"Just tell me which one is cheaper."
That's the question behind almost every search for cloud computing cost, and it's not an easy one to answer. Amazon Web Services (AWS), Microsoft Azure, and Google Cloud each organize their pricing differently, with their own instance families, purchasing models, and discount structures. Comparing costs across providers isn't as simple as matching hourly rates.
This guide makes that comparison easier by using the same workload specifications across all three providers: identical compute, the same region, and the three main purchasing models, on-demand, committed, and spot. It also compares GPU pricing, where the differences between providers are often the largest.
So which provider is actually cheaper?
The answer depends less on the provider than on the purchasing model you choose. On-demand pricing is remarkably similar across all three. The biggest differences emerge with committed discounts, spot pricing, and GPU infrastructure, where each provider takes a noticeably different approach.
What drives cloud computing cost
Every cloud bill is made up of four primary cost components: compute, storage, networking, and licensing.
- Compute is the cost of the virtual machines, containers, or serverless resources running your workloads.
- Storage covers the data you keep at rest, whether it's block, object, or file storage.
- Networking includes data transfer charges, particularly egress, the cost of moving data out of a provider's network.
- Licensing covers commercial software running on your infrastructure, such as Windows Server or enterprise databases.
Of these four, networking, particularly egress charges, is the cost most organizations underestimate. While providers typically charge little or nothing for inbound traffic, moving data out of the cloud often incurs additional fees that can significantly increase your monthly bill.
Compute, storage, and licensing are usually visible in pricing calculators before you deploy a workload. Networking costs, especially egress, often become apparent only after the workload is running. That's why the hourly price of a virtual machine is only one part of your total cloud computing cost.
Compute pricing compared: on-demand, committed, and spot
To keep the comparison fair, every provider is measured using the same workload: 4 virtual central processing units (vCPUs), 16 GB of memory, Linux, and a standard United States region. That corresponds to an Amazon Web Services (AWS) m6i.xlarge instance in us-east-1, a Microsoft Azure D4s v5 instance in East US, and a Google Cloud n2-standard-4 instance in us-central1.
As a bonus, we've also included Rackspace Spot in the comparison. Rather than relying solely on fixed pricing, Rackspace Spot offers both standard virtual machines through an auction market. For this comparison, we've used a Large VM (4 vCPUs, 16 GB RAM) with a current market price of $0.04/hour from the San Jose region, giving a like-for-like comparison against the hyperscalers.
On-demand pricing by provider
On-demand pricing is remarkably similar across the three hyperscalers. Amazon Web Services (AWS), Microsoft Azure, and Google Cloud closely monitor one another's list prices, making headline compute rates a poor differentiator. Rackspace Spot takes a different approach. Rather than fixed pricing, its marketplace reflects real-time supply and demand, which can result in significantly lower compute costs for the same virtual machine configuration.
One-year committed pricing by provider
Google Cloud offers the deepest one-year discount of the three providers and is also the only one that automatically applies Sustained Use Discounts for eligible workloads without requiring every resource to be reserved in advance.
Spot pricing by provider
Although the Spot discounts appear similar across the hyperscalers, the pricing model differs. Amazon Web Services (AWS), Microsoft Azure, and Google Cloud determine Spot pricing internally, while Rackspace Spot uses an open-market auction where customers bid for capacity and prices reflect real-time supply and demand.
To better understand the differences between auction-based pricing and traditional Spot pricing, read our guide to the history of Spot instances and how cloud Spot markets have evolved.
Price stability also matters. According to Cast AI's 2025 Kubernetes Cost Benchmark Report, Amazon Web Services (AWS) repriced Spot capacity roughly 197 times per month across the instance types it tracked, while Google Cloud repriced comparable capacity only 0.35 times per month over the same period. Fewer price changes can reduce the operational overhead of running Spot workloads.
Compute is only one component of cloud computing cost. Storage, networking, and GPU pricing often have an even greater impact on your total monthly bill.
Block storage pricing comparison
Compute isn't the only contributor to cloud costs. Persistent block storage is typically billed per gigabyte (GB) per month, and the price differences between providers are often larger than the differences in compute pricing.
As with the compute comparison, we've included Rackspace Spot alongside the three hyperscalers as a bonus comparison. This provides a like-for-like view of how its storage pricing compares with Amazon Web Services (AWS), Microsoft Azure, and Google Cloud.
Standard SSD block storage pricing (per GB/month)
Rackspace Spot offers the lowest starting storage price in this comparison, with its NVMe-backed ssdv2 Persistent Volumes priced at $0.06 per GB/month. Amazon Web Services (AWS), Microsoft Azure, and Google Cloud all provide comparable premium solid-state drive (SSD) storage, but at higher starting rates.
Google Cloud's Persistent Disk SSD costs more than double Amazon Web Services (AWS) Elastic Block Store (EBS) gp3 for the same capacity. On a Kubernetes cluster running ten nodes with 100 GB persistent volumes each, that difference alone adds roughly $1,080 per year in storage costs. Google Cloud does offer Hyperdisk Balanced, priced at approximately $0.084 per GB/month, which narrows the gap considerably, but it isn't the default StorageClass in Google Kubernetes Engine (GKE) and must be configured explicitly.
Microsoft Azure's pricing also deserves a closer look. The advertised $0.113 per GB/month covers only the base storage capacity. Provisioned input/output operations per second (IOPS) above 3,000 and throughput above 125 MB/s are billed separately, increasing the cost of high-performance database and transactional workloads.
Storage costs accumulate steadily over time. GPU pricing, however, can change the cost of a workload in a single training run.
GPU and AI instance pricing
Graphics Processing Unit (GPU) pricing is more difficult to compare than Central Processing Unit (CPU) pricing because providers package GPUs differently. Amazon Web Services (AWS) bundles NVIDIA A100 GPUs into fixed-size instances, while Microsoft Azure and Google Cloud allow customers to provision as few as a single GPU. That packaging difference alone can significantly affect the total cost of artificial intelligence (AI) and machine learning workloads.
A100 on-demand pricing (normalized per GPU-hour, US region)
On paper, Microsoft Azure is the most expensive at approximately $4.41 per GPU-hour, roughly 60% higher than Amazon Web Services (AWS). However, the Amazon Web Services (AWS) price comes with an important trade-off. The p4d.24xlarge instance includes eight GPUs, meaning you pay for all eight regardless of whether your workload uses only one. Microsoft Azure and Google Cloud allow single-GPU provisioning, making them a better fit for smaller training or inference jobs.
Spot and preemptible GPU capacity narrows the gap further, though availability is the trade-off. Amazon Web Services (AWS) Spot pricing for P4d instances typically runs 50% to 70% below on-demand pricing when capacity is available. Microsoft Azure Spot Virtual Machines can reduce costs by 70% to 82%, with a 30-second eviction notice, compared with the two-minute interruption notice Amazon Web Services (AWS) provides for most Spot Instances.
That eviction window matters for long-running GPU training jobs, where an interruption before a checkpoint can result in hours of lost work. Ultimately, however, GPU pricing is only one part of the equation. Poorly managed storage, networking, and purchasing decisions can outweigh the savings from selecting the lowest-priced GPU.
What actually moves your cloud bill
The provider you choose often has less impact on your monthly bill than how you use the platform. Three factors consistently have the biggest effect: data egress, idle resources, and purchasing strategy.
1. Data egress is the hidden cost many teams overlook. A workload transferring 5,000 GB of data out of Google Cloud can incur roughly $600 in egress charges alone, potentially exceeding the cost of the compute resources themselves. Amazon Web Services (AWS) and Microsoft Azure generally charge less for the same transfer, but all three providers bill for outbound data, and those costs aren't reflected in the hourly compute prices you compare before deployment.
2. Idle resources are the easiest source of unnecessary spend. Unattached storage volumes, virtual machines left running after development work finishes, idle GPU instances over a weekend, and forgotten load balancers continue generating charges even when they're doing no useful work.
3. Purchasing strategy can also increase costs when it doesn't match workload behaviour. A team that commits to 50 virtual machines for one year and later migrates part of the workload to containers still pays for the unused committed capacity until the reservation expires.
Spot pricing is another example of how purchasing strategy affects cost, but in a different way. Instead of committing to capacity, organizations trade guaranteed availability for significantly lower prices. Here, the hourly rate is only part of the equation. How often Spot prices change also matters, since frequent repricing can increase the operational effort required to rebalance workloads and handle interruptions.
That's where pricing models begin to differ. Amazon Web Services (AWS), Microsoft Azure, and Google Cloud all manage Spot pricing internally, while Rackspace Spot uses an open-market auction where customers bid for compute capacity and pay the amount they bid when it clears the market. Bids start as low as $0.001 per hour, giving organizations another way to reduce infrastructure costs for interruption-tolerant workloads such as batch processing, continuous integration and continuous deployment (CI/CD) pipelines, extract, transform, and load (ETL) jobs, and distributed artificial intelligence (AI) training.
The real answer
The question, "Which cloud is cheapest?", doesn't have a single answer because cloud computing cost depends as much on how you buy infrastructure as where you buy it. On-demand compute pricing across Amazon Web Services (AWS), Microsoft Azure, and Google Cloud is remarkably similar. The biggest cost differences emerge when you compare purchasing models, storage, networking, GPU packaging, and Spot pricing.
The key is to compare providers using the workload you actually plan to run, not a headline hourly rate. For many organizations, the lowest total cost comes from choosing the right pricing model rather than switching providers altogether. For interruption-tolerant workloads, it's also worth evaluating alternatives such as Rackspace Spot, whose open-market auction model offers another approach to reducing compute costs.
Ultimately, effective cloud cost management isn't about finding the cheapest provider. It's about finding the provider and pricing model that best fit your workload before your monthly bill tells you otherwise.
Frequently asked questions
How much does cloud computing cost?
Cloud computing costs vary by provider, instance size, and purchasing model. A general-purpose virtual machine with 4 virtual central processing units (vCPUs) and 16 GB of memory costs around $0.19 per hour on-demand across Amazon Web Services (AWS), Microsoft Azure, and Google Cloud. Storage typically ranges from $0.06 to $0.17 per GB per month, while Graphics Processing Unit (GPU) instances cost between $2.75 and $4.41 per GPU-hour on-demand. Your total bill also depends on networking, software licensing, and data egress.
Which cloud provider is the cheapest?
Rackspace Spot is one of the cheapest cloud providers for interruption-tolerant workloads, with Spot instance bids starting as low as $0.001 per hour through its open-market auction model. For general-purpose on-demand compute, however, there isn't a single cheapest provider. Amazon Web Services (AWS), Microsoft Azure, and Google Cloud have very similar pricing, with the biggest cost differences coming from committed discounts, Spot pricing, storage, networking, and GPU packaging rather than headline compute rates.
Is AWS cheaper than Azure or Google Cloud?
For standard on-demand virtual machines, pricing is nearly identical across all three providers. Google Cloud generally offers the deepest one-year committed discounts, while Microsoft Azure often provides the largest Spot discounts. Organizations looking to reduce compute costs for interruption-tolerant workloads should also evaluate Rackspace Spot, whose auction-based Spot marketplace can significantly undercut traditional Spot pricing.
How do you calculate cloud computing costs?
Calculate cloud computing costs by adding together compute, storage, networking, and software licensing. Compute is charged by the hour or second, storage by the gigabyte per month, and networking primarily through outbound data transfer (egress). Don't forget additional costs such as load balancers, snapshots, managed databases, and monitoring services.
What increases cloud computing costs the most?
The biggest cost drivers are overprovisioned compute, data egress charges, idle resources, and choosing the wrong purchasing model. Many organizations spend more on unused infrastructure and outbound data transfer than they expect because these costs aren't obvious during initial deployment.
How does cloud computing reduce costs?
Cloud computing reduces costs by eliminating upfront hardware purchases and allowing organizations to scale infrastructure on demand. Additional savings come from committed-use discounts, autoscaling, and Spot pricing for interruption-tolerant workloads. Platforms such as Rackspace Spot can reduce compute costs even further through an open-market auction where Spot instance bids start at $0.001 per hour.
Is cloud computing cheaper than on-premises infrastructure?
It depends on the workload. Cloud infrastructure is usually more economical for variable or unpredictable demand because you pay only for the resources you use. Long-running, predictable workloads may become less expensive on owned infrastructure if utilization remains consistently high over several years.
What is the cheapest way to run workloads in the cloud?
For predictable workloads, committed pricing generally offers the best long-term value. For fault-tolerant workloads such as batch processing, continuous integration and continuous deployment (CI/CD), extract, transform, and load (ETL), and machine learning training, Spot instances usually provide the largest savings. Rackspace Spot is designed around this model, offering Spot instance bids from $0.001 per hour alongside managed Kubernetes, virtual machines, and managed PostgreSQL Database as a Service (DBaaS).
Why is cloud computing so expensive?
Cloud computing becomes expensive when resources are oversized, left running unnecessarily, or generate high networking charges. Poor cost governance, idle infrastructure, and unexpected data egress fees are among the most common reasons organizations exceed their cloud budgets.
What is the biggest hidden cost in cloud computing?
For many organizations, the biggest hidden cost is data egress, the fee charged for transferring data out of a cloud provider's network. Other frequently overlooked costs include idle virtual machines, unattached storage volumes, snapshots, logging, monitoring, and load balancers.
