Ask what hosting a custom web application costs and you will hear two kinds of answer. One says “a few dollars a month”, which is true of a hobby project on a single server. The other is a horror story about a five-figure invoice. Neither helps you budget a real business application.
Here is the bill for two realistic setups, priced line by line from AWS’s own published price lists for US East (N. Virginia), followed by the parts of the cost that no price list shows.
The prices, from AWS’s own price list
All figures are on-demand, Linux, US East, as published by AWS at the start of November 2026. A month is 730 hours, which is how AWS bills it.
| Resource | Published rate | Per month |
|---|---|---|
| EC2 t4g.small (2 vCPU, 2 GiB) | $0.0168 an hour | $12.26 |
| EC2 t4g.medium (2 vCPU, 4 GiB) | $0.0336 an hour | $24.53 |
| EC2 m7g.large (2 vCPU, 8 GiB) | $0.0816 an hour | $59.57 |
| RDS PostgreSQL db.t4g.micro, single zone | $0.016 an hour | $11.68 |
| RDS PostgreSQL db.t4g.medium, multi-zone | $0.129 an hour | $94.17 |
| RDS PostgreSQL db.m7g.large, multi-zone | $0.337 an hour | $246.01 |
| RDS gp3 storage | $0.115 per GB, doubled to $0.23 multi-zone | per GB |
| Application Load Balancer | $0.0225 an hour, plus capacity units | $16.43+ |
| Data transfer out to the internet | first 100 GB free, then $0.09 per GB up to 10 TB | per GB |
| CloudWatch Logs ingestion | $0.50 per GB | per GB |
Two things stand out before any modelling. A multi-zone database costs about twice the single-zone one: storage exactly double, the instance within a cent an hour of it, because you are paying for a standby copy in another data center. And the database is the most expensive thing on the list at every size. Application servers are cheap. State is not.
Example one: a small production business application
An internal tool or a B2B portal with a few hundred active users. Built the way we would build it for a client: two application servers behind a load balancer so one can fail, a database with a standby, and a smaller staging copy so changes are tested before they reach production.
| Line | Basis | Per month |
|---|---|---|
| Application servers | 2 × t4g.medium | $49.06 |
| Load balancer | 1 ALB, hours only | $16.43 |
| Database | db.t4g.medium, multi-zone | $94.17 |
| Database storage | 100 GB gp3, multi-zone | $23.00 |
| Data transfer out | 500 GB, first 100 GB free | $36.00 |
| Logs | 20 GB ingested | $10.00 |
| Staging | t4g.small, db.t4g.micro, 20 GB storage | $26.24 |
| Total | about $3,060 a year | $254.90 |
Look at what drives it. The database and its storage are $117 of the $255. Traffic, the thing people worry about, is $36. If usage doubled tomorrow, the bill would barely move. If usage fell to zero, most of it would still be there.
Example two: the same application, grown
A customer-facing product with real traffic: larger servers, a larger database, five times the storage, and 5 TB a month going out to users as pages, API responses, images and downloads.
| Line | Basis | Per month |
|---|---|---|
| Application servers | 2 × m7g.large | $119.14 |
| Load balancer | 1 ALB, hours only | $16.43 |
| Database | db.m7g.large, multi-zone | $246.01 |
| Database storage | 500 GB gp3, multi-zone | $115.00 |
| Data transfer out | 5 TB, first 100 GB free | $451.80 |
| Logs | 100 GB ingested | $50.00 |
| Total, before staging | about $11,980 a year | $998.38 |
Now the shape has changed. Data transfer is the biggest line and the only one that scales directly with success. That is why images and downloads usually go behind a CDN early, and why “how much data do we send, and to whom” belongs in an architecture review rather than a surprise on an invoice.
The lines that catch people out
Environments nobody turned off
A staging copy, a demo built for one sales meeting, a load-test cluster from March. Each runs at full price until someone deletes it, and nobody owns deleting it.
Tag every resource with an owner and a purpose, and review untagged spend monthly. Anything without an owner is a candidate for deletion.
Logging everything, forever
At $0.50 per GB ingested, a chatty application writing debug logs in production can cost more than the servers it runs on. Then the logs are kept indefinitely by default.
Set log levels deliberately and set retention periods on every log group when it is created, not after the first large bill.
Sizing for the launch-day fantasy
Servers and databases sized for the traffic in the business plan rather than the traffic that arrived. A database two sizes too large costs twice as much every hour of every month.
Start small and measure. Moving to a larger instance takes minutes on a managed service. Paying for unused capacity takes a year of invoices to notice.
Paying on-demand rates for things that never stop
The prices above are on-demand, the most expensive way to pay for a server that runs around the clock. AWS discounts committed usage through Savings Plans and reserved instances.
Commit only to the steady floor, usually the production database and baseline servers, once you have a few months of real usage to size it against.
This is not a rare failure. Flexera’s 2026 State of the Cloud report, from a survey of more than 750 cloud decision-makers and users, put wasted cloud spend at 29%. That was the first increase in five years, and Flexera attributed it to cloud AI workloads.
The cost no price list shows: people
Here is the comparison that changes how the decision looks. A median US software developer costs an employer about $820 per available working day, which we derived in what custom software actually costs.
For a small application, the decisions that matter are not “AWS or Azure” or “which instance type”. They are about how much engineering time the infrastructure consumes. Managed databases, automated backups, infrastructure defined in code and deployments that run themselves cost a little more on the invoice and a lot less in people. A setup that is cheap to run but needs someone to nurse it every week is the expensive option.
For very small or internal applications, a single well-managed virtual server can be cheaper than everything above, and that is a legitimate choice. What you give up is the standby database, automatic failover and the ability to scale without a migration. The right answer depends on what an hour of downtime costs you, which is a business question rather than a technical one.
What to ask before you sign off a hosting plan
- What is the fixed monthly floor if nobody uses the system at all? That is your minimum commitment.
- Which line grows with usage, and at what rate? Usually data transfer, sometimes storage or a per-request service.
- How many environments are there, and who switches the temporary ones off?
- How many hours a month of engineering does this setup need to stay healthy, and who provides them?
- Who receives the billing alerts? If the answer is nobody, set a budget alarm before launch.
The running cost of software is mostly people, as we covered in what software maintenance really costs after launch. Hosting is the part with a price list, which makes it the easiest to argue about and the least likely to sink a budget.
We design and run cloud infrastructure for the custom software we build. Backups, monitoring and cost alerts are set up before launch rather than after the first surprise, and access rules are agreed up front, as in security requirements to specify before development starts. See our cloud and DevOps work, or send us your current bill and we will tell you which lines are doing the damage.
Common questions
How much does it cost to host a web application on AWS?
A small production business application with two application servers, a load balancer, a standby database and a staging environment comes to about $255 a month, roughly $3,060 a year, at AWS’s published on-demand prices for US East in November 2026. A grown application sending 5 TB a month to users comes to about $1,000 a month. Most of a small app’s bill is the database, which costs the same whether anyone uses the system or not.
Why is my AWS bill higher than expected?
Usually one of four things: environments that were never switched off, log data ingested and retained without limits, servers and databases sized for traffic that never arrived, or steady workloads paid at on-demand rates. As traffic grows, data transfer out to the internet, at $0.09 per GB after the first 100 GB, becomes the line that moves. Flexera’s 2026 survey put average wasted cloud spend at 29%.
Is a multi-zone database worth twice the price?
If an hour of downtime costs you more than the difference, yes. A multi-zone database keeps a standby copy in a second data center and fails over automatically, and on AWS it costs about twice the single-zone price: storage exactly double, the instance within a cent an hour of double. For internal tools that can tolerate an outage while a backup is restored, a single zone with tested backups is often enough.
What costs more, cloud hosting or the people who run it?
For small and medium applications, the people. At the BLS-derived employer cost of about $820 per working day for a median US developer, one day a month spent on infrastructure costs about $9,840 a year, more than three times the hosting bill of a small production application. Managed services that cost slightly more but need less attention are usually the cheaper choice overall.
Sources
- Amazon Web Services, public Price List files for Amazon EC2, Amazon RDS, AWS Data Transfer, Elastic Load Balancing and Amazon CloudWatch, US East (N. Virginia), published 29 October to 4 November 2026. EC2 on-demand Linux: t4g.small $0.0168, t4g.medium $0.0336, m7g.large $0.0816 per hour. RDS for PostgreSQL on-demand: db.t4g.micro single-AZ $0.016, db.t4g.medium multi-AZ $0.129, db.m7g.large multi-AZ $0.337 per hour; gp3 storage $0.115 per GB-month single-AZ and $0.23 multi-AZ. Data transfer out to the internet: $0.090 per GB for the first 10 TB a month beyond the free tier, $0.085 for the next 40 TB. Application Load Balancer $0.0225 per hour plus capacity units. CloudWatch Logs $0.50 per GB ingested in the Standard log class. AWS states customers receive 100 GB of data transfer out to the internet free each month.
- Flexera, Flexera finds cloud value is rising while AI waste grows, press release on the 2026 State of the Cloud Report, 18 March 2026. Survey of more than 750 cloud decision-makers and users; wasted cloud spend estimated at 29%, the first increase in five years.
- US Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025, and Employer Costs for Employee Compensation, June 2026, for the $820 per working day employer cost of a median software developer.
- Monthly figures use 730 hours a month. Load balancer capacity-unit charges, backups beyond the included allowance, DNS, monitoring tools and taxes are not included, so a real invoice will be somewhat higher.
Written November 2026. Cloud prices change and vary by region. Price your own setup in the AWS Pricing Calculator before committing.
