How much does it cost to run a SaaS app with 100,000 users?

At 1,000 users a SaaS bill is a flat spread of small components. At 100,000 users it is a different animal: a few lines dominate, the fixed plumbing becomes a rounding error, and categories that were invisible at small scale (observability, cross-AZ traffic, request fees) turn into four-figure monthly charges. Here is what the bill actually looks like.

The workload we are pricing

Assume a multi-tenant B2B SaaS in us-east-1 with 100,000 users, about 8,000 concurrent at weekday peak. Traffic is roughly 250 million API requests a month, bursty with a 4x peak-to-trough ratio. The primary database is a 2 TB Aurora PostgreSQL cluster with one writer and one reader. There is 20 TB of tenant documents in object storage, 12 TB of CDN egress a month, a search index over tenant data, and an asynchronous worker tier handling exports, webhooks, and scheduled jobs. Everything spans three availability zones.

The monthly breakdown

Observability costs more than the application

The line that stops people is 1,060 dollars for logs and metrics, against 605 dollars for the compute running the product. Two terabytes of ingestion a month sounds like a lot until you do the arithmetic: 250 million requests emitting a single structured access log of roughly 1.5 KB is 375 GB before anyone writes a debug statement. Add application logs, load balancer logs, VPC flow logs, and database logs, and 2 TB is conservative rather than extravagant.

The fix is not to log less in an emergency but to decide up front what tier each log stream belongs to. Sample successful request logs at 5 percent and keep errors at 100 percent, ship VPC flow logs to object storage rather than the log service, and set retention to 7 days for hot logs with an archive tier behind it. That routinely cuts the line by 60 to 70 percent, worth 600 to 700 dollars a month here.

The CDN tiering effect

Twelve terabytes of CDN egress is not billed at one rate. The first 10 TB bills at 0.085 per gigabyte (870.40 dollars) and the next 2 TB at 0.080 (163.84 dollars), plus about 150 dollars of HTTPS request charges. This tiering is why egress cost per user falls as you grow, and why moving egress from direct object-storage transfer at 0.09 per gigabyte to a CDN is worth doing even for API responses. The detail is inCloudFront cost optimization anddata transfer out explained.

Unit economics against the small-scale case

Cost per user falls by two thirds between 1,000 and 100,000 users. The reason is that the components that scale linearly (object storage, CDN egress, request fees) are only about 40 percent of the bill, while the rest scales in steps or barely at all. A load balancer serving 8,000 concurrent users costs little more than one serving 150. This is the mechanical reason gross margin improves with scale in SaaS, and it is worth measuring as cost per tenant rather than a company-wide average, because a handful of heavy tenants usually distort the mean.

The three levers worth pulling

First, commitment discounts. The app tier, database, cache, and search cluster are all steady, and a one-year Savings Plan or reserved instance coverage on all four is worth roughly 900 dollars a month against the on-demand equivalent. Second, storage lifecycle. Twenty terabytes of tenant documents at 0.023 per gigabyte includes plenty of data nobody has opened in a year; moving cold objects to an infrequent-access or archive tier via a lifecycle policy can halve that 531 dollar line. Third, observability discipline, as above.

Together those take the bill from 6,540 to roughly 4,700 dollars a month without touching the architecture, which is about 0.047 dollars per user. Price the whole stack from Terraform against theresource catalog before the next capacity increase, and the growth curve stops being a surprise.

FAQ

What to do next

Forecast your SaaS bill before it grows. C3X reads your Terraform and prices your resources against a live catalog. Start with the quickstart.