Cloud: Managing Costs Across Compute, Storage, and Data Pipelines

In Germany, operating costs for cloud solutions rose at 64 percent of companies last year, sharply at 19 percent. For 2026, 14 percent expect sharply rising and 40 percent slightly rising expenses, according to the Bitkom Cloud Report 2026. At the same time, 67 percent of companies use the cloud with the goal of reducing costs.

In distributed architectures, costs arise in many places at once: in compute instances with unused capacity, in storage without deletion periods and in data pipelines that move more data than necessary on every run. Organizations that want to reduce cloud costs therefore need cost control that covers all three areas and is shared by engineering, finance and business units.

What Cloud Cost Control Covers

Cloud cost control connects financial management with the operation of variable cloud resources. Traditional IT cost management plans budgets on an annual cycle and assigns costs to fixed cost centers. In the cloud, costs arise continuously, triggered by deployments, scaling rules and data volumes. Cost accountability therefore sits where resources are created: with the teams that provision infrastructure and run applications.

Cost control follows a recurring cycle of three steps. The first step creates transparency about which costs belong to which team, product or customer. In the second, resources are adjusted and pricing models selected. In the third, the results are embedded in operations as rules, budgets and automation. Dashboards and cost management tools mainly cover the first step. The effect on the bill comes from the two that follow.

Compute: Aligning Capacity with Actual Demand

Compute costs follow three decisions: instance size, runtime and pricing model. Rightsizing compares actual utilization over several weeks with provisioned capacity and adjusts instances step by step. Development and test environments can be shut down automatically outside working hours. For baseline load, reservations or savings plans lower unit costs, while spot instances suit interruptible batch jobs.

The trade-off lies in commitment. Reservations run for fixed terms and tie usage to a provider, a region or an instance family. Organizations that plan to relocate workloads later or move them into a sovereign environment align the term and scope of commitments with their own platform strategy.

In Kubernetes environments, requests and limits determine how densely pods are packed onto nodes. Oversized requests reserve capacity that is paid for and not used. Autoscalers at pod and cluster level adjust capacity to load. Cost allocation per namespace shows which team consumes which resources.

Storage: Placing Data by Access and Lifecycle

Storage costs grow with every file nobody deletes: snapshots, logs, orphaned volumes and backups without an expiry date. A central lever is assigning data to storage classes by access frequency. Frequently used data stays in fast classes, while rarely used data moves automatically into cheaper classes or the archive via lifecycle policies. Deletion periods follow legal and internal retention requirements.

Archive classes are cheap to store and more expensive and slower to retrieve. Data transfers out of the cloud, known as egress costs, occur during migrations and provider changes. This is where cost and sovereignty questions overlap: 59 percent of companies name lock-in effects such as difficult data exports as the biggest obstacle to switching providers. Export paths and data formats therefore belong in the storage architecture before large volumes of data accumulate.

Data Pipelines: Making Costs Visible per Run

Data pipelines generate costs in compute, storage and network at the same time. This makes them hard to allocate in per-resource cost reports. A metric per pipeline run or per data product is more meaningful.

Incremental processing loads only changed records and replaces full reloads. Partitioning and columnar formats such as Parquet reduce the volume of scanned data, which directly affects costs under query-based pricing models. Intermediate results receive an expiry date, and schedules follow the actual needs of downstream consumers. Data contracts define which data is delivered at what frequency and quality, limiting unnecessary processing.

AI workloads increase the pressure on these pipelines. 42 percent of companies use AI services from the cloud today, and 69 percent plan to do so within five years. Training, feature computation and inference draw on the same data flows and increase their volume when operated without joint planning.

Cloud Cost Control Operating Models Compared

How cost control is anchored in the organization depends on size, cloud maturity and existing skills. Three models can be distinguished.

The models can be combined. A central team sets standards, a partner provides analysis and automation, and product teams carry cost accountability for their workloads. 38 percent of companies rely on multi-cloud. In such environments, cost control requires consistent cost data across all providers.

How to Reduce Cloud Costs: Three Principles

01 Allocate costs before optimizing.

Mandatory tags for team, product, environment and cost center form the basis of every analysis. Showback makes costs visible to teams, chargeback bills them internally. The path usually starts with showback, because teams first need to learn how to read their costs.

02 Measure costs per unit of output.

Total costs rise as a business grows. Costs per transaction, per customer or per pipeline run are more meaningful. Falling unit costs alongside rising total costs indicate efficient scaling.

03 Automate rules.

Budgets with thresholds, policies as code for permitted instance types, automatic shutdown schedules and lifecycle policies work without manual intervention. Cloud automation embeds cost rules directly in infrastructure-as-code templates and deployment pipelines.

Cost Control as Part of Platform Architecture

Cloud cost control delivers results when it is embedded in architecture, operations and data platforms. The starting point is cost transparency across compute, storage and data pipelines. Tagging standards, metrics per unit of output and automated rules build on this and are integrated into existing platforms and deployment processes. Which operating model fits depends on cloud maturity, team structure and sovereignty requirements. Further levers of cloud cost optimization build on these foundations.