Cloud pricing is hard to compare because the three largest providers, AWS, Microsoft Azure, and Google Cloud, use different units, discount structures, and billing rules. A single virtual machine can be priced per second, per hour, or per request, depending on the service. This guide explains how each provider charges, where the real differences sit, and what to check before estimating a bill.
How cloud pricing works across the big three
All three providers use pay-as-you-go pricing as the default. You pay for what runs and what is stored, and you can stop paying when you stop using a resource. On top of that base, each provider offers ways to lower the price in exchange for a commitment, or in exchange for capacity that can be taken back at short notice.
Most of the list prices for common services are close to each other. The bill is usually shaped by a smaller set of factors: how long resources run, whether they are covered by a commitment, how much data leaves the cloud, and which support and licensing options are added. Those factors are where two bills for similar workloads can diverge.
Pricing models at a glance
| Pricing model | AWS | Azure | Google Cloud |
|---|---|---|---|
| Pay-as-you-go | Default for most services | Default for most services | Default for most services |
| Long-term commitment | Savings Plans and Reserved Instances, up to about 72% off for 1 or 3 years | Reservations up to about 72% off; compute savings plan up to about 65% off | Committed use discounts up to about 57% off (about 70% for memory-optimized machines) |
| Automatic usage discount | None for most compute | None for most compute | Sustained use discounts up to about 30% on eligible machine families |
| Interruptible capacity | Spot Instances, up to about 90% off, 2-minute notice | Spot VMs, up to about 90% off, 30-second notice | Spot VMs, roughly 60–91% off, can be reclaimed |
| Licensing benefit | Bring-your-own-license options | Azure Hybrid Benefit for existing Windows Server and SQL Server licenses | Bring-your-own-license options on some products |
Discount percentages are maximums from each provider’s published terms. The discount you get depends on the instance type, region, term, and payment option, so treat these figures as ceilings, not promises.
Pay-as-you-go pricing
On-demand pricing has no commitment and no minimum spend. It suits short-lived tests, unpredictable traffic, and early projects where the workload is still changing. It is also the most expensive way to run a steady workload for a long period.
Billing granularity varies. Linux virtual machines on AWS and Google Compute Engine are billed per second with a one-minute minimum. Windows images and some licensed images can be billed differently, so check the specific image before estimating. Serverless services often bill by request and by resource use, which is covered in our serverless comparison.
Commitment discounts
Commitment discounts reward steady usage. They are most useful for the baseline capacity you already know you need, not for peaks.
AWS: Savings Plans and Reserved Instances
Savings Plans let you commit to a fixed hourly spend for one or three years. Compute Savings Plans apply to a broad set of compute usage, while EC2 Instance Savings Plans are tied to an instance family in a region. Reserved Instances are a billing discount on a specific instance configuration, with standard and convertible variants. Savings Plans and regional Reserved Instances do not reserve capacity, so availability is not guaranteed by the discount itself. Only zonal Reserved Instances include a capacity reservation.
Azure: reservations and savings plans
Azure reservations are commitments on specific resources, usually for one or three years. Azure also offers compute savings plans, which work like a flexible hourly commitment. Azure Hybrid Benefit lets organizations apply existing Windows Server and SQL Server licenses to Azure workloads, which can change the total cost more than the reservation discount itself for some Microsoft-heavy estates.
Google Cloud: committed use and sustained use discounts
Google Cloud has two discount types. Committed use discounts are contracts for vCPUs and memory in a region, typically for one or three years. Sustained use discounts apply automatically when a machine runs for a large share of the month, which means no upfront commitment is needed. Sustained use discounts are not available on every machine family, so check eligibility for the family you plan to use.
Spot and preemptible capacity
Spot instances on AWS, Spot VMs on Azure, and Spot VMs on Google Cloud all use spare capacity at a much lower price. The provider can reclaim them when it needs the capacity back, with a notice period that differs by provider. AWS gives a two-minute warning, and Azure gives 30 seconds.
Spot capacity suits batch processing, rendering, build runners, and other work that can restart from a checkpoint. It is a poor fit for a primary database or a service that cannot tolerate interruptions. Teams often combine a small committed baseline with spot capacity for the rest of a fault-tolerant workload.
Storage pricing: hot and archive tiers
Storage prices look similar at the top tier and diverge in the archive tiers. Based on approximate US list prices per GB per month:
- Hot object storage: Amazon S3 Standard is about $0.023. Azure Blob Storage hot tier is roughly $0.018–$0.021 depending on redundancy. Google Cloud Storage Standard is about $0.020.
- Archive storage: S3 Glacier Deep Archive is about $0.001. Azure Archive is roughly $0.001–$0.002. Google Cloud Storage Archive is about $0.0012.
Archive tiers come with rules that affect the bill. Each has a minimum storage duration, and deleting data early can still incur charges for the full period. Deep Archive and Azure Archive have a 180-day minimum, Google Cloud Storage Archive has 365 days, Coldline has 90 days, and Nearline has 30 days. Retrieval also takes time and costs money. S3 Glacier Deep Archive retrieval can take 12 to 48 hours, Azure Archive rehydration can take up to 15 hours, and Google Cloud Storage Archive returns data in milliseconds with retrieval fees.
PAA: How much does 1 TB of Google Cloud storage cost? At the Standard tier’s approximate list price of $0.02 per GB per month, 1 TB (about 1,000 GB) comes to roughly $20 a month before operations charges, network egress, and any regional differences. Check the Google Cloud Pricing Calculator for your region and class.
Data transfer (egress) charges
Data coming into a cloud provider is generally free on all three. Data leaving the cloud is where costs can grow quickly. AWS and Azure each include the first 100 GB per month of outbound internet transfer, then charge about $0.09 per GB. Google Cloud’s Premium Network Tier is about $0.12 per GB for the first tier, and the rate varies by destination. Google’s Standard Tier uses the public internet and is cheaper, with different performance characteristics.
Egress is also charged between some regions and services, so a design that moves data between regions or to another cloud can add costs that do not show up in compute pricing.
In 2024, all three providers announced that they would waive egress fees for customers leaving the platform. That waiver applies to migration out of the cloud, not to normal ongoing traffic, so check current terms before a migration.
Free tiers and new-account credits
Free tiers help with learning and early testing. Each provider works differently:
- AWS: Since July 2025, new accounts have received credits, up to $200 under the current offer, instead of the earlier 12-month free tier. Some always-free allowances also exist. Check the current terms before relying on them.
- Azure: A new-account credit of $200 is available for 30 days, along with a set of free services.
- Google Cloud: A $300 credit is available for 90 days, plus an Always Free allowance for small usage.
Free allowances are capped. Once a project exceeds them, normal pricing applies, and the bill can arrive at the end of the month when nobody is watching usage. For AWS in particular, see how to avoid unexpected AWS charges as a beginner.
A worked example for a steady workload
Suppose a virtual machine costs $0.10 per hour on demand and runs all month. A 730-hour month then costs about $73 at on-demand rates. A commitment that cuts the rate by 40% would bring the same machine to about $44. These figures are hypothetical and only show the arithmetic. Real rates depend on the machine type, region, and term.
Now add 1 TB of internet egress. At about $0.09 per GB after the first 100 GB free, that traffic costs roughly $81 a month on AWS or Azure. In this example the data transfer costs more than the discounted machine, which is common for content-heavy applications. The lesson is to model egress before committing to a design.
Costs that are easy to miss
- Public IPv4 addresses: AWS has charged $0.005 per hour for each public IPv4 address since February 2024. That is about $3.65 per address per month.
- Support plans: Paid support is extra on every provider. AWS Business Support starts at $100 per month or a percentage of spend, depending on the plan.
- Idle resources: Stopped virtual machines can still incur storage, snapshot, and IP charges.
- Logs and monitoring: Verbose logging and metrics can add charges that grow with traffic.
- Cross-region transfer: Replication across regions is billed as data transfer.
- Licensing: Windows Server, SQL Server, and some third-party images can change the per-hour rate.
Who is cheaper, AWS or Azure?
Neither is cheaper across the board. For many common services, list prices are close, and the gap that matters comes from discounts, egress, support, licensing, and region. An organization with existing Windows Server and SQL Server licenses may find Azure cheaper for those workloads. A team that stores large volumes of data and serves them to the internet should model egress carefully on either platform. The answer depends on the workload, so compare totals for a realistic configuration, not single line items.
How to compare prices for your own workload
Start with a simple inventory: the machines, storage, databases, and expected monthly outbound data. Then use each provider’s calculator with the same region, term, and usage assumptions:
- AWS Pricing Calculator
- Azure Pricing Calculator
- Google Cloud Pricing Calculator
Run at least three scenarios: on-demand, a one-year commitment, and a version with a realistic share of spot capacity. Add egress, support, and licensing before comparing totals. Prices change often, so date every estimate and recheck it before a purchase decision.
For a broader view of how the providers differ beyond price, see the AWS vs Azure vs Google Cloud comparison. Small teams comparing fixed-price virtual servers may also want the cloud hosting comparison chart, and the DigitalOcean vs AWS guide explains how flat pricing differs from usage-based billing. Teams running containers should check how control-plane fees compare in the managed Kubernetes comparison.