MongoDB Atlas vs self-hosted: the real cost of managing it yourself

The self-hosting argument for MongoDB is usually framed as a large saving. It is worth running the numbers, because at small and medium scale the saving is modest and easily consumed by operational time, while at large scale the picture genuinely reverses.

Atlas pricing, tier by tier

Atlas hourly rates cover the full three-node replica set, a baseline of storage, continuous monitoring, and automated failover. Backups are billed separately: continuous cloud backup runs about $1.00 per GB-month for the snapshot store plus download charges, which on a 200 GB database with 30 day retention can add $150 to $250 per month depending on change rate. That backup line is the one most often missing from comparisons.

The self-hosted bill

So the comparable Atlas configuration at roughly $394 plus backup against roughly $300 self-hosted. Add Atlas continuous backup at $150 and the gap widens to about $244 per month, or $2,928 per year. That sounds worth chasing until you price the work.

The labour line

A self-hosted replica set needs: version upgrades (MongoDB ships a major release roughly annually and supports each for a few years), OS patching on three hosts, TLS certificate rotation, backup verification with periodic restore drills, oplog sizing, index build coordination, failover testing, and an on-call rotation that knows how to resolve a stuck primary election at 3am. Call it eight hours a month in steady state and a multi-day project once or twice a year.

At a fully loaded engineering cost of $200,000 per year, roughly $96 per hour, eight hours a month is $768. The $244 infrastructure saving does not come close. The break-even point sits at roughly 2.5 hours of operational attention per month, which is not a realistic steady state for a production replica set with an on-call obligation attached.

Where self-hosting genuinely wins

The economics invert at scale, for three reasons. First, Atlas tiers are fixed shapes; if your workload needs 128 GB of RAM and two vCPUs of compute, you buy an M80 and pay for CPU you do not use, whereas self-hosting lets you pick an r6i or x2gd shape matched to the actual ratio. Second, commitment discounts: three-year Reserved Instances or Savings Plans take 50 to 60 percent off EC2, a discount Atlas reserved capacity does not match at equivalent terms. Third, at high node counts the fixed operational cost amortizes: eight hours a month covers three nodes or thirty nodes at roughly similar effort with good automation.

A concrete inversion point: at M60 scale, about $2,884 per month, the equivalent self-hosted setup on three r6i.2xlarge instances is about $1,105 on-demand and roughly $500 with a three-year commitment. Even adding $1,000 per month of engineering time, self-hosting saves over $1,300 per month. Somewhere between M40 and M60, which is roughly the $800 to $2,000 per month range, the decision genuinely flips for a team that already has database operations capability.

The factors that are not money

Atlas gives you multi-cloud and multi-region deployment with a configuration change, online archive to cheaper storage, and a support relationship with the people who wrote the database. Self-hosting gives you control over the version, the storage engine configuration, the exact instance shape, and the ability to use Spot or Reserved capacity you already own. Neither of those columns converts cleanly to dollars, and both should weigh on the decision.

Whichever path you take, model it from Terraform before you commit so the storage copies, cross-AZ transfer, and backup retention appear in the estimate. Compare instance families against the resource catalog, and see self-hosted versus managed economics for the relational equivalent.

FAQ

What to do next

Cost the whole topology, not just the instances. C3X reads your Terraform and prices your resources against a live catalog. Start with the quickstart.