Key takeaways
Gartner forecast worldwide public cloud end-user spending of $723.4 billion in 2025, up from $595.7 billion in 2024.
Organizations estimate that 29 percent of their IaaS and PaaS spend is wasted, up from 27 percent. Flexera attributes the first increase in five years to AI workloads.
84 percent of organizations name managing cloud spend as their top cloud challenge, and cloud budgets were already exceeded by 17 percent.
The fix is allocation before optimization: every dollar with an owner and a unit cost, so that rightsizing and commitments are decided on evidence.
Spending keeps rising
Gartner forecast that worldwide end-user spending on public cloud services would reach $723.4 billion in 2025, up from $595.7 billion in 2024, an increase of about 21 percent.1 Respondents to Flexera’s 2025 survey expected their own cloud spending to rise by 28 percent in the following year.2
Note: Forecast published November 2024.
Source: Gartner1.
Waste has stopped falling
Flexera asks organizations each year to estimate how much of their cloud spend is wasted. In 2026 the answer was 29 percent of IaaS and PaaS spend, up from 27 percent, the first increase in five years. Flexera links the rise to cloud-based AI workloads.3
The governance response is visible. Dedicated FinOps teams reached 63 percent of respondents and cloud centers of excellence 71 percent.3 Yet managing cloud spend remains the most frequently named challenge, cited by 84 percent of respondents in 2025, when budgets were already exceeded by 17 percent.2
Where waste comes from
Waste is easier to remove once it is named. Most of it falls into a few categories, and the first is the one that makes the others hard to see.
- Unallocated spend
- Resources without a tag, project or owner. Nobody is accountable for them, so nobody turns them off.
- Idle resources
- Environments, volumes, snapshots and addresses left running after the work that needed them has ended.
- Overprovisioning
- Instances and databases sized for a peak that rarely comes, or for a load estimate that was never revisited.
- Commitment mismatch
- Reserved capacity or savings plans that no longer match the workload, paid for whether used or not.
- Reserved AI capacity
- GPU capacity held for training or inference that runs well below its utilization for most of the day.
Allocation before optimization
Discounts and rightsizing tools work only on spend that someone understands. The first step is allocation: every resource tagged to a product, team and environment, and every bill line assigned to an owner who sees it each month.
Discounts and rightsizing work only on spend that someone owns and understands.
The second step is a unit cost: cloud spend per transaction, per customer or per model call, tracked over time. A rising bill with a falling unit cost is growth. A rising unit cost is a problem, whatever the total.
Five tests
- Allocation coverageThe share of spend with an owner and a product is measured, and is above 90 percent.
- A unit cost metricEach major workload has a cost per unit of business output, reviewed monthly.
- Rightsizing on evidenceInstance changes follow measured utilization over a full business cycle, not one week.
- Commitments sized to the baselineReserved capacity covers the steady load only; peaks are paid on demand.
- AI capacity budgeted by use caseGPU and model spend is attributed to the workflow it serves, with a target cost per completed task.
References
- 1.Gartner. “Gartner Forecasts Worldwide Public Cloud End-User Spending to Total $723 Billion in 2025.” Press release, 19 November 2024. gartner.com
- 2.Flexera. “New Flexera Report Finds that 84% of Organizations Struggle to Manage Cloud Spend.” Press release, 19 March 2025. flexera.com
- 3.Flexera. “Flexera Finds Cloud Value is Rising While AI Waste Grows” (2026 State of the Cloud Report). Press release, 18 March 2026. flexera.com
