Multi-AZ and high availability: what it adds to the bill

Rohit Athaide · 2026年11月3日 公開

A cloud region is made of several availability zones: separate data centers close enough to talk quickly, far enough apart that one fire or power cut does not take down both. Running across more than one zone is how you survive the loss of one. It costs more, and the extra is easy to underestimate.

What high availability adds

• More compute. A second server in a second zone doubles the compute for that tier, whether it handles traffic or waits as a standby.
• A standby database. A managed database with a standby in another zone charges for both instances. In the AWS PostgreSQL rows of our databases catalog, the Multi-AZ price is about twice the Single-AZ price for the same size, because you pay for two. Storage is billed for both copies as well.
• A load balancer to spread traffic and detect failures. See load balancer costs.
• Data transfer between zones. Replication and cross-zone traffic is often billed per gigabyte, even within a region.
• Extra capacity headroom. If one of two zones fails, the other must carry the full load, so each should be able to.

High availability is not backup

A standby protects against a failed machine or zone. It does not protect against a deleted table, a bad deployment or a corrupted dataset, because the mistake is copied to the standby. You still need backups. See backup and retention costs.

Disaster recovery is a different step up

Surviving the loss of a whole region means a second region, with data copied across and capacity ready. That usually adds more than a second zone does, because inter-region transfer is billed and a warm standby runs all the time. The common patterns, from cheapest to most expensive, are backup and restore, a minimal “pilot light” copy, a warm standby at reduced size, and an active copy in each region.

How much do you need

Ask what an hour of downtime costs, and what you can accept.

1. Internal tools and development: a single zone is usually fine. Restore from backup if it fails.
2. A customer-facing app with modest stakes: two zones for the stateless tier, and a managed database with a standby.
3. A service with contractual availability promises: two zones for everything, tested failover, and a recovery plan for a region loss.

Pay for the level your customers and contracts actually need. Both over- and under-building are costly.

Pricing it in an estimate

Our estimates model a typical single deployment. To see the effect of a standby, price the database in the catalog for both deployment types and compare. For the whole app, run the web app estimator with your server count and add a standby by hand where it applies.

The managed-versus-self-run trade is in managed database versus self-hosted.