AWS Cost Estimator

Work out what AWS would actually cost you, for a migration or a new build. The estimate includes the line items people leave out — NAT Gateway, egress, Multi-AZ, snapshots and Support — and every rate and formula is printed with your numbers in it.

Prices as at 27 May 2026 On-demand, USD, taken from the AWS Price List API. That date is the oldest of the seven price files behind this tool, not the newest — the per-service dates are in the assumptions below. AWS has published newer prices since. Treat the figures as indicative.

Migrating what I have You are running servers somewhere else and want to know the AWS bill. Building something new Nothing exists yet. Size it from what the application needs.

01 / Region — prices differ, sometimes by a lot

AWS region Changing this re-prices everything below.

Spread across two availability zones. Adds a second NAT Gateway, which is the part that shows up on the bill.

02 / Servers — the sizing, and what it maps to

How many servers Identically sized. Mixed fleets: run it once per group.

vCPU per server What each one needs, not what the host has.

RAM per server, GB Rounded up to the nearest instance that fits.

Disk per server, GB gp3 volumes at the included 3,000 IOPS and 125 MB/s.

Snapshot storage kept, GB Total across all retained snapshots. Blank means none.

03 / Database — RDS, or none at all

Engine SQL Server and Oracle are out of scope, with Windows licensing.

Database vCPU

Database RAM, GB

Database storage, GB Multi-AZ stores it twice, and bills for both.

Backup storage kept, GB Backup up to your allocated size is free. Only the excess bills.

Multi-AZ. Roughly doubles the database instance and its storage.

Changing engine as part of the move. Widens the effort estimate a long way.

04 / Traffic — where the surprises live

Traffic out per month, GB To the internet. First 100 GB a month is free.

NAT Gateway, so private servers can reach the internet.

Through the NAT Gateway, GB Blank matches your traffic out. NAT bills per GB, and egress bills again.

Application Load Balancer in front of the servers.

Average LCUs One is the minimum and covers most small sites.

05 / Everything else — optional, blank means none

S3 storage, GB Standard class.

S3 uploads, thousands/month

S3 downloads, thousands/month

Custom CloudWatch metrics Standard EC2 metrics are free. Custom ones are not.

Logs ingested, GB/month

Logs kept, GB

Route 53 hosted zones

DNS queries, millions/month

AWS Support plan A percentage of everything above, not a flat fee.

06 / Commitment and comparison — both optional

Apply a savings plan or reserved instance discount.

Discount, percent Around 30% for a one-year, no-upfront commitment. Change it to whatever you have been quoted. It applies to EC2 and RDS instance hours only — storage, traffic, NAT and the load balancer are never discounted.

Current hosting bill, USD/month Optional, and the comparison below only appears if you fill it in. It stays in your browser like everything else.

Data to move across, TB Counts towards the effort estimate, not the monthly bill.

Runs in your browser. Your figures never leave this page.

Estimate

What this would cost on AWS

Estimated monthly bill

Monthly breakdown

Sizing

Override the instance type

Override the database class

Sizing picks the cheapest current-generation instance that meets both numbers. Graviton is excluded for EC2, because ARM is an architecture migration rather than a sizing choice. RDS keeps it, because AWS runs the engine and nothing needs porting. The override list holds one instance per size, the cheapest at that vCPU and memory. So a type you already run may not be in it — pick the entry with the same vCPU and memory instead. It costs the same or less for the same size.

What people do not budget for

Effort to get there

Engineer-hours, as a range, broken into components. This tool never publishes an hourly rate — multiply by whatever your own engineering time costs, or by a quote you have been given.

Every assumption behind these numbers

A month is 730 hours, which is how AWS bills one. Not 720, not 744.
Every rate is on-demand, in USD, for the region you picked, taken from the AWS Price List API. Taxes are not included.
Sizing picks the cheapest current-generation instance meeting both your vCPU and your RAM. You can override it, and the override is used exactly as chosen even if it is smaller than you asked for.
EBS is priced at the gp3 baseline — 3,000 IOPS and 125 MB/s are included at no charge, and this tool never charges for provisioned IOPS or throughput above that. If you genuinely need more, add it separately.
The first 100 GB of traffic out each month is free. Above that, the tiered rates apply in order.
NAT Gateway bills per hour and per gigabyte processed, and traffic leaving through it is then billed again as egress. That double charge is real and it is the most commonly missed item here.
Multi-AZ bills the database instance at the Multi-AZ rate and stores the data twice. Where AWS publishes no Multi-AZ rate for an engine, twice the single-AZ rate is used and the sizing line says so.
RDS backup storage up to your allocated database storage is free. Only the excess is charged.
A savings plan or reserved instance discount applies to EC2 and RDS instance hours only. Storage, traffic, NAT Gateway, the load balancer, CloudWatch and Route 53 are never discounted, and this tool does not pretend otherwise.
The discount percentage is a starting point you can edit, not a rate read from AWS. The price table carries on-demand terms only.
AWS Support is a percentage of everything else, with a monthly minimum. Developer is 3% or $29. Business is 10% or $100. Both are simplified — AWS tiers the percentage down at high spend, so heavy bills pay slightly less than this shows.
Effort is engineer-hours only. It contains no rate, no markup and no assumption about who does the work.
Out of scope and deliberately absent: Lambda, ECS, EKS, Fargate, DynamoDB, Redshift, CloudFront, Windows licensing and Marketplace AMIs.
There is no multiplier, weighting or industry average anywhere in this tool. Every figure comes from a published AWS rate and something you typed.

Each service is priced from its own file with its own publication date. The headline date at the top of this page is the oldest of them, so most of these are fresher than it suggests.
Price file	Covers	Published
AmazonRoute53	Route 53 hosted zones and queries	27 May 2026
AWSDataTransfer	Data transfer out to the internet	20 July 2026
AWSELB	Application Load Balancer	20 July 2026
AmazonCloudWatch	CloudWatch metrics and logs	6 August 2026
AmazonS3	S3 storage and requests	7 August 2026
AmazonEC2	EC2 instances, EBS volumes, snapshots, NAT Gateway	10 August 2026
AmazonRDS	RDS instances, database storage and backups	12 August 2026

No email needed. The PDF is built in your browser.

How it works

What this estimator actually does

Most AWS calculators either ask for forty fields or hide the rates behind a slider. This one asks what your servers look like, prices them at AWS’s own published on-demand rates, and prints every multiplication it performed. Nothing is rounded up to make a point.

The rates are not typed in by hand. They come from the AWS Price List API, refreshed automatically, and stamped with the date AWS published them. So the numbers move when AWS moves them, and the page tells you when it last looked.

Sizing works the same way. Give it the vCPU and memory each server needs, and it picks the cheapest current-generation instance that meets both. Then it shows you which one it picked and lets you change it. There is no hidden rounding and no “recommended” tier quietly costing double.

Two modes, one engine

Migrating and building from scratch ask the same question: what will this cost per month? So both run the same arithmetic. Only two things differ. Migration mode asks what you pay today, so it can compare. Greenfield mode asks whether you are using a second availability zone, because nobody migrating has that choice yet.

Why the effort estimate is in hours, not money

Because an hourly rate on this page would turn it into a quote, and a quote for work nobody has scoped is worthless. Instead you get engineer-hours, as a range, split into components you can argue with. Multiply by your own cost, or by a quote you have already been given.

The range is deliberately wide at the top end. Migrations run long when something unexpected turns up, and something usually does.

Line items

Every charge this tool prices

Ten line items, and no more. Each one is listed here with what drives it, so you can tell at a glance whether the estimate covers your setup or misses something.

EC2 instances Charged per hour, per server, whether busy or idle. This is usually the largest single line and the one people already expect.
EBS volumes Your disks, charged per gigabyte per month whether you fill them or not. Priced at the gp3 baseline, which includes 3,000 IOPS and 125 MB/s at no extra cost.
EBS snapshots Backups of those disks. Cheap per gigabyte, and easy to forget. Snapshots accumulate quietly until somebody looks.
RDS instances The managed database, charged per hour like EC2. Multi-AZ roughly doubles it, because AWS runs a standby for you.
RDS storage and backups Charged separately from the instance. Multi-AZ stores your data twice and bills for both copies. Backup storage up to your allocated size is free; only the excess costs anything.
Application Load Balancer An hourly charge plus a capacity charge measured in LCUs. Small sites sit at the one-LCU minimum, so the hourly part dominates.
NAT Gateway Hourly, plus a fee on every gigabyte through it. Two zones means two gateways. This is the line item that surprises people most.
Data transfer out Every gigabyte leaving AWS for the internet, after the first 100 GB each month. The rate falls in tiers as volume rises.
S3 storage and requests Per gigabyte stored, plus a small charge per thousand uploads and downloads. Usually modest unless you serve media from it.
CloudWatch and Route 53 Custom metrics, log ingestion and log retention, plus your hosted zones and DNS queries. Individually small, collectively not.

What it deliberately does not price

Lambda, ECS, EKS, Fargate, DynamoDB, Redshift, CloudFront, Windows licensing and Marketplace AMIs are all out of scope. Their cost depends on request counts and runtime durations that a form cannot sensibly ask for, and a confident wrong number is worse than an honest gap. AWS Support is handled as a percentage instead of an eleventh line, because that is how AWS charges for it.

Common problems

Where an AWS bill goes wrong

Almost nobody is caught out by the instance price. They are caught out by the charges that have no equivalent on a VPS invoice, and by the work nobody costed. Here are the five that come up most, and what to do about each.

The NAT Gateway bills twice, and never sleeps

It costs roughly $33 a month before a single byte moves. Then every gigabyte through it is charged again, and that same traffic is charged a third time on the way out as egress. Two availability zones means two gateways.

The fix. Small setups often do not need one. Put the instances in a public subnet with tight security groups and they reach the internet directly. Where traffic is only going to AWS services, VPC endpoints for S3 and DynamoDB cost nothing per gigabyte and remove the gateway from that path entirely.

Bandwidth was included before, and is not now

This is the single biggest reason a VPS at $40 becomes an AWS bill at $300. Your old host bundled several terabytes. AWS gives you 100 GB a month and charges for the rest.

The fix. Put CloudFront in front of anything static. Its egress is cheaper than EC2’s, and cached responses never touch your servers. Compress what you serve, and check nothing is shipping uncompressed images or unminified bundles.

Multi-AZ doubles the database, quietly

Ticking Multi-AZ runs a second instance you never see, and stores your data twice. Both are billed. It is not a mistake, and it is not free.

The fix. Decide it against your own downtime cost rather than by reflex. If restoring from a snapshot and losing an hour is survivable, single-AZ with good backups is defensible. If it is not, keep Multi-AZ and budget for it properly.

Snapshots accumulate until somebody looks

Automated snapshots with no retention policy grow forever. Each one is cheap, which is exactly why nobody notices for eighteen months.

The fix. Set a lifecycle policy on day one. Decide how far back you genuinely need to restore, then delete everything older. Most businesses need days, not years.

The migration itself never made the budget

The monthly bill gets planned carefully. The work to get there gets estimated as “a couple of weekends”, and then runs into networking, IAM, DNS and a database that will not cut over cleanly.

The fix. Cost it before you commit. The effort breakdown above splits it into components you can hand to different people, absorb in-house, or drop. A rehearsed rollback is the one item worth keeping whatever else goes.

FAQ

Questions people ask about AWS costs

How accurate is this estimate?

The rates are exact. They come straight from the AWS Price List API, and the page shows every one of them with your numbers beside it. The sizing is where the uncertainty sits. If you guess your traffic, the total is a guess too. Treat it as a realistic starting figure rather than a quote.

Why is my AWS estimate so much higher than my current hosting?

Usually because a VPS bundles things AWS charges for separately. Bandwidth is the big one. Your VPS probably includes several terabytes; AWS bills every gigabyte out. Then add the NAT Gateway, the load balancer and Multi-AZ, none of which have an equivalent line on a VPS invoice. You are not comparing the same thing, which is exactly why the breakdown is itemised.

What is a NAT Gateway, and why does it cost so much?

It lets servers in a private subnet reach the internet without being reachable from it. AWS charges for it twice: an hourly fee that runs whether you use it or not, and a per-gigabyte fee on everything passing through. That traffic is then billed again on the way out as egress. Small setups are often better served by putting the instances in a public subnet with tight security groups, or by using VPC endpoints for AWS-only traffic.

Do I actually need Multi-AZ?

It depends what an hour of downtime costs you. Multi-AZ roughly doubles the database line, and in exchange AWS fails over automatically when a zone goes down. If you can tolerate restoring from a snapshot and losing an hour, skip it. If your customers cannot, it is cheap insurance. Run the estimate both ways and compare the difference against your own downtime cost.

Are savings plans and reserved instances worth it?

Generally yes, if the workload is steady. A one-year no-upfront commitment typically saves around 30% on instance hours. The catch is that it only covers compute. Storage, traffic, NAT Gateway and load balancers stay at full price, so the discount applies to a smaller slice of the bill than people expect. This tool applies it to EC2 and RDS hours only, for that reason.

Why are Lambda, containers and Kubernetes not included?

Because their cost depends on things a form cannot ask for. Lambda bills per request and per millisecond. Fargate bills per task per second. Getting either wrong by a factor of ten is easy, and a wrong number is worse than no number. The same goes for CloudFront, DynamoDB, Redshift and Windows licensing. They are named as out of scope rather than quietly omitted.

How current are the prices?

The table is refreshed automatically from the AWS Price List API and stamped with the date AWS published it. The headline date on the page is the oldest of the seven price files behind the tool, never the newest, so it understates freshness rather than overstating it. Each file also carries its own date in the assumptions block.

What does the effort estimate include?

Building the network and access layer, moving or building each server, the database work, the DNS and TLS cutover, and testing with a rehearsed rollback. Every component shows its own range and they add up to the total, so nothing is hidden in a multiplier. It is expressed in engineer-hours only. There is no rate anywhere in this tool.

Do you store the numbers I enter?

No. The calculator runs entirely in your browser. Nothing is sent to a server, nothing is logged, and closing the tab discards everything. Your current hosting bill never leaves the page, which is also why the results are not encoded into the address bar. The PDF is generated on your machine too.

Want a second opinion on the sizing?

Run an estimate above and this section will point at whatever the numbers say is worth talking about. If you would rather just talk it through, book a call.

Send me a message ↓

Prefer to talk? Book a free call ↗ · Or hire me on Upwork ↗ · Typical reply within one business day.