Cloud cost drift, the top challenge for CIOs
The cloud fulfilled its promise of agility, yet at a price that is often unforeseen. According to Flexera’s State of the Cloud 2025 report, cloud cost management is the number one challenge facing organizations for the third year running – security having occupied the top spot in previous years – cited by roughly 84% of respondents.
The figures speak for themselves. Companies waste on average 27% of their cloud spend due to oversized instances and neglected resources. And cloud budgets exceed their forecasts by an average of 17%. In other words, nearly a third of the bill goes to pure waste, and the final envelope regularly escapes budgetary control.
This drift stems from the very nature of the cloud. Its self-service model lets any team provision a resource in a few clicks – but also forget it. Unattached storage volumes, idle virtual machines left running, test environments never shut down: the public cloud accumulates resources that no longer serve a purpose but continue to incur charges.
The contrast with the old model is striking. In the era of physical servers, every expense passed through a validated purchase, planned, amortized – a slow but controlled process. The cloud inverted this logic: spending became instant, continuous, and decentralized, escaping traditional budgeting loops. Finance discovers the invoice after the fact, without a direct lever on the decisions that generated it. It is precisely this gap that FinOps aims to bridge.
Why the problem is worsening
Several factors amplify the difficulty in 2025-2026. The first is the generalization of multi-cloud: most large organizations operate across multiple providers, which fragments visibility into spending and makes consolidation arduous.
The second is the growing abstraction. With Kubernetes and containers, costs are no longer tied to identifiable machines but to shared workloads, far harder to attribute to a team or project. The third, most recent factor, is the explosion of AI/ML workloads: GPU instances, essential for training and inference of models, can push bills to unprecedented speeds.
Finally, the pay-as-you-go pricing model becomes de-empowering when not properly bounded. The on-demand rate is still the most expensive: without optimization, a company pays a premium for flexibility it does not always exploit.
An organizational challenge adds to this: according to an Apptio study, 55% of executives report lacking the information necessary to properly evaluate their technology spend. The problem isn’t merely drifting costs, but opacity: decision-makers operate blindfolded, unable to link an expense to the value it yields. This opacity is as costly as the waste itself, because it prevents any rational decision.
FinOps: a cultural fix before a technical one
FinOps (a contraction of Finance and Operations) is a discipline aimed at mastering cloud costs by bringing together finance, engineering, and business around a shared responsibility. Its aim is not to cut spending, but to maximize value: spend better rather than spend less.
The FinOps Foundation, the leading organization that structures the discipline, summarizes the approach in three phases: Inform (visibility into who spends what and why), Optimize (reducing waste and optimizing pricing), and Exploit (governance and automation to sustain savings). It’s an iterative and ongoing process, not a one-off audit.
The cultural dimension is essential. FinOps makes technical teams accountable for the cost of their choices, without turning them into bookkeepers. The goal is to make spend visible and understandable, so everyone can arbitrate intelligently between performance, speed, and cost. Results can be spectacular: several FinOps case studies report cloud cost reductions of around 20 to 30% within a few months.
Where to start
Embarking on a FinOps journey does not require a large program. A few initial actions, with strong symbolic and tangible impact:
- Establish visibility: break down costs by service, team, environment, and project – you only manage what you measure.
- Eliminate obvious waste: remove unused resources (orphaned volumes, forgotten machines), the first symbolic FinOps action.
- Rightsize capacity: adjust oversized resources to match actual needs (rightsizing), the lever with the strongest impact for most organizations.
- Lock in commitments: for predictable workloads, reserved instances or savings plans markedly reduce on-demand pricing.
The challenge for a leader isn’t to turn engineers into managers, but to recognize that cost has become a dimension of architecture, on par with performance and security. In a context where AI workloads are driving bill spikes, ignoring FinOps means letting a growing portion of the IT budget slip away. Conversely, embedding it as a culture – not as policing – turns cost control into a durable advantage: more value per euro spent, and the means freed for innovation.
The moment is all the more opportune as generative AI workloads shift the scale of the problem. An forgotten GPU instance or an overprovisioned model can cost in days what a traditional infrastructure would in a month. Organizations that implement FinOps before aggressively scaling their AI use will approach this wave with guardrails; others risk discovering the bill too late. Plan ahead rather than suffer: that is the essence of starting the journey now, even at a small scale.
This content is published by Mentioned
