Why 28% of Your Cloud Budget Is Wasted and how to fix it

The cloud is supposed to reduce costs. In reality, companies spend an average of 28% of their cloud budget on resources they’ve provisioned but never actually used. This isn’t a technical failure, it’s a structural problem that can be specifically addressed with the right measures.

Why the numbers don’t add up

According to the Flexera State of the Cloud Report 2024, 59% of surveyed companies spent more on cloud services last year than planned. Gartner predicts that by 2025, more than 80% of companies will leave a significant portion of their cloud investments unused, primarily due to a lack of pricing transparency.
The real reason lies not in the technology itself, but in a fundamental systemic shift: cloud computing has completely changed the rules of IT cost control. Many organizations have underestimated this consequence.

So the question is not whether to use the cloud, but how to use the cloud.
Companies that seriously invest in cost management achieve the promised efficiency.

The Structural Core of the Problem

CapEx Became OpEx and that Changes Everything

In traditional IT environments, infrastructure costs were capital-intensive and one-time investments with long depreciation periods and clear responsibilities. The budget served as a disciplining factor because the consequences of a decision were immediately felt: You bought a server, and the amount was missing from the budget.
In the cloud, costs are incurred continuously, variably, and decentrally. The decision is made today and the bill arrives in four weeks. By then, the architecture is set and difficult to change. Traditional budgeting cycles are simply not designed for this model.
Whoever provision a cloud resource, usually a developer or an operations team, bears no immediate financial responsibility. The budget is managed centrally, the bill arrives at the end of the month or year, and it is rarely obvious which technical decision led to which financial outcome. This is not a management failure; it is a structural consequence of the model.

Decentralized Provisioning without Governance

In organizations with multiple teams across various cloud platforms, a situation quickly arises where no one has a complete overview: What resources are available? Who owns them? Are they still needed? Without systematic tagging and centralized inventory management, a significant portion of all resources cannot be assigned to any cost center, project, or team.

The four largest cost generators, explained in detail

1. Unproductive resource usage in non-production setups

By far the most prevalent and most often overlooked type of inefficiency. Testing, development, and staging environments usually operate during office hours or, to be precise, 30% of the total 168 hours in a week. The other 70% generate costs without any actual work being done.
As far as the technical fix goes, the concept of instance scheduling is pretty much standard. Any major cloud vendor will offer tools to that effect. But there are three key reasons why this particular strategy falls through: lack of central inventory management (you simply don’t have an overview of your non-production resources), opposition from developers, and lack of cost consciousness within teams.

Potential Cost Savings
Automatic instance scheduling in non-production could slash expenses by up to 72%.

2. Overprovisioning and Large Instance Sizes

The concept of provisioning to the highest level possible was valid in the days of physical servers. However, it is not cost-efficient in the cloud environment. Most groups continue using oversized instances simply because they were previously used or provision too much capacity just in case. Thus, we observe instances running on maybe less than 10% of their CPUs.
Rightsizing refers to a comprehensive assessment that considers CPU and memory usage for at least 14 days, alongside network traffic and storage usage. Also, there is an option to take advantage of new generations of instances which have more computational capacity for the same price or cheaper.

3. Orphaned Resources

Unmounted volumes, stale snapshots, and forgotten load balancers, none of these triggers alerts, none impacts the services provided and none can be seen in any monitoring system dashboard. However, the cost must still be paid; for instance, a 500 GB EBS volume from AWS (eu-central-1) costs roughly 90 EUR per month, irrespective of whether it’s mounted or not.
Here, the key resides in automation of scanning procedures to detect orphans as well as processes that help in their cleanup. However, all of this cannot be achieved without first establishing an inventory, which brings us to the core issue of tagging.

4. Suboptimal Payment Structures

Unused commitments and suboptimal license utilization are less obvious but significant. If you purchase Reserved Instances or Savings Plans and do not fully utilize the capacity, you still have to pay. Conversely, large workloads without any commitment strategy miss out on the substantial discount compared to on-demand prices.
 License optimization offers another, often untapped opportunity: The Azure Hybrid Benefit allows you to transfer existing on-premises Windows licenses to Azure and use them there at up to 85% lower cost. Oracle offers a comparable option in OCI with its BYOL program.

What to do now: The first step

Getting started with systematic cloud cost management doesn’t have to be complex. Three measures deliver immediate results:

• Create a complete inventory of all cloud resources across platforms, with a focus on non-production environments
• Conduct an initial joint cost review with Finance, IT, and Business not to assign blame, but to understand the problem
• Enable instance scheduling for non-production environments. This delivers immediate savings and is technically straightforward
  

The most critical step is cultural rather than technical. It involves the conversation between the people who provide the clouds and the ones paying for them. That is precisely what FinOps starts.