A product leader asks what it would cost to run the platform in a second region for resilience. Your single-region cloud spend is about $80k a month. What is the honest order of magnitude?

The arithmetic, shown

Start with what actually gets duplicated. Compute and storage double, because a region that can serve traffic must be able to serve it. That is the \(80k becoming roughly \)160k before anything new appears.

Then add the costs that exist only because there are two regions:

• Cross-region data transfer, priced at around \(0.02 per GB in 2024-2025 for many provider-to-provider region pairs, which at a sustained 50 MB/s of replication is roughly 130 TB a month and on the order of \)2,600.
• Database replication licensing or managed-service tier changes, frequently a step change rather than a percentage.
• A second set of everything non-obvious: observability ingest, secrets management, CI runners, network appliances.

So the infrastructure answer is a little over 2×.

Why the other options fail

• About 10% more describes a standby that is not a standby: a small pilot-light footprint that cannot take traffic. It is a real and useful option, and it is not what was asked for, and quoting it invites the expectation of failover that the footprint cannot deliver.
• About 30% with no engineering work assumes the data services do not need duplicating, which is the expensive half, and that replication and failover come free with the managed tier. They come with a higher tier and a configuration project.
• Nothing extra because the cloud handles it confuses multi-zone defaults with cross-region resilience. No major provider gives you cross-region failover automatically, and believing otherwise is the assumption that turns a regional outage into a surprise.

The part that dominates and is not on the invoice

The engineering programme. Data has to be replicated with a defined RPO, routing has to fail over, sessions and caches have to be regional, schema migrations have to be safe across versions in two places, and something has to resolve conflicting writes if both regions accept them. Plan in quarters and in people, not in cloud line items, and expect the ongoing tax: every future feature now ships to two regions and every migration is a distributed one.

Which assumption dominates the error

Whether the second region is active or standby. A warm standby at reduced capacity is materially cheaper, perhaps 1.4× to 1.6×, and it buys a longer recovery time and carries the risk that the untested path does not work when needed. An active-active pair costs more and is the only version that is continuously proven. Ask which one is being requested before quoting a number, because they differ by more than the number does.

What the number rules in or out

At \(80k a month, **the question worth putting back to the product leader is what outage duration they are buying protection against, and what it is worth.** If the answer is a regional cloud failure lasting hours, once every couple of years, the annual cost of the insurance is around \)1M in infrastructure plus the programme, and that is a business decision, not an engineering one.

The cheaper alternatives deserve a sentence before anyone signs: multi-zone within one region, which most workloads should already have and which addresses the far more common failure; backups with a tested restore into a second region, which converts hours of downtime into a day at a fraction of the cost; and a static degraded mode that keeps the product partially usable. Multi-region is the right answer when the business cannot tolerate hours, and an expensive one when it merely dislikes them.

When this is the wrong answer

If the last three incidents were caused by deploys, dependencies and database migrations rather than by infrastructure failures, a second region multiplies the surface that produced them while addressing none of them. The honest conversation names the actual incident history before it prices the insurance.