For five straight years, the cloud waste story had a reassuring shape. Organizations built FinOps teams, adopted tagging discipline, negotiated commitments, and the share of cloud spending that produced nothing slowly ticked down. In 2026 that trend broke. Flexera's fifteenth annual State of the Cloud Report, based on a survey of more than 750 cloud decision-makers, found estimated wasted cloud spend rose to 29 percent of IaaS and PaaS budgets, the first increase in five years.¹ ²
The reversal is not a failure of FinOps discipline. It is the arrival of a workload class that existing cost frameworks were never designed to handle. AI moved from experiment to production faster than governance could follow, and it brought with it consumption patterns that resist every optimization technique the last decade of cloud financial management was built on.
The scale makes this expensive. Gartner forecasts public cloud services growth of 21.3 percent in 2026, with the market projected to reach $1.48 trillion by 2029.³ Waste at 29 percent of a rapidly expanding base is not a rounding error. For a mid-market organization spending $5 million annually on cloud infrastructure, it is roughly $1.45 million. For an enterprise at $50 million, it approaches $14.5 million. Those are dollars that could fund security modernization, product development, or margin.
At Accelerate Partners, we advise CFOs and CTOs across regulated industries where this pressure lands hardest. Budgets are constrained, compliance obligations are expanding, and boards are asking harder questions about AI returns. What we see consistently is that the organizations struggling most are not the ones lacking tools. They are the ones applying 2022 cost controls to 2026 workloads.
What Changed: AI Broke the Optimization Playbook
The FinOps Foundation's sixth annual State of FinOps survey, covering 1,192 practitioners representing more than $83 billion in annual cloud spend, captured the shift with a single number. In 2024, 31 percent of practitioners managed AI spend. In 2025, 63 percent. In 2026, 98 percent.⁴ ⁵ That is the fastest adoption curve in the survey's history, and it happened faster than most organizations could build governance around it.
The problem is structural. Traditional cloud cost optimization rests on three assumptions: that usage is predictable enough to forecast, that workloads are stable enough to commit to, and that utilization correlates with load in ways autoscaling can act on. AI training and inference workloads satisfy none of these conditions. They are bursty, they are driven by experimentation rather than steady demand, and GPU fleets provisioned statically for peak capacity sit idle between runs.
Practitioners are candid about the difficulty. The FinOps Foundation reports that AI cost management is now the single most sought-after skillset teams plan to add, ranking above tooling expertise and automation development.⁴ Many organizations also report being asked to self-fund AI investment through optimization savings elsewhere, which creates an uncomfortable dynamic: teams are squeezing efficiency from mature infrastructure to pay for workloads that resist efficiency measures.
There is a second, less obvious driver. Optimization has hit diminishing returns. The obvious waste categories, idle development environments, grossly oversized instances, orphaned storage volumes, have largely been addressed at organizations with functioning FinOps practices. What remains requires architectural expertise rather than cleanup, and the effort-to-savings ratio has deteriorated sharply. The easy wins are gone.
Where the Waste Actually Lives
Understanding waste categories matters because the remediation for each is different, and applying the wrong one wastes effort as surely as the underlying inefficiency wastes money.
Idle compute remains the largest single category. These are resources provisioned and billed but never meaningfully used: development environments running through weekends, virtual machines spun up for a one-week test and never terminated, load balancers serving applications that were decommissioned months ago. The Harness FinOps in Focus research found that 48 percent of developers do not track and shut down idle resources at all, and only 43 percent have access to real-time data on idle cloud resources in the first place.⁶ You cannot eliminate what you cannot see.
Overprovisioning is the second major category and the one most rooted in habit rather than technology. Capacity planning in traditional data centers meant buying hardware sized for anticipated peak load three to five years out, which necessarily meant buying too much. That instinct persists in environments where it no longer serves any purpose, because cloud platforms allow resources to be resized in minutes. The same Harness research found 61 percent of developers do not rightsize instances.⁶
Orphaned and untiered storage compounds quietly. Storage volumes persist by design, since automatic deletion would be a data-loss hazard, which means they require deliberate cleanup. The cost differential across tiers is dramatic and underexploited. In AWS US East, S3 Standard runs $0.023 per GB-month while S3 Glacier Deep Archive runs $0.00099 per GB-month.⁷ For a petabyte, that is roughly $23,500 per month versus roughly $1,000. Lifecycle policies that move data to appropriate tiers based on access patterns are among the highest-return, lowest-risk actions available. The caveat matters for regulated organizations: Deep Archive carries retrieval fees and a 180-day minimum storage duration, so archives you may need to produce for an audit on short notice belong in a different tier.⁷ ⁸
Unused commitments and guesswork purchasing deserve more attention than they typically get. When 55 percent of developers say purchasing commitments are ultimately based on guesswork, and 58 percent do not use reserved instances or savings plans at all, the result is a two-sided loss: money left on the table through uncommitted on-demand spend, and money wasted on commitments that do not match actual consumption.⁶
AI and GPU capacity is the fastest-growing category and the least mature. Statically provisioned accelerators are expensive per hour and frequently idle between training runs. This is where 2026 waste growth is concentrated.
The Organizational Problem Underneath the Technical One
Cloud waste is not primarily an engineering failure. It is a structure-and-incentive failure that shows up on an engineering bill.
The most consequential gap sits between the people who provision resources and the people who own the budget. Engineering teams are measured on reliability, velocity, and feature delivery. Provisioning generously is the rational response to those incentives. Finance teams own the budget but generally lack the technical context to identify waste or the standing to challenge an architecture decision. Harness found that 52 percent of engineering leaders say the disconnect between FinOps and development teams is directly producing wasted infrastructure spend.⁶
The traditional budgeting model made this worse by removing a control point. Capital expenditure approval created a natural checkpoint before spending occurred. The cloud's operational expenditure model removed it. By the time the bill arrives, the money is spent and cannot be recovered.
What is encouraging is that the reporting structure is shifting in a productive direction. In 2026, 78 percent of FinOps practices report into the CTO or CIO organization, up 18 percentage points, and practices with executive engagement show substantially greater influence over technology decisions: 53 percent versus 24 percent on cloud service selection, and 47 percent versus 16 percent on cloud provider selection.⁴ ⁵ FinOps is moving upstream, from explaining last month's bill to shaping next quarter's architecture.
FinOps in 2026 Is No Longer About Cloud Alone
The most significant development for CFOs is a scope change. The FinOps Foundation formally updated its mission from advancing the people who manage the value of cloud to advancing the people who manage the value of technology.⁴ That is not a marketing adjustment. It reflects what practitioners are already being asked to do.
In 2026, 90 percent of FinOps teams manage or plan to manage SaaS spending, up from 65 percent the prior year. Sixty-four percent manage software licensing, up from 49 percent. Fifty-seven percent manage private cloud, up from 39 percent. Forty-eight percent manage data center costs, and an emerging 28 percent are including labor.⁵ The discipline has become technology financial management.
For organizations in financial services, healthcare, and manufacturing, this is a meaningful reframing. The waste sitting in redundant SaaS subscriptions and overlapping security tooling is frequently larger and easier to capture than the remaining waste in a well-optimized IaaS estate. Our cyber FinOps and cost optimization work routinely surfaces overlapping endpoint, monitoring, and vulnerability management tools purchased by different teams over several budget cycles.
The governance infrastructure is catching up. Flexera found 71 percent of organizations now operate a Cloud Center of Excellence and 63 percent have an established FinOps team.¹ ² The dominant operating model is a small central team, cited by 60 percent of respondents, supported by federated champions embedded in business units rather than a large centralized headcount.⁹
What to Actually Do: A Sequenced Approach
The following sequence reflects what produces results in mid-market and enterprise environments, ordered by return relative to effort.
Start with visibility and allocation, not cuts. You cannot govern what you cannot attribute. Establish tagging that identifies business unit, application, environment, cost center, and owner, then enforce it through policy rather than goodwill. Nearly half of organizations, 49 percent, now track a unit metric to understand cost per service, up from 40 percent the prior year.¹⁰ Unit economics, cost per customer, per transaction, per model inference, is what turns a cloud bill into a business conversation.
Capture idle and orphaned resources next. This is the highest-certainty savings available, with essentially zero business impact. Define idle explicitly, for example under 5 percent CPU utilization for 30 consecutive days, automate discovery, and implement scheduled shutdown for non-production environments outside business hours. A development instance running continuously costs roughly 3.4 times what the same instance costs running only during a 50-hour work week.
Rightsize systematically, not occasionally. Cloud platforms expose CPU, memory, network, and disk metrics that support evidence-based decisions. Analyze over 30 to 90 days to account for variability, then establish a recurring review cycle. Rightsizing done once is a project. Rightsizing done quarterly is a capability.
Tier storage with automated lifecycle policies. Given the roughly 23-fold cost differential between S3 Standard and Deep Archive, and comparable spreads on Azure and Google Cloud, lifecycle automation delivers durable savings with no ongoing effort.⁷ Set retention and tiering policies that match your actual regulatory retention obligations rather than defaulting to indefinite Standard-tier storage.
Then address commitments. Analyze historical consumption to identify genuinely stable workloads, cover 60 to 80 percent of steady-state demand with commitments, and leave headroom on demand for growth and flexibility. Do this after rightsizing, not before, or you will commit to capacity you are about to eliminate.
Shift left on AI spend specifically. Pre-deployment cost estimation emerged as a top desired tooling capability in the 2026 survey, and shift-left is a leading practitioner priority.⁴ For AI workloads, that means model right-sizing by task, caching stable context rather than paying for it repeatedly, and full token attribution to cost centers before usage scales rather than after.
The Business Case
The economics remain compelling even as the easy wins disappear. Mature FinOps programs consistently deliver cost reductions in the 20 to 30 percent range, and organizations that implement FinOps best practices realize 27 to 40 percent cost savings compared to on-premises environments, according to the AWS Cloud Value Benchmark referenced in the Harness and AWS joint research.¹¹
For a mid-market organization spending $5 million annually, a 25 percent reduction is $1.25 million recovered against a FinOps investment that typically includes a cost management platform, one to two dedicated practitioners, and initial program design. Payback measured in months rather than years is the normal outcome, not the optimistic one.
The more important shift is what leading organizations now measure. Flexera found that value delivered to business units jumped 12 percentage points as a top metric while cost efficiency and cost avoidance declined.¹⁰ The question has moved from how much did we save to whether the spending is producing proportionate return. For private equity portfolio companies, where cloud efficiency shows up directly in EBITDA and exit multiples, that reframing is particularly consequential.
What This Means for 2027 Planning
Three things are worth carrying into your next budget cycle.
First, treat the 29 percent figure as a signal rather than a benchmark to accept. The increase is concentrated in AI workloads, which means organizations with disciplined AI governance can materially outperform it while organizations without governance will likely exceed it.
Second, extend your FinOps scope past IaaS. If your practice still stops at AWS, Azure, and Google Cloud bills, you are governing a shrinking share of your technology spend. SaaS, licensing, and AI services now carry comparable waste risk with far less mature controls.
Third, move cost conversations earlier in the lifecycle. Every dollar of waste prevented at the architecture stage is cheaper than a dollar recovered through remediation, and the organizations with executive-engaged FinOps practices are demonstrably influencing those earlier decisions.
The cloud waste problem in 2026 is not that organizations forgot how to optimize. It is that the workload mix changed underneath a set of practices that were working. The organizations that adapt their governance to match, rather than applying old controls harder, will fund their AI ambitions out of recovered waste. The ones that do not will fund them out of margin.
Works Cited
1. "Flexera Finds Cloud Value is Rising While AI Waste Grows." Flexera, March 18, 2026. https://www.flexera.com/about-us/press-center/flexera-finds-cloud-value-is-rising-while-ai-waste-grows
2. "Flexera 2026 State of the Cloud Report." Flexera, 2026. https://info.flexera.com/CM-REPORT-State-of-the-Cloud
3. "Forecast: Public Cloud Services, Worldwide, 2023-2029, 3Q25 Update." Gartner, September 26, 2025. https://www.gartner.com/en/documents/6996966
4. "State of FinOps 2026 Report." FinOps Foundation, 2026. https://data.finops.org/
5. "State of FinOps Survey: AI Value and Skills Top Priorities as FinOps Matures Across Technology Value." The Linux Foundation, February 19, 2026. https://www.linuxfoundation.org/press/state-of-finops-survey-ai-value-and-skills-top-priorities-as-finops-matures-across-technology-value-98-manage-ai-90-saas-64-licensing-48-data-center-1
6. "$44.5 Billion in Infrastructure Cloud Waste Projected for 2025 Due to FinOps and Developer Disconnect." Harness, February 26, 2025. https://www.harness.io/press-and-news/finops-in-focus-report-2025
7. "Amazon S3 Pricing." Amazon Web Services. https://aws.amazon.com/s3/pricing/
8. "Managing the lifecycle of objects." AWS Documentation. https://docs.aws.amazon.com/AmazonS3/latest/userguide/object-lifecycle-mgmt.html
9. "State of FinOps Data Library." FinOps Foundation, 2026. https://data.finops.org/library/
10. "New Flexera Report Finds that 84% of Organizations Struggle to Manage Cloud Spend." Flexera, March 19, 2025. https://www.flexera.com/about-us/press-center/new-flexera-report-finds-84-percent-of-organizations-struggle-to-manage-cloud-spend
11. "FinOps in Focus." Harness. https://www.harness.io/finops-in-focus
12. "What is FinOps?" FinOps Foundation. https://www.finops.org/introduction/what-is-finops/
13. "FinOps Framework." FinOps Foundation. https://www.finops.org/framework/
14. "Cost Optimization Pillar." AWS Well-Architected Framework. https://docs.aws.amazon.com/wellarchitected/latest/cost-optimization-pillar/welcome.html
15. "Amazon EBS Pricing." Amazon Web Services. https://aws.amazon.com/ebs/pricing/
16. "AWS Budgets." Amazon Web Services. https://aws.amazon.com/aws-cost-management/aws-budgets/
17. "Instance Scheduler on AWS." Amazon Web Services. https://aws.amazon.com/solutions/implementations/instance-scheduler-on-aws/
18. "Reduce service costs by using Azure Advisor." Microsoft Learn. https://learn.microsoft.com/en-us/azure/advisor/advisor-cost-recommendations
19. "Create, edit, or delete budgets and budget alerts." Google Cloud Documentation. https://cloud.google.com/billing/docs/how-to/budgets
20. "Tagging Best Practices." AWS Whitepaper. https://docs.aws.amazon.com/whitepapers/latest/tagging-best-practices/tagging-best-practices.html
21. "FinOps Certified Practitioner." The Linux Foundation. https://training.linuxfoundation.org/certification/finops-certified-practitioner/
22. "Amazon EC2 Spot Instances." Amazon Web Services. https://aws.amazon.com/ec2/spot/
