The essentials at a glance

  • Expect about one-quarter to one-third to be wasted: Gartner estimates 29% in 2026.

  • The biggest savings come from usage (rightsizing, cleanup) and procurement (commitments). Cleanup can be done within a few weeks, rightsizing and commitments within two months.

  • Lasting results require ownership: teams that see their own costs and are held accountable for them.

  • AI costs will be the top priority for FinOps teams in 2026 (State of FinOps 2026). Start by gaining insight into each application before you begin optimizing.

1. Why are cloud costs rising?

Cloud costs are rising because whoever incurs the costs doesn’t see the bill. An engineer can purchase capacity with just a few clicks and never sees the bill. The bill ends up with the finance department or the IT manager, and they can’t tell which team, which application, or which forgotten test environment is driving the costs. Without clear agreements, the bill therefore rises faster than usage.

According to research, cost control has been the biggest challenge in the cloud for years. Gartner estimates that in 2026 an average of 29% of cloud budgets is wasted; that estimate is rising again, primarily due to AI workloads and new PaaS services.

The most common causes

  • Overprovisioned resources: servers, databases, and Kubernetes requests with much more capacity than they actually use.

  • Unused resources: forgotten test environments, unattached disks, old snapshots, unused IP addresses, and load balancers.

  • Test and development environments that run 24/7, even though they are only needed during business hours. About 70% of the hours in a week fall outside of business hours.

  • No procurement agreements or incorrect ones: commitments that are missing, have expired, or no longer align with actual usage.

  • Data traffic: egress, NAT gateways, and traffic between zones or regions.

  • The wrong storage class: data that is rarely read but is stored on expensive storage, and snapshots without a retention policy.

  • Logging and observability: Teams log everything and store it indefinitely.

  • Architectural choices that no longer fit, for example serverless where a container would be cheaper under constant load, or vice versa.

2. Where does your cloud budget go?

In most organizations, five categories account for the bulk of the expenses:

  • Compute: virtual machines, containers, and serverless.
  • Databases: managed databases, often generously sized and rarely decommissioned.
  • Storage: volumes, object storage, snapshots, and backups.
  • Network: data traffic to the Internet, between zones and regions, and NAT gateways.
  • Managed services: logging, monitoring, load balancers, and, increasingly, AI services.

Start with the five largest services and the five largest accounts or subscriptions. That's almost always where the biggest savings are to be found.

3. Ten cost-saving measures

OrderClean up and rightsize first, and only then buy commitments. If you buy commitments first, you lock yourself into capacity that you no longer need after rightsizing.A real-world exampleAt Colect, the database optimizations alone saved $6,759 per month right away. In total, the AWS bill fell by 43%.

Quick wins: immediate results, minimal effort

No. Measure Effort
1 Clean Up Unused Resources
Unallocated disks, old snapshots, unused IP addresses, and load balancers
Low
2 Automatically Shut Down Test and Development Environments
Outside of business hours
Low
3 Rightsize resources
Based on actual usage (at least two to four weeks of usage data)
Low-medium
4 Commitments for stable baseline consumption Low

Structural measures: for the long term

No. Measure Effort
5 Migrate storage to a lower-cost tier
With lifecycle and retention policies
Medium
6 Newer instance generations or ARM processors
Such as AWS Graviton, Azure Cobalt, Google Axion
Medium
7 Autoscaling
So that capacity scales with usage
Medium
8 Limiting Data Traffic
Zones, NAT, CDN, caching
Medium-high
9
Spot Capacity For interruptible workloads
Medium
10 Architecture Review
Serverless or containers, database design, managed or self-managed
High

Reducing cloud costs on AWS: databases, compute, and networking

In the AWS environments analyzed by SUE, 34.9% of waste is in databases, 23.4% in compute, and 18.9% in networking. These are therefore the three areas with the greatest potential for savings.

Databases (RDS and Aurora). Rightsize instances based on actual usage, consolidate databases where possible, choose between Aurora Serverless v2 and fixed capacity, and migrate storage from gp2 to gp3.

Compute (EC2, containers, and Lambda). First, rightsize using the recommendations from AWS Compute Optimizer. Then, cover your stable baseline usage with Compute Savings Plans (see Section 4) and use Graviton processors and Spot instances where workloads allow. Lambda isn’t always cheaper: with high request volumes and slow downstream services, you’re mainly paying for wait time; see “How Much More Expensive Is AWS Lambda?”

Network. Route traffic to S3 and DynamoDB via gateway endpoints instead of through a NAT gateway, limit traffic between availability zones, and use CloudFront for traffic to users. Also, clean up unused public IPv4 addresses. To learn how CloudFront reduces data traffic costs, read the Monaliro case study.

Reducing cloud costs on Azure: licensing, procurement, and rightsizing

On Azure, a large portion of the bill often comes from licenses: Windows Server and SQL Server may be included in the price of a VM or database.

Licenses. With Azure Hybrid Benefit, you can use existing Windows Server and SQL Server licenses with active Software Assurance in Azure. You then pay only the base rate for a Windows VM, which is the same as for Linux. Separate Dev/Test rates apply for testing and development, with a Visual Studio subscription.

Procurement. Buy reservations for VMs and databases that run continuously, and the Savings Plan for Compute for usage that shifts between VM series and regions (see Section 4).

Rightsize and shut down. Follow Azure Advisor's recommendations, automatically shut down test VMs outside of business hours, and choose Azure SQL Database serverless for databases that are frequently idle.

You can view costs per subscription and resource group in Azure Cost Management. The Cloud Cost Optimization Assessment shows you what savings are possible in your environment.

4. Procurement models and commitments

Commitments offer a discount in exchange for a one- or three-year commitment. They reduce your bill the fastest. At the same time, they are the only option that requires you to commit.

How to do it right:

  • Cover only the stable baseline usage. A common goal is to cover 70–80% of that; peaks and the remainder remain flexible.
  • Choose flexibility if your roadmap changes. If a migration or transition to ARM or Kubernetes is planned within three years, Compute Savings Plans or flexible CUDs are a safer choice than type-based reservations.
  • Spread out your commitments over time (laddering), so they don’t all expire at the same time and you can make adjustments each quarter.
  • Understand the risk. You can only cancel AWS Savings Plans within the first seven days and up to a limited amount. Microsoft currently does not charge cancellation fees for reservations, but reserves the right to charge 12%. You’ll still have to pay for any commitment you don’t use.
  • Measure the result using the effective savings rate (see Section 8).
Provider Model Maximum discount according to the provider Flexibility
AWS Compute Savings Plans Up to 66% Each instance family, region, Fargate, and Lambda
EC2 Instance Savings Plans / Reserved Instances Up to 72% Tied to an instance family and region
Spot Up to 90% AWS can reclaim capacity with two minutes' notice
Microsoft Azure Compute Savings Plan Up to 65% Across VM series and regions
Reserved VM Instances Up to 72% Tied to the VM series and region; exchanges are limited, and will no longer be available for purchases made on or after 1 February 2027
Spot VMs Up to 90% Azure can reclaim capacity with 30 seconds' notice
Google Cloud Resource-based committed use discounts Up to 55%, memory-optimized up to 70% Tied to machine type and region
Flexible (spend-based) committed use discounts 28% (1 year) to 46% (3 years) Across machine types and regions
Spot VMs Up to 91% Google can reclaim capacity with 30 seconds' notice

Source: Pricing pages and documentation from AWS, Microsoft, and Google, November 2026. Maximum discounts apply to three-year terms and specific configurations.

5. Data traffic and the EU Data Act

Data traffic is often the least understood cost item, because you pay for it by the gigabyte and it’s generated in many different places.

What helps:
Gateway endpoints for S3 and DynamoDB instead of routing traffic through NAT; limiting traffic between zones where availability permits; using a CDN and caching for traffic to users; compression; and processing data in the region where it is stored.

The EU Data Act.
As of 12 September 2025, cloud providers may only charge costs they actually incur when a customer switches to another provider or to an on-premises solution. Starting 12 January 2027, these switching charges will be prohibited, including egress fees for transferring your data. However, standard data traffic during use—such as traffic to your users or between regions—will still be subject to charges. AWS, Azure, and Google each offer their own procedure for waiving egress fees during a migration, with their own terms and deadlines. For the assessment in Section 13, this means that exit costs will decrease while ongoing data traffic costs will remain.

6. What is FinOps?


FinOps is an approach in which finance, engineering, procurement, and management work together to drive the value of technology. Those who incur costs are also the ones who see them, and decisions are based on the value they deliver. The FinOps Foundation (part of the Linux Foundation) manages the framework.

Three phases that keep recurring:

  1. Inform: Show who is spending what, and allocate costs to teams and products.
  2. Optimize: Save money through usage (rightsizing, cleanup, architecture) and through pricing (commitments, negotiation).
  3. Operate: Embed budgets, alerts, regular reviews, and ownership within the organization.

Since 2025, FinOps has extended beyond the public cloud. The framework now includes the following scopes: public cloud, SaaS, AI, licenses, and data centers. According to the State of FinOps 2026 (1,192 respondents, with a combined cloud budget of over $83 billion), 98% of FinOps teams manage AI costs, 90% manage SaaS, and 64% manage licenses. 78% report to the CTO or CIO. FinOps for AI is the top priority for 2026.

Maturity: Organizations grow from crawling to walking to running. Start with insight and ownership; perfection comes later.

Who does what in FinOps?

Role Responsibility
Engineering Technical decisions, implementing optimizations, factoring costs into the design
Finance Budget, forecasting, allocation, and chargeback
Product or Business Owner Balancing costs and value, unit economics
Procurement Commitments and Contracts with Providers and SaaS Vendors
IT Management Goals, Priorities, and the Mandate for FinOps
FinOps Manager Data, tools, reporting, and the frequency of reviews

7. Tagging, cost allocation, and FOCUS

Cost allocation starts with tagging: each resource is assigned tags such as team, application, environment, and cost center. Define required tags in policies and enforce them in code using AWS Tag Policies and Service Control Policies, Azure Policy, or Google Organization Policies, along with checks in your Terraform pipeline.

  • Showback: Teams see how much they cost, but they don't pay for them themselves. A good first step.
  • Chargeback: Costs are charged to the team’s budget. This fosters a stronger sense of ownership but requires reliable allocation.
  • Shared Costs (network, support, shared clusters) are allocated according to a fixed formula, such as based on usage or revenue.

FOCUS

The FinOps Open Cost and Usage Specification is an open standard for cost and usage data, currently at version 1.4. AWS, Microsoft, Google, and Oracle provide cost data in this format. European providers such as OVHcloud and STACKIT, as well as SaaS providers such as Snowflake and Databricks, are also on the FinOps Foundation’s list. This allows you to compare costs across multiple clouds and services without having to convert the data yourself.

8. Key metrics to steer by

Unit cost is the key metric you use when discussing this with management. Rising costs aren't a problem if the cost per customer is falling.

Key figure What it says Guideline (indicative)
Allocated costs Percentage of costs with an owner 90% or more
Commitment Coverage The portion of stable usage covered by a commitment 70–80%
Commitment utilization The portion of the purchased commitments that is actually used Almost 100%
Effective savings rate Discount realized on your total expenses compared to on-demand Higher is better; follow the trend
Waste Costs of Unused or Underutilized Resources Declining
Unit cost Cloud costs per customer, transaction, order, or user Declining as you grow
Deviation from the forecast How Accurately You Predict Your Costs As small as possible; set a goal each quarter

9. Costs in the design phase: shift-left

Costs that you avoid during the design phase won’t need to be cut later on. That’s why mature organizations shift their focus on cost awareness to the design phase:

  • Cost as a non-functional requirement in architectural decisions, alongside performance and security.
  • Cost estimates in the pipeline: Tools like Infracost show the monthly cost of every Terraform change.
  • Reference architectures with cost-effective standard choices, such as ARM, gateway endpoints, and lifecycle policies.
  • The "Cost Optimization" pillar of the AWS Well-Architected Framework—or the comparable guidelines from Azure and Google—as a standard benchmark.

10. Kubernetes costs

In Kubernetes, applications share the same nodes, so the cloud bill doesn't show which team or application is responsible for which costs. Utilization is also low. We often see an average CPU usage of 8% and memory usage of 20%.

What helps:

  • Allocate costs by namespace or label using OpenCost (open source, CNCF) or a commercial tool.
  • Realistic requests and limits based on measured usage, for example, using the Vertical Pod Autoscaler in recommendation mode. Overly generous requests are the biggest source of waste, because the scheduler reserves capacity that is never used.
  • Bin packing and node autoscaling using Karpenter or the Cluster Autoscaler, so that nodes run at full capacity and idle nodes are removed.
  • Spot nodes for stateless and interruptible workloads, with pod disruption budgets.
  • Fewer isolated clusters: every control plane and every set of system components costs money. See the guide on Kubernetes.

11. Costs of AI and GPUs

AI adds two new cost items: the use of language models (per token) and GPU capacity.

Reducing token costs:

  • Allocate costs by application and by team, using separate API keys or projects. Without allocation, no one knows how much each application costs.
  • Choose the smallest model that's good enough, and route simple queries to a less expensive model.
  • Use prompt caching for recurring contexts. With Anthropic's API, a cache read costs 10% of the normal input price; other providers offer similar discounts.
  • Use batch processing for tasks that don't need to be completed immediately: many providers offer a 50% discount.
  • Measure costs per outcome— for example, per completed customer interaction—and compare them to the value.

Reducing GPU costs:

  • Measure utilization per GPU and per workload. Without that data, you won't know if reserving resources is worth it.
  • Disable GPU capacity when it is not in use, and share GPUs across workloads whenever possible.
  • Only reserve capacity for inference that is demonstrably constant.
  • Consider an on-premises solution or a European provider for stable, large-scale AI workloads, especially if the data cannot leave the country.

12. Licenses and SaaS

Licenses can represent a larger expense than the underlying infrastructure. Windows Server and SQL Server are billed per vCPU in the cloud; with the Azure Hybrid Benefit or your own licenses (BYOL), you only pay for them once. Even smaller instances with the same performance provide immediate savings on per-core licensing. The same principle applies to SaaS as to the cloud: gain insight into usage per user, cancel unused licenses, and align contracts with actual usage. 90% of FinOps teams now include SaaS in their management.

13. Optimize, switch, or go back to on-premises?

High costs alone are rarely a good reason to switch cloud providers or return to an on-premises data center. A migration takes time and resources, and the underlying issues come along with it: if you oversize your resources in the cloud, you’ll do the same on-premises. Therefore, optimize first, and then compare.

For highly stable, predictable workloads, on-premises or a European cloud may indeed be less expensive, and sovereignty requirements can be the deciding factor.

Make that decision based on the total costs over five years

  • Hardware, data center, power, and cooling.

  • Staff for operations, 24/7 coverage, and security.

  • Licenses and support.

  • The cost of the migration itself.

  • The loss of flexibility and managed services.

  • Exit costs, which will be lower starting in 2027 under the EU Data Act.

14. Plan for the first 90 days

01

Days 1–30: insights and quick wins


  • Enable cost reporting by account, service, and team, preferably in FOCUS format; identify the top five cost items.
  • Clean up unused resources; shut down test and development environments outside of business hours.
  • Set up budgets and anomaly detection (AWS Cost Anomaly Detection, Azure Cost Management, Google Cloud Billing).

02

Days 31–60: rightsizing and procurement


  • Resize the largest resources and Kubernetes requests based on measured usage.
  • Only then should you assess and supplement the coverage with commitments.
  • Establish and enforce a tagging policy in code.

03

Days 61–90: embed and sustain


  • Showback per team: a monthly cost review with engineering and finance.
  • Define KPIs, including unit cost for one product.
  • Develop a roadmap for structural measures: architecture, ARM, data traffic, AI costs.

15. Frequently asked questions about cloud costs

All frequently asked questions
On AWS, most of the waste is found in databases, compute, and networking: in the environments analyzed by SUE, these account for more than three-quarters of the total. On Azure, significant savings can be found in licensing (Azure Hybrid Benefit), procurement (reservations and the Savings Plan for Compute), and rightsizing.

All three offer discounts in exchange for a one- or three-year commitment. A Reserved Instance on AWS or a reservation on Azure is tied to a specific resource type in a region, with discounts of up to 72% on AWS. A Compute Savings Plan on AWS, or the savings plan for compute on Azure, is a fixed hourly rate that applies to other instance types and regions as well—and on AWS, it also applies to Fargate and Lambda—with discounts of up to 66%. Google Cloud offers committed use discounts: resource-based for fixed vCPUs and memory, and spend-based for a fixed amount of usage. The more predictable your usage, the more you’ll benefit from the fixed-rate option.

A common target is 70–80% of steady-state baseline usage. Never commit to peaks or to workloads that will disappear within the contract term, and clean up and rightsize before you buy.

With showback, teams can see their costs without having to pay them. With chargeback, the costs are charged to their budget. Start with showback and switch to chargeback once cost allocation is reliable.

Cloud costs expressed per unit that matters to your business, such as per customer, transaction, or order. This lets you see if you’re becoming more efficient as you grow.

The team that uses the resources, supported by finance and a FinOps manager. Ownership must be explicitly assigned; otherwise, no one will feel responsible.

By serving data closer to the user (CDN, caching), limiting traffic between zones and regions, compressing data, and using gateway or private endpoints instead of NAT gateways.

Starting 12 January 2027, cloud providers will no longer be allowed to charge fees for transferring your data when you switch to another provider or move to an on-premises solution. Until that date, they may only pass on their own costs. You will still be charged for regular data traffic during use.

The FinOps Open Cost and Usage Specification: an open standard for cost and usage data from the FinOps Foundation. AWS, Azure, Google Cloud, and an increasing number of SaaS providers deliver their data in this format, allowing you to compare costs across multiple sources.

Allocate costs by application, choose the smallest model that is sufficient, use prompt caching and batch processing, and measure GPU utilization. Focus on costs per outcome, such as per query processed.

Residual capacity with up to a 90% discount, which the provider may reclaim at short notice. Suitable for batch processing, CI/CD, and stateless Kubernetes workloads that can tolerate interruptions.

At least once a month with engineering and finance, with automatic alerts for deviations. You review commitments on a quarterly basis.

Schedule a free cloud cost analysis

SUE will help you identify where you can save money in your organization. Learn more about Cloud Cost Optimization at SUE and the Cloud Cost Optimization Assessment.

Jonah Sanderse
Jonah Sanderse

Let’s talk


Schedule a free cloud cost analysis

SUE will help you identify where you can save money in your organization. Learn more about Cloud Cost Optimization at SUE and the Cloud Cost Optimization Assessment.

* required

We will respond within one business day. By submitting this form, you confirm that you have read and understood our privacy policy.
SUE Cloud & AI Native Solutions
Privacy overview
This website uses cookies. We use cookies to ensure that our website and services function properly, to gain insight into the use of our website, and to improve our products and marketing. For more information, please read our privacy and cookie policy.