The cloud was supposed to transform infrastructure from a large, inflexible investment into a flexible cost tailored to the company’s current needs. In many organisations, however, this flexibility works only one way: new resources are deployed in a matter of minutes, whilst their usefulness, ownership and impact on the bottom line are only analysed once the invoice has been received.
This is why the price of a single service rarely turns out to be the problem. The greatest costs arise from hundreds of small decisions that nobody subsequently checks. An oversized virtual machine, a test environment left running, an extra copy of data or an unused cluster do not look like a major problem on their own. However, on the scale of a large-scale environment, they create a constant stream of expenditure that grows automatically until someone consciously stops it.
According to a Flexer study, organisations estimate that they waste around 29 per cent of their expenditure on infrastructure and cloud platforms. The figure itself is not the most important thing. What is more significant is that it has risen after several years of improvement, despite the development of reporting tools, cost recommendations and automated detection of inefficiencies. Companies no longer lack data on expenditure. What is lacking is a mechanism that turns this data into accountability and decisions.
Cost visibility does not mean control
A dashboard can show that the bill has risen by 20 per cent, but it does not answer the most important question: who can do something about it? The finance department sees the invoice, the infrastructure team sees the resources, and the product team understands their significance for the user. Without bringing these three perspectives together, the company monitors costs but does not manage them.
The key problem, therefore, is not the lack of a report, but the lack of ownership. If an expense has not been assigned to a specific product, application, environment or team, it is difficult to determine whether it is still justified. A resource may be unused, but it may just as well be responsible for system reliability, a backup, or an important process carried out once a month. The algorithm alone cannot determine this.
As a result, optimisation recommendations end up on a to-do list that nobody treats as a priority. Potential savings compete with product development, security, system migration and ongoing outages. If responsibility for costs remains fragmented, the lack of a decision becomes the default decision, and the company continues to pay.
Cloud management begins not with the purchase of yet another tool, but with identifying who is responsible for the economic impact of usage. Every significant cost needs an owner who understands both the technical significance of the resource and the value of the process that the resource supports.
The bill is the result of earlier decisions
Cloud costs are often treated as an operational problem that only becomes apparent at the end of the month. Yet a significant portion of the future invoice is determined as early as the system design stage.
The choice of data storage method affects the cost of processing and transferring data. A fault-tolerant architecture increases the number of resources that need to be maintained. Managed services reduce the team’s workload but may increase the unit cost. Automatic scaling mechanisms help to handle peak demand, but if poorly configured, they increase consumption faster than revenue.
In this sense, the invoice is not the start of the problem, but its final stage. By the time costs become visible in a financial report, the company often already has an application designed around specific services, integrations and data flows. Changing such an architecture costs more than assessing its economics at an earlier stage.
That is why cost management is shifting to the design and implementation phase. Technology teams are increasingly analysing the projected cost before the environment goes live, rather than only after several months of operation. The value of this change lies not solely in reducing the bill. The company can see at an earlier stage whether the technology model will scale alongside the product, or whether it will become a barrier to its profitability.
AI accelerates cost growth and makes it harder to assess
AI workloads exacerbate all the problems previously present in the cloud. Usage is harder to predict; it depends on user behaviour, the chosen model, the length of the context, the number of queries and how the agents operate. The cost of a single process can fluctuate without any changes to the application itself.
This creates a new type of risk. The system may function correctly whilst simultaneously becoming less and less cost-effective. A higher number of users, more complex queries or agents automatically performing subsequent operations drive up costs faster than a traditional budget can account for.
Under such circumstances, a monthly invoice provides information too late. Effective management requires knowledge of how much it costs to handle a single case, generate a document, analyse a request or have an agent carry out a task. Without this, the company knows the cost of the infrastructure but not the cost of the service it provides.
AI therefore clearly demonstrates why the total bill is becoming an increasingly less useful indicator. Two teams may spend similar amounts, but one uses that expenditure to generate thousands of paid transactions, whilst the other is conducting an experiment with no clear business outcome. From a financial perspective, these are not comparable situations.
What matters is the unit cost, not the invoice amount
Rising cloud expenditure does not always indicate a problem. If a company is acquiring more customers, processing more transactions and handling a greater volume of data, a higher invoice may be a natural consequence of growth.
What matters is the pace of this growth. If the cost of serving a single customer falls, the environment is scaling effectively. If the number of users rises by 20 per cent, whilst infrastructure costs rise by 50 per cent, the technology model begins to undermine the product’s economics.
That is why mature organisations are moving away from controlling the total budget towards measuring unit costs. They analyse the cost per transaction, per active user, per order, per AI query or per unit of revenue. Such metrics link technological decisions to business outcomes and enable an assessment of whether increased usage is creating value.
This is also an important shift for boards of directors. The mere fact that cloud expenditure has risen does not in itself indicate whether the situation requires intervention. Only by comparing cost with volume and product performance can it be determined whether the company is investing in scaling or funding inefficiency.
FinOps goes beyond cost-cutting programmes
Cloud economics management has long been associated with identifying underutilised resources, renegotiating contracts and purchasing commitments. This model is becoming insufficient, as optimising price alone does not solve the problem of uncontrolled consumption.
Discounts may reduce the unit cost, but they do not eliminate resources that do not generate value. Furthermore, long-term purchasing commitments can entrench flawed assumptions about demand. The company then pays less per unit, but still buys too many of them.
A strategy based solely on shifting the burden from the cloud to an in-house data centre is equally limited. It may improve the economics of stable and predictable systems, but it does not remove the need to manage capacity, licences, energy, hardware and staff. Some of the costs simply become less visible, as they do not appear on a single monthly invoice.
FinOps therefore increasingly refers to managing the value of the entire technology portfolio: public cloud, SaaS, AI, licences and on-premises infrastructure. The aim is not the lowest possible cost, but the best balance between expenditure, speed of operation, resilience, risk and product performance.
Anonymous consumption is the most expensive
The scale of the cloud means that manual control is not enough. Automatic deactivation of test environments, mandatory tagging, anomaly detection and limits on experiments reduce the number of issues before they appear on the invoice. However, automation only works when it is based on clearly defined accountability rules.
The greatest threat is not a single expensive resource. It is anonymous consumption: expenditure that cannot be linked to an owner, a product or a measurable business outcome. Such a cost can remain in the environment for months, as everyone sees only a fragment of it.
