Cloud users estimated that 29% of their infrastructure and platform spending was wasted in Flexera’s 2026 State of the Cloud survey. The finding illustrates a cost-control problem: buying computing capacity on demand can remove a large upfront hardware purchase, while leaving businesses paying for resources they do not need.
The survey covered 753 technical professionals and executive leaders worldwide. Its waste figure comes from respondents’ estimates, not an audit of every resource or an industry-wide measurement. Flexera sells technology-spending management products.
The report, released in March, found that estimated waste had increased after several years of decline. It also identified forecasting difficulties as companies expanded their use of artificial intelligence.
A growing bill can reflect successful products serving more customers. The financial problem is separating that growth from idle capacity, inefficient applications and purchases nobody has reviewed.
What is cloud computing?
Cloud computing provides access over a network to shared computing resources, including servers, storage and applications. The National Institute of Standards and Technology identifies on-demand access, rapid expansion and contraction, and measured usage among its defining characteristics.
For a business using a public cloud provider, this can mean renting computing capacity without buying and operating the underlying equipment. A retailer can increase capacity during a sales event and reduce it afterward.
That flexibility changes the purchasing process. Developers can obtain resources quickly, and applications can add capacity automatically. Spending decisions become part of daily operations, instead of being concentrated in occasional hardware orders.
Resources are easy to add and easy to overlook
A test environment can keep running after an experiment ends. An application can retain enough capacity for its busiest period even when demand falls. Backups and stored data can accumulate long after the team that created them has moved on.
Google Cloud’s cost-optimization guidance advises customers to match resources to workload requirements and consumption patterns. Allocating more capacity than an application needs, known as overprovisioning, can increase costs without improving its performance.
The invoice also covers more than computing time. Storage, managed databases, monitoring and data transfers can have separate charges. AWS, for example, publishes internet data-transfer allowances and charges alongside its computing prices. How an application moves and stores information affects its bill.
The wider subscription problem is familiar. As we reported in our coverage of unused software, finance, technology and procurement teams can each hold different parts of the record. An invoice alone cannot show whether the underlying service is needed.
AI puts forecasting and accountability under pressure
Flexera found that 58% of respondents were using generative AI public-cloud services, up from 50% in the previous survey. Unpredictable usage was among the difficulties reported in managing AI workloads.
AI costs can change with the model selected, the amount of information processed and the number of requests. A tool that attracts more users can also create more computing demand, making the relationship between adoption and cost part of the product decision.
FinOps brings engineering, finance and business teams together to manage these decisions. The FinOps Foundation’s framework focuses on financial accountability and the business value obtained from technology spending.
That involves assigning costs to the teams or products generating them, reviewing unused resources and forecasting demand before approving new services. Measuring cost per customer, transaction or completed task can help explain whether a rising bill is funding productive growth.
The Foundation’s 2026 State of FinOps report also shows practitioners extending their remit into software subscriptions, licensing, private cloud and data centers. Its findings describe the FinOps community surveyed, rather than adoption across all businesses.
Moving workloads back has its own costs
Some companies have moved applications from public clouds to their own hardware, a process often called cloud repatriation.
In an October 2024 account, 37signals co-owner and Chief Technology Officer David Heinemeier Hansson said moving seven applications out of AWS had reduced its annual cloud bill from a $3.2 million run rate to $1.3 million. The remaining spending was on AWS storage.
He reported about $700,000 in new computing hardware and said the company had fitted it into existing data-center racks and power capacity. His projection of more than $10 million in five-year savings included a further planned storage migration. It was a company forecast, not a completed five-year result.
Another business would have to account for its own hardware, facilities, maintenance, staffing and migration costs. Existing contracts can also delay savings, as they did at 37signals.
Flexera found that 73% of respondents used hybrid cloud, combining public and private cloud environments. For companies reviewing where an application should run, the comparison needs to cover its expected demand, reliability requirements and total operating cost over time. A lower provider invoice is only one part of that calculation.
