without owners, tags or expiry dates.
Where Cloud Waste Hides
| Category | Examples | Effort to fix |
|---|---|---|
| Orphaned resources | Unattached volumes, old snapshots, idle load balancers, unused Elastic IPs | Low |
| Always-on non-production | Dev, test and staging environments running nights and weekends | Low |
| Idle AI capacity | GPU nodes, SageMaker or Vertex endpoints with no traffic, abandoned notebooks | Low to medium |
| Oversized compute and databases | Instances sized for a peak that never came, over-provisioned database tiers | Medium |
| Storage and data | Hot-tier data that is rarely read, long log retention, cross-region transfer | Medium |
| Commitment gaps | Unused Savings Plans or Reserved Instances, or none at all on steady workloads | Medium |
A 30-Day Plan
Week 1: Baseline and ownership
- Connect all accounts and subscriptions and get one view of spend by cloud, service and environment.
- Assign an owner to every account and large cost center. Where tags are missing, assign by account first and refine later. The method is in our allocation guide.
- Record your baseline monthly spend so you can show savings against it.
Week 2: Quick wins
- Delete unattached volumes, unused IPs and idle load balancers, and expire snapshots past your retention policy.
- Schedule non-production environments to stop outside working hours.
- Shut down AI endpoints and GPU nodes with no traffic. Our FinOps for AI guide covers the checks.
- Confirm each change with the owner before removing anything stateful.
Week 3: Rightsizing and storage
- Review instances and databases with sustained low utilization and resize them in non-production first.
- Move rarely accessed data to cheaper storage tiers and set lifecycle rules.
- For container platforms, apply the steps in Kubernetes cost optimization.
Week 4: Commitments and guardrails
- With waste removed, size Savings Plans or Reserved Instances to the steady baseline that remains. Our commitment guide shows how.
- Set budgets and anomaly alerts so new waste is caught in days, not at month end. See anomaly detection.
- Add a recurring review so each owner sees their own findings.
Report Savings in Terms Finance Will Accept
Savings claims fall apart when they mix categories. Keep three numbers separate: realized reductions (spend that went down against your week-one baseline), cost avoidance (spend that would have happened but was prevented, such as a rejected oversized deployment), and identified but not yet actioned opportunities. Adjust for growth, since a bill that is flat while usage doubled is still a saving, and say how you adjusted. Finance trusts a small, well-defined number more than a large, vague one.
Keep the Savings From Drifting Back
One-off cleanups decay. Waste returns because new resources keep arriving. The teams that hold their gains do three things: they run detection continuously instead of quarterly, they put cost in front of the people who create it, and they require an owner and an expiry for every non-production environment. The broader operating model is in the FinOps framework and our cloud cost optimization guide.
How Varcio Helps
The Varcio platform runs 309 waste detectors across AWS, Azure, GCP, OCI, Kubernetes and cross-cloud checks continuously, so orphaned resources, idle capacity and oversized services show up as ranked findings with estimated savings rather than a monthly report. Talk to our FinOps team if you want a baseline of your own waste in the first week.
Frequently asked questions
How much cloud spend is wasted in 2026?
Flexera's 2026 State of the Cloud report estimates that 29% of IaaS and PaaS spend is wasted, up from 27% in 2025. It is the first increase in five years. The figure is a self-reported estimate from survey respondents, so treat it as a benchmark for how common waste is, not a measurement of your own bill.
Why is cloud waste going up again?
Flexera attributes the reversal to the added cost complexity of AI workloads and newer PaaS and SaaS offerings: dynamic usage, harder rightsizing decisions and unfamiliar pricing metrics make visibility and optimization harder than they were for plain virtual machines.
What are the fastest ways to reduce cloud waste?
The quickest wins are deleting unattached volumes, old snapshots and idle load balancers, scheduling non-production environments to shut down outside working hours, and removing idle AI endpoints. These need little engineering effort and carry little risk.
How long does it take to cut cloud waste?
Quick wins can land within the first two weeks. Rightsizing and storage tiering take another few weeks to test safely, and commitments should be sized last. Ongoing detection and ownership are what keep the savings from drifting back.
