There is no typical price for cloud services, because you pay for what you use: the services you run, times their unit price, times how much you consume. Four things set most bills: compute hours, stored data, data sent out to the internet, and support. Commitment discounts then cut the compute line by up to 72% on AWS and Azure.
Key takeaways
- Cloud bills follow a simple formula: service × unit price × volume. Change any one of the three and the bill moves.
- On-demand compute is billed by the second or hour. One- or three-year commitments cut it by up to 72% (AWS EC2 Instance Savings Plans, Azure reservations); spare-capacity Spot instances by up to 90%.
- The lines people forget are data transfer out, storage retrieval and minimum-duration fees, support plans and software licences.
- Waste is common. In Flexera's 2026 State of the Cloud survey, respondents estimated that 29% of their cloud spend is wasted.
- Get a real number by pricing your own workloads in the official AWS, Azure or Google Cloud calculators, then add support, licences and operations.
How is cloud pricing calculated?
Every public cloud bill is a sum of metered items. For each one, the provider multiplies a unit price by your usage: vCPU and memory per second or hour, gigabytes stored per month, gigabytes transferred, requests, or licences per user.
Three details make the formula behave differently from a traditional IT budget:
- Granular billing. AWS bills Linux On-Demand instances per second with a 60-second minimum, according to its EC2 On-Demand pricing page. Switch something off and the charge stops.
- Regional prices. The same instance type or storage class can cost different amounts in Stockholm, Frankfurt, Mumbai or Virginia. Data residency rules often decide the region before price does.
- Free allowances. AWS, for example, includes the first 100 GB per month of data transfer out to the internet free, aggregated across services and regions. Google Cloud gives new customers $300 in credit and a free tier on more than 20 products.
The main cloud pricing models
Most organisations end up using several models at once: commitments for the steady baseline, on-demand for variable load and Spot for interruptible jobs.
| Model | How it works | Published maximum discount | Best for |
|---|---|---|---|
| On-demand (pay as you go) | No commitment; pay the list price for what you use | None (list price) | New, spiky or short-lived workloads |
| AWS Savings Plans | Commit to a $/hour spend for 1 or 3 years | Up to 72% (EC2 Instance Savings Plans); up to 66% (Compute Savings Plans) | Steady compute on AWS |
| Azure Reserved VM Instances | Reserve VM capacity for 1 or 3 years | Up to 72% vs pay-as-you-go | Steady VMs on Azure |
| Google Cloud committed use discounts | Commit to resources or spend for 1 or 3 years | Up to 55% for most machine series; up to 70% for memory-optimised (resource-based) | Steady workloads on Google Cloud |
| Spot / spare capacity | Use spare capacity that can be reclaimed at short notice | Up to 90% vs On-Demand (AWS EC2 Spot) | Batch, CI/CD, rendering, fault-tolerant jobs |
| Per-user subscription (SaaS) | Fixed fee per user per month or year | Varies by vendor and volume | Email, collaboration, business apps |
Sources: AWS Savings Plans pricing, Azure Reserved VM Instances, Google Cloud committed use discounts and Amazon EC2 Spot Instances. These are maximums. Actual discounts depend on instance family, region, term and payment option, and three-year terms give the largest cuts. Our AWS Savings Plans guide explains how to size a commitment without overbuying.
What drives cloud costs?
Compute
Virtual machines, containers and serverless functions are usually the largest line. Cost depends on the instance size (vCPU and memory), the family (general purpose, compute-, memory- or GPU-optimised), the operating system, how many hours it runs and which pricing model applies. Oversized instances are the most common source of avoidable spend.
Storage
You pay per gigabyte-month, at a rate set by the storage class. Cheaper archive classes come with retrieval fees and minimum storage durations; S3 Glacier Deep Archive, for example, has a 180-day minimum. Snapshots and backups grow quietly unless a retention policy deletes them.
Data transfer
Data coming into the cloud is generally free, while data going out to the internet is charged per gigabyte after any free allowance. Traffic between regions, and in some cases between Availability Zones, is also metered. Applications that serve large files or replicate across regions feel this most.
Managed services and databases
Managed databases, Kubernetes control planes, load balancers, NAT gateways, monitoring and logging each add their own hourly or per-request charges. They save engineering time, but they need to be in the estimate.
Support plans
Basic support is included, but business support is priced as a share of your bill. AWS's support pricing, for example, charges Business Support+ at the greater of USD 29 per month or a sliding percentage of monthly AWS charges, starting at 9% of the first USD 10,000. Enterprise Support starts at USD 5,000 per month.
Software licences
Windows Server, SQL Server, Red Hat and commercial databases can cost as much as the infrastructure underneath. Bring-your-own-licence options, such as Azure Hybrid Benefit, change the comparison, so check which licences you already own.
Why is there no reliable "average cloud cost"?
Because cloud costs follow usage, two companies of the same size can have bills that differ by an order of magnitude. A firm running email and a CRM as SaaS pays per user; a firm running its own data platform pays for compute and storage at scale. Averages across such different estates do not help you budget, which is why this article does not quote one.
What survey data does show is that managing the bill is hard. In Flexera's 2026 State of the Cloud report, based on more than 750 cloud decision-makers, 85% named managing cloud spend as a top challenge, and estimated waste rose to 29%, the first increase in five years, with AI workloads adding to the problem.
How to estimate what cloud services will cost you
- List the workloads. For each application, note servers, CPU and memory, storage, databases, peak and average load, and how much data leaves the system each month.
- Map them to cloud services. Decide what moves as-is (virtual machines) and what changes (managed databases, containers, serverless). The architecture choice affects the bill more than the provider choice.
- Price them in the official calculators. Use the AWS Pricing Calculator, the Azure pricing calculator or the Google Cloud pricing calculator, with the region you will actually use.
- Add the forgotten lines. Data transfer, backups and snapshots, logging and monitoring, support plan, licences, security tools, and the people or partner who will run it.
- Model the discounts. Price the steady baseline with one- and three-year commitments and keep only the variable part on-demand.
- Compare total cost of ownership. Against on-premises, include hardware refresh, data centre space and power, maintenance contracts, and staff time, not just server prices.
Expect the first estimate to be off. Run the workload for a month or two, compare the bill with the plan, and adjust before you commit to a three-year term.
How much can moving to the cloud save?
It depends on what you move and how you run it afterwards. Moving servers as-is and leaving them running around the clock often costs more than expected. Savings come from rightsizing, auto-scaling, switching off idle environments and buying commitments for the baseline.
One example: when Opsio migrated Swedish ski resort group Branäsgruppen from VMware to AWS, the case reports a 40% reduction in total cost of ownership, achieved by pairing right-sized auto-scaling with FinOps practices. Their traffic is highly seasonal, which is where paying only for capacity in use helps most.
How to keep cloud spend under control
- Tag everything by team, application and environment, so cloud costs have an owner.
- Set budgets and alerts per account or project, so surprises surface in days, not at month end.
- Rightsize monthly using utilisation data, and delete unattached disks, old snapshots and idle load balancers.
- Schedule non-production environments to stop outside working hours.
- Review commitments quarterly so coverage follows your real baseline.
- Make it a shared process between engineering and finance. The FinOps Framework from the FinOps Foundation describes how to set this up.
How Opsio helps
Opsio prices and runs workloads on AWS, Azure and Google Cloud for organisations in the Nordics, the EU and India. We build cost estimates before cloud migration, and afterwards our cloud cost optimisation and FinOps services handle tagging, budgets, rightsizing and commitment planning. To compare the providers first, see our AWS vs Azure vs GCP guide.
Frequently asked questions
How much do cloud services cost per month?
Anything from nothing, within free tiers, to millions for large enterprises. Your monthly cost is the sum of compute, storage, data transfer, managed services, support and licences you use. Price your own workloads in the provider's calculator to get a realistic figure.
What is the average cost of cloud computing?
There is no meaningful average, because usage differs so much between organisations. A better benchmark is unit cost, such as cost per customer, per transaction or per environment, tracked over time.
Is cloud cheaper than on-premises?
It can be, but not automatically. The cloud is usually cheaper for variable or seasonal workloads and when you rightsize and use commitment discounts. Steady workloads moved unchanged and left running on-demand can cost more than the servers they replaced.
What are the hidden costs of cloud services?
The usual surprises are data transfer out of the cloud and between regions, storage retrieval and minimum-duration fees on archive tiers, growing snapshots and logs, support plans priced as a percentage of spend, and software licences.
How can I reduce my cloud costs?
Start with visibility: tag resources and set budgets. Then rightsize, delete idle resources, schedule non-production environments, move suitable jobs to Spot capacity, and cover your steady baseline with one- or three-year commitments.
Related reading
Related Services
Written By
Country Manager, Sweden
Johan leads Opsio's Sweden operations, driving AI adoption, DevOps transformation, security strategy, and cloud solutioning for Nordic enterprises. With 12+ years in enterprise cloud infrastructure, he has delivered 200+ projects across AWS, Azure, and GCP — specialising in Well-Architected reviews, landing zone design, and multi-cloud strategy.
Editorial standards: This article was written by the cloud practitioner named in the byline above. We revise it when the technology or regulation it covers changes; the dates on this page show the latest update. Opsio maintains editorial independence.