The cloud exceeds 875 billion euros, and AI drives useless spending

The cloud billing has entered a much more uncomfortable phase for companies. It’s no longer just about provisioning more capacity to scale, deploying products faster, or accelerating AI projects. The problem is that an increasing portion of the spending remains before generating value: idle machines, overprovisioned clusters, forgotten storage, outdated snapshots, poorly designed traffic, unused licenses, and GPUs reserved for AI workloads that aren’t well utilized.

The market is now measured in figures hard to digest. Gartner estimated that global end-user spending on public cloud would reach $723.4 billion in 2025, roughly €614.9 billion at the BCE reference exchange rate of May 11, 2026. Forrester projects the market to exceed $1.03 trillion in 2026, approximately €875.5 billion. The conversion is not minor: for European budgets, cloud is no longer just an IT line item but a budget category comparable to entire industrial sectors.

Artificial intelligence is accelerating this pressure. Flexera reports that wasted cloud spend has risen to 29% in 2026, after five years of decline, due to the complexity of AI workloads and the proliferation of new IaaS and PaaS offerings. Applying that percentage to Forrester’s forecast for 2026, the potential waste approaches €254 billion. This is not marginal inefficiency; it’s a structural hole.

From Cloud Growth to Financial Chaos

For years, cloud promised a very attractive proposition: pay only for what you use. In theory, this model was more flexible than buying servers, overprovisioning own data centers, and waiting years to amortize hardware. In practice, many companies have transferred the same bad habits to the cloud, but with a bill that updates every month.

The mechanism is familiar. One team provisions for peak levels, but the system often runs at average values. Another team creates a testing environment and never shuts it down. A project is canceled but leaves orphaned disks, snapshots, or databases. Kubernetes reserves more CPU and memory than it actually consumes. An AI model starts as a pilot but ends up generating ongoing costs in tokens, GPUs, and storage without clear attribution to product, client, or outcome.

Data collected by Zop from Flexera, Gartner, FinOps Foundation, Datadog, and HashiCorp all point in the same direction: cloud spending grows faster than management discipline. The 2025 market is expected to be around €614.9 billion, and the layer of wasted spend could range between €154.7 billion (based on Zop’s cited figure of $182 billion) and about €166 billion if the 27% percentage from Gartner’s estimate is directly applied.

By 2026, the situation becomes more serious. On a forecast of €875.5 billion, a 29% waste would amount to roughly €254 billion. That’s paying for infrastructure that’s not used, capacity that remains unchecked, or resources not tied to any business metric.

AI Turns FinOps Into a Management Priority

Adoption of artificial intelligence has changed the conversation. Previously, FinOps focused mainly on instances, storage, traffic, usage commitments, non-production environments, and Kubernetes. Now, more challenging variables come into play: cost per token, inference, training, GPUs, multimodal models, agents making multiple calls, long prompts, and managed AI services growing without the same level of oversight as traditional infrastructure.

The FinOps Foundation already detects this shift: 63% of FinOps professionals manage AI-related expenses, compared to 31% the previous year. It’s a significant advance, but it also reveals the opposite problem: many organizations use GenAI without a complete view of the actual costs associated with their models, agents, and GPU workloads.

AI has a dangerous feature for budgets: it scales very quickly when working and just as fast when it’s not. A team can launch dozens of experiments, index documents, run agents over repositories, create sandboxes, test different models, or extend context windows without the cost being immediately apparent. The bill arrives later.

In traditional infrastructure, an underutilized virtual machine was already a problem. In AI, a costly underused GPU can be an even bigger issue. According to analysis shared by Zop, the average utilization of cloud GPUs is around 23%, leaving enormous capacity paid for but not used. One cited example involves an NVIDIA H100 instance on AWS costing between $25,000 and $98,000 per year, roughly €21,250 to €83,300. If actual usage is low, the effective hourly cost skyrockets.

Where the Money Slips Away

Cloud waste isn’t just one thing. It’s a combination of bad technical habits, lack of governance, and unclear accountability. Using a potential waste estimate of €254 billion in 2026, the approximate breakdown by category provides a pretty clear picture:

Idle computation remains the biggest leak. Datadog has noted in cloud cost analyses that many organizations keep low-utilization instances running for long periods, making it difficult to identify overprovisioned or obsolete resources in ephemeral and complex environments. Similar issues exist with containers: clusters are designed to prevent shortages, but rarely are they reviewed as thoroughly as they are created.

Non-production environments are another easily fixable leak. Development, QA, testing, staging, demos, and sandboxes are necessary, but they don’t need to be on 24/7 if used only 20, 30, or 40 hours a week. Many companies could quickly save by turning these off outside working hours, scheduling windows, creating ephemeral infrastructure, or destroying resources after a merge.

Here’s a key point: not all savings require long migrations or complex redesigns. Some are purely operational actions. Proper tagging, assigning owners, reviewing orphaned resources, scheduling shutdowns, resizing instances, leveraging committed use where demand is stable, automating anomaly alerts—these can all produce quick wins. And most importantly, cloud billing should no longer be treated solely as a finance issue.

Cheap Cloud Doesn’t Exist Without Discipline

FinOps isn’t about spending less by default. It’s about spending intentionally. A company may need to increase its cloud bill if it’s growing, serving more customers, or deploying revenue-generating AI products. The real issue arises when no one can explain which part of the spend is working and which is just turned on.

Maturity in this area remains limited. Many organizations have dashboards, but actions often lag. They see the bill, break it down by teams, and discuss it in meetings, but actual infrastructure changes tend to be manual, slow, or politically difficult. HashiCorp, in its cloud complexity report, highlights lack of visibility as a major barrier to managing modern infrastructure, especially in multicloud and hybrid environments.

AI will make this gap between visibility and action even more costly. A monthly spending dashboard is no longer enough when an agent can make multiple calls, or a team can reserve GPUs for days. Companies will need limits, per-model budgets, near-real-time alerts, automatic model selection based on cost and quality, caches, quantization, idle load shutdowns, and much finer attribution.

The CFO will also need to get involved earlier in the conversation. Cloud is no longer just a technical decision. In 2026, with a market exceeding €875 billion and a potential waste approaching €254 billion, cloud governance becomes a matter of margins, competitiveness, and strategy.

A clear takeaway for any tech media: AI isn’t just increasing cloud demand; it’s exposing its worst practices. Companies that already mismanaged virtual machines, storage, and Kubernetes will also struggle more with GPUs, tokens, and agents unless they change their processes. Cloud didn’t suddenly become expensive; what’s changing is that the bill now shows, more clearly than ever, how much it costs not to govern it.

Frequently Asked Questions

How much money will be wasted on cloud in 2026?
Applying Flexera’s estimated 29% waste to Forrester’s forecasted public cloud market of 2026, the unspent amount would be around €254 billion.

Why does AI increase unnecessary cloud spending?
Because it introduces more variable and costly workloads, such as GPUs, inference, tokens, agents, and long context windows. Without control, these demands grow faster than financial oversight can keep up.

What is FinOps and why is it more important now?
FinOps is a discipline combining engineering, finance, and business to provide visibility, assign accountability, and optimize cloud costs without stifling innovation. With AI, it becomes even more urgent.

Where are the quickest savings usually found?
In development, testing, and staging environments running outside business hours; overprovisioned instances; orphaned resources; unused storage; and poorly managed usage commitments.

Source: Noticias inteligencia artificial