Cost to host a small web app: real monthly prices

A few hundred daily users barely touch a CPU. The monthly bill for the same app can still run from $0 to about $40, and almost all of that gap is fixed cost: things that bill by the hour whether or not anyone visits. Here is one small app priced four ways with real rates, then the design choices that change the bill after launch.

One always-on web service with 0.25 vCPU and 512 MB of memory. One PostgreSQL database holding 1 GB of data. A custom domain with HTTPS, a few hundred daily users, production only. Prices are monthly at 730 hours, taken from each provider's pricing pages in October 2026. AWS is US East.

• AWS. Express Mode creates the load balancer ($0.0225 per hour) and the Fargate service. The database is RDS db.t4g.micro ($0.016 per hour, Single-AZ), which has 2 vCPU and 1 GiB, more than the 256 MB databases below, with the 20 GiB minimum of storage at $0.115 per GB-month. Load balancer traffic, logs, public IPv4 addresses, and data transfer are extra.
• Render. A Starter web service ($7, 0.5 CPU, 512 MB) plus the smallest paid Postgres, Basic-256mb ($6, 256 MB, 1 GB of storage included).
• Railway. It bills what the services actually use: $20 per vCPU-month, $10 per GB-month of memory, $0.15 per GB-month of volume. I assumed the app averages 0.1 vCPU and 400 MB, and the database 0.05 vCPU, 250 MB, and 1 GB of volume. Hobby costs $5 and includes $5 of usage, so you pay whichever is larger.
• ProductEcho. The Free plan allows the Starter size (250m CPU, 512Mi) and includes one 1 GB database, so this app costs $0. On Pro, the console's rates are $0.035 per ACU-hour for app compute, $0.020 per DCU-hour for database compute, and about $0.12 per GB-month of storage. The app is 0.25 ACU and the smallest database 0.25 DCU, so usage is $10.16. Pro costs $10 and includes $7 of credit, so you pay the $10 plus the $3.16 above the credit: $13.16.

AWS bills the load balancer and the database instance every hour. Those two alone are about $28 before a single request. Render and ProductEcho Pro price the size you pick, not the traffic. Railway meters use, so an idle app is cheap and a busy one climbs: one vCPU kept busy all month is $20. A free tier costs nothing but is small and shared.

A lower number is not always the same product. Check what each includes: database backups and recovery, high availability, autoscaling, and who patches and renews certificates. The AWS database above is Single-AZ, and Multi-AZ costs more. A staging environment adds a second copy of whatever is fixed.

• Held-open requests. An app that keeps every request open for a long AI task needs more concurrent capacity than one that queues the work.
• Verbose logs. Unbounded debug logging turns into storage and ingestion charges.
• Connections. Limits are real. Render's smallest paid database allows 100 connections, and an app that opens a connection per request can hit it long before it hits a CPU limit.

Set a budget alert before public traffic, and make sure someone owns it. Tag each environment and app. After the first week, compare actual compute, database, traffic, and logs with this worksheet. Remove idle staging, trim log retention, and resize from observed peaks, not guesses.

If your app fits the reference size, the Free plan runs it for $0 with a custom domain and HTTPS. The tradeoff is a shared cluster and a small allowance, so Pro starts at $10 a month when you outgrow it. To compare any of these fairly on your own app, run the one-afternoon platform test, then connect your coding agent to try ProductEcho on Free.