Why use this calculator
Almost every cloud comparison you will be shown compares instance prices, and almost every real bill is decided by something else. Compute is the line the vendors advertise and the line that commitment discounts attack, but storage and data transfer out are unaffected by Savings Plans, Reserved Instances or Committed Use Discounts, and on a content-heavy or analytics workload egress can be the largest single item on the invoice.
This calculator separates the three so you can see which provider is genuinely cheaper for your shape of workload rather than for a generic one. It also lets you flip between on-demand and committed pricing, which frequently changes which provider wins, because the three do not discount at the same depth. Use it to sanity-check a migration business case before anyone commits to a three-year term.
How to use it
- Enter the total number of vCPUs across the instances in scope, which you can read off the instance type in your console or from the sizing in your infrastructure code.
- Enter the matching total memory in gigabytes from the same instance specification.
- Put the provisioned block storage volume size in the storage field, not the amount of data you have actually written, because you are billed on what is provisioned.
- Set running hours to 730 for an always-on server, or to your scheduler uptime if non-production machines are stopped outside business hours.
- Take data transfer out from the egress line of last month's bill or from your CDN report, since this is the figure most estimates omit entirely.
- Finally choose the pricing model you are willing to commit to, remembering that a three-year term is a balance-sheet decision and not just a discount.
What each result means
- AWS estimated monthly cost
- Indicative AWS bill for this shape of workload at list price, split in the note into compute, storage and egress.
- Azure estimated monthly cost
- The same workload priced against Azure list rates, so the three columns can be compared like for like.
- Google Cloud estimated monthly cost
- The same workload on Google Cloud, whose larger free egress allowance often shows up here first.
- Cheapest provider
- Which of the three is lowest on these inputs. Treat it as a tie-breaker, not a decision, because services and support differ.
- Annual saving from a three-year commitment
- How much a year the cheapest bill would fall if you committed to three years of compute instead of your current option, with storage and egress unchanged. The note gives the one-year figure, which keeps you more flexible.
- Lowest monthly cost
- The winning provider's total before support plans, licences, managed database premiums and observability tooling are added.
- Lowest annual cost
- Twelve months of the winning total. Useful for a budget line, but real estates grow as environments multiply.
- Annual difference, cheapest versus dearest
- The yearly spread between the best and worst of the three. Small spreads rarely justify a migration programme.
- Egress cost on the cheapest provider
- What data leaving the platform costs each month, and what share of the bill that represents.
- Compute share of the cheapest bill
- The only portion a commitment discount can reduce. A low share means a reservation will disappoint you.
The formula
Monthly cost = compute + storage + egress
- compute
- (vCPUs multiplied by the vCPU hourly rate plus memory in GB multiplied by the memory GB-hour rate) multiplied by running hours multiplied by one minus the commitment discount; storage = provisioned GB multiplied by the GB-month rate; and egress = the greater of zero and (transfer GB minus the provider's free allowance) multiplied by the per-GB egress rate. vCPU and memory rates are per hour
- storage
- per gigabyte per month
- egress
- per gigabyte
The commitment discount is applied to compute only because no provider discounts storage or egress under a compute commitment.
How this number is used in practice
Egress is where cloud estimates and cloud invoices part company. Compute and storage are predictable because you provision them, but egress is driven by user behaviour, by how chatty your services are across regions or availability zones, and by anything that copies data out of the platform such as backups to another vendor, log shipping to a third-party observability tool, or a data science team pulling extracts down to laptops.
All three majors now waive an initial tranche of internet egress each month, but once you are past it the per-gigabyte rate is charged on every byte and it is billed at retail even while your compute sits on a negotiated discount. On a media, analytics or public API workload it is common for transfer to be the single largest line, which is exactly why the comparison above splits it out rather than folding it into a total.
Committed pricing is the second place estimates go wrong, and it goes wrong in both directions. A one-year commitment typically removes somewhere between a quarter and a third of compute cost, and a three-year commitment roughly half, but the discount only touches compute. A workload that is 70 per cent storage and egress will see its total bill fall by a fraction of the headline percentage, and teams who budgeted on the headline end the year over.
In the other direction, organisations refuse to commit at all because they fear lock-in, then pay on-demand for years on a baseline that never moved. The practical rule most mature platform teams settle on is to commit the steady baseline, usually the floor of the last six months of usage, and leave the variable layer on demand or on spot capacity.
Price is also not the whole comparison. Egress between availability zones, inter-region replication, NAT gateway data processing, managed database premiums, support plan tiers charged as a percentage of spend, and licence costs for Windows or commercial databases all sit outside this model and can easily exceed the difference between providers.
So does the cost of your own people: a team fluent in one provider will ship faster and break less there than on a platform that is nominally 5 per cent cheaper. Use the comparison to rule out an option that is dramatically more expensive for your shape of workload, then confirm the shortlist in each vendor's own pricing calculator with your actual region and instance families.
Worked examples
A single always-on 4 vCPU, 16 GB server with 500 GB of disk and 100 GB of egress, on-demand
AWS $222.21, Azure $239.55 and Google Cloud $220.53 per month, with Google Cloud cheapest
The same server under a three-year commitment
AWS drops to $131.10 and now wins, against Azure at $134.20 and Google Cloud at $142.09
An egress-heavy media workload: 8 vCPUs, 32 GB, 2 TB of disk and 20 TB out per month
Google Cloud is cheapest at $2,224.06 a month, of which $1,683.00 is egress alone
When people use it
- Sanity-checking a cloud migration business case before it reaches a finance committee
- Working out whether egress or compute is the right target for a cost reduction programme
- Testing how much a one or three year commitment would actually take off the total bill
- Comparing providers for a new workload where no incumbent platform decision exists
- Building an infrastructure budget line for a startup plan or an annual forecast
Tips
- Pull your real egress figure from last month's bill before you model anything. Guessing it is the single biggest source of error in cloud estimates.
- Commit only the baseline you are confident will still exist in a year, and leave the variable layer on demand. Unused commitment is a pure loss.
- Check the free egress allowance separately for each provider. It differs between them and it can flip the winner on a small workload.
- Turn off non-production environments outside business hours. Dropping from 730 to about 200 running hours a month cuts non-production compute by roughly 70 per cent.
- Tag everything from day one. Without cost allocation tags you cannot attribute a bill to a team and no reduction programme will hold.
- Storage is billed on what you provision, not what you use. Oversized volumes and forgotten snapshots are the most common quiet waste in any estate.
- Verify the shortlist in each vendor's own pricing calculator with your real region and instance family before anyone signs a term.
What it assumes, and where it stops being accurate
- Rates are indicative September 2026 list prices for a US region, a general-purpose instance family and balanced SSD block storage. Every provider charges differently by region, and enterprise agreements are not list price.
- Compute is modelled as a linear blend of vCPU and memory rates. Real instance types are sold as fixed bundles, so a workload that needs an unusual CPU to memory ratio will price differently.
- Only internet egress is charged. Cross-zone, cross-region and NAT gateway data processing charges are excluded and are frequently larger than internet egress on microservice architectures.
- Support plans, licences, managed service premiums, load balancers, backups, snapshots, observability and marketplace software are all excluded.
- Commitment discounts are the published headline rates for a flexible commitment applied to compute only, and assume you keep the commitment fully used. Unused commitment is money burned.
- Not accurate for serverless, container-per-request or spot-heavy architectures, where billing is per request or per second of actual execution rather than per provisioned hour.
- Currency is US dollars throughout. Local billing currency and taxes are not applied.
Terms used on this page
- Egress
- Data transferred out of the cloud provider to the internet or another network, billed per gigabyte after a free allowance.
- On-demand
- Pay-as-you-go pricing with no commitment, the most expensive per-hour rate and the most flexible.
- Commitment discount
- A reduced compute rate in exchange for a one or three year spend commitment. Known as Savings Plans, Reserved Instances or Committed Use Discounts.
- vCPU
- A virtual CPU, usually one hardware thread. The unit cloud compute is sold and priced in.
- FinOps
- The practice of managing cloud spend as a shared engineering and finance responsibility rather than a procurement exercise.
Questions people ask
- Are these prices exact?
- No. They are indicative September 2026 list rates for a US region and a general-purpose instance family, used for relative comparison. Your actual cost depends on region, instance type, storage class, support tier and any enterprise agreement, so confirm the shortlist in each vendor's own pricing calculator before committing.
- Why is egress so expensive compared with storage?
- Storage is a commodity the provider buys once and amortises, whereas egress consumes network capacity and peering that is priced as an ongoing service. It is also the charge that makes moving data off the platform costly, which is why regulators in several markets have taken an interest in it and why some providers now waive egress when you leave entirely.
- How much do reserved instances and committed use discounts actually save?
- Published headline rates sit near a quarter to a third off compute for a one-year commitment and around half for three years, varying by provider and by how flexible the commitment is. Because they apply to compute only, a bill dominated by storage and egress will fall by far less than the headline percentage suggests.
- Should I pick a provider on price alone?
- No. The spread between the three for a typical workload is often under 10 per cent, which is smaller than the productivity difference between a team working on a familiar platform and an unfamiliar one. Weigh managed service maturity, compliance certifications, regional coverage, existing skills and exit cost alongside the monthly figure.
- What is missing from this estimate?
- Cross-zone and cross-region transfer, NAT gateway data processing, load balancers, snapshots and backups, managed database premiums, support plans charged as a percentage of spend, and commercial licences. On a microservice estate these can exceed the difference between providers entirely.
- Does this work for serverless or container workloads?
- Not well. Serverless is billed per request and per gigabyte-second of execution rather than per provisioned hour, so a provisioned model like this one will usually overstate it for bursty traffic and understate it for sustained traffic. Use the provider's own serverless calculator for those workloads.
Use this with
Cloud Cost Comparison Calculator answers one part of the question. These pick up where it stops: IT Downtime Cost Calculator, Technical Debt Cost Calculator, SaaS Gross Margin Calculator, Startup Burn Rate Calculator and ROI Calculator.
Where these figures come from
This calculation is jurisdiction-neutral arithmetic. No country-specific rules are applied.
- AWS EC2 On-Demand Pricing. Compute and storage list rates
- AWS Savings Plans. Published commitment discount ranges
- Azure Pricing Calculator. Azure compute, storage and bandwidth rates
- Google Cloud Pricing. Compute Engine rates, network tiers and committed use discounts
- FinOps Foundation. Cloud cost management practice and terminology
Cloud Cost Comparison Calculator produces an estimate for general information. It is not financial, tax, medical, legal or engineering advice. read the full disclaimer. Spotted a wrong number? Report it and it gets fixed. The form picks up this calculator and the inputs you used.