Cost per customer on AWS, Azure, Google Cloud and Oracle

Cost per customer on each cloud

Say what one customer does in a month: requests, data downloaded, files and database rows. The estimate sizes a web app for your customers and for twice, five and ten times as many, prices it on the four clouds, and divides by customers. Where a server or database tier is added, the cost per customer steps instead of sliding.

Cost per customer a month

At 1,000 customers the lowest complete estimate is Oracle Cloud, at $0.1505 per customer.

Cost per customer by cloud at each customer count
Customers	Microsoft Azure	Amazon Web Services	Google Cloud	Oracle Cloud
1,000 (now)	$0.6026	$0.4139	$0.3982	$0.1505 lowest
2,000 (2x)	$0.3411	$0.2445	$0.2432	$0.0812 lowest
5,000 (5x)	$0.1842	$0.1429	$0.1474	$0.0397 lowest
10,000 (10x)	$0.1466	$0.1238	$0.1324	$0.0326 lowest

Added cost per extra customer

The extra monthly cost between one row and the one before it, divided by the extra customers. A jump means a server or database tier was added.

Added cost per extra customer by cloud
From to	Microsoft Azure	Amazon Web Services	Google Cloud	Oracle Cloud
1,000 to 2,000	$0.0795	$0.0752	$0.0882	$0.0120
2,000 to 5,000	$0.0795	$0.0752	$0.0835	$0.0120
5,000 to 10,000	$0.1090	$0.1047	$0.1174	$0.0255

What was sized

Servers and storage at each customer count
Customers	Requests a second (average)	App servers	Database	Data out (GB a month)
1,000	5.79	2 (2 vCPU, 8 GB)	2 vCPU, 8 GB, 98 GB	488
2,000	11.57	2 (2 vCPU, 8 GB)	2 vCPU, 8 GB, 195 GB	977
5,000	28.94	2 (2 vCPU, 8 GB)	2 vCPU, 8 GB, 488 GB	2,441
10,000	57.87	4 (2 vCPU, 8 GB)	2 vCPU, 8 GB, 977 GB	4,883

Prices as of 2026-11-04, list prices in USD. A cloud with a gap is ranked below complete ones and never called cheapest.

Assumptions

The busiest moment is 3 times the average request rate. One app server (2 vCPU, 8 GB) is assumed to take 50 requests a second at the busiest moment, with a floor of 2 servers. These are round planning figures, not benchmarks: measure your own and use the web app estimate for exact sizes.
The database is 2 vCPU / 8 GB up to 200 requests a second at the busiest moment and 4 vCPU / 16 GB above that. Its storage is the rows per customer times customers, with a floor of 20 GB.
Everything else is the web app estimate: an HTTP load balancer, object storage for files and data out to the internet, at list prices, on-demand, before discounts, tax, support, logging and monitoring.
Each point (1x, 2x, 5x, 10x your customers) is sized and priced again from scratch. Cost per customer therefore moves in steps when a server or database tier is added, not along a straight line.
"Added per customer" is the extra monthly cost between one point and the one before it, divided by the extra customers. It is the marginal cost over that range, not for the very next customer.
A cloud with a service it does not offer, or one that cannot be sized from public information, is ranked below complete ones and never called cheapest.

What is included, and what is not

Included. App servers, an HTTP load balancer, a managed PostgreSQL database, object storage for files and data out to the internet. This is the same model as the web app estimate, sized from customers instead of servers.
Not included. Support plans, logging and monitoring, email and SMS, payment fees, and your own staff. These are often the largest part of a real cost per customer, so add them to your own bill in the bill-based calculator.
Round sizing. How many requests a server takes depends on your code. The figures used are listed under the assumptions and are meant for planning, not capacity tests.

Prices come from the vendors' own price data, as on the other estimates; see how it works. The sizing has tests. For the method, readcloud cost per customer: unit economics for engineers. This page is English only for now.