Say "MongoDB in the cloud" and most people mean Atlas. It runs on all three major public clouds, from a free starter cluster up through large dedicated tiers, and now there's Infinite, a version that just entered public preview. This post goes through it in order — clouds and regions, tiers, elasticity, backup and security, the new edition — and finishes with how to choose between a managed service and running it yourself.
Storage limits per tier are from the Atlas docs (AWS deployments); the Infinite figure is the logical storage limit per replica set
Bar length is on a logarithmic scale so everything from 0.5 GB to 128 TB fits in one chart
Self-hosting MongoDB means handling high availability, scaling, backup and security yourself. A managed cloud service turns those into options you configure.
The service builds and maintains the replica set. In supported regions, a cluster can be spread across three availability zones.
Move to a bigger tier, or turn on auto-scaling and let capacity follow load — instead of migrating data in the middle of the night.
Cloud backups, private network connectivity and disk encryption are part of the service, not components you assemble yourself.
Atlas supports AWS, Google Cloud and Microsoft Azure. On AWS alone, the docs list 34 regions across the Americas, Asia Pacific, Europe, and the Middle East and Africa.
Not every region supports every tier: the docs say all listed regions support dedicated M10+ clusters, but some don't support free or Flex clusters. AWS regions in China and GovCloud are handled separately (government regions are covered by MongoDB Atlas for Government). Check the latest region table before choosing.
Atlas cluster tiers fall into three broad bands: the shared starter tier, lightweight on-demand Flex, and dedicated clusters starting at M10.
Dedicated tiers also come in Low-CPU, NVMe and Gen2 variants; on AWS, Gen2 applies to M30 and above. The storage and memory figures above come from the Atlas AWS docs page. Tier tables change between releases, so rely on the latest official page when sizing.
Atlas auto-scaling works in two directions, compute and storage, and the compute side has two mechanisms: reactive and predictive.
Adjusts the tier based on the cluster's current resource usage, and applies to every eligible cluster.
Uses historical usage patterns to scale up ahead of an expected peak. For spikes with no cyclical pattern, it falls back to reactive scaling.
On Atlas Core, storage increases when the cluster reaches 90% of its disk capacity (provided it's configured or you haven't opted out).
On Atlas Core, an auto-scaling event waits until the oplog window covers the estimated scaling time. If you don't want the window to block scaling, you can set a minimum oplog retention window.
Backup and security are the two things people ask about most when evaluating a managed service. Here is how Atlas approaches both on AWS.
mongodump / mongorestore insteadWhen restoring a backup, the target cluster's version has to fit the rules: the same major version with an equal or higher minor version, or the next higher major version. Writes to the cluster aren't allowed while a restore is running — factor that into your downtime window when you rehearse restores.
Atlas is more than a hosted database — several commonly needed capabilities come as part of the service.
Embedded full-text search. Build indexes right on the cluster and query with the $search and $searchMeta aggregation stages, with autocomplete, pagination and faceting. No separate search system to run next to your database.
Vector search for AI applications, positioned officially as a way to give AI apps accurate context. See its dedicated documentation for usage and limits.
The docs also mention Online Archive, Data Federation, Stream Processing, the SQL Interface and more, to combine as needed. What's available depends on the cluster type.
Both use the same interface, drivers and APIs, so existing queries and connection code keep working. The difference is how compute and storage are organized.
MongoDB's pages cite scaling time cut by more than 96% and higher throughput per dollar for Infinite. Those numbers come from MongoDB's internal tests and haven't been independently verified. The preview also has a long list of unsupported items: multi-region and multi-cloud clusters, Atlas Search and Vector Search, Online Archive, the database profiler, an uptime SLA and more. Check each one before using it in production.
A managed service saves effort, but it isn't right for every situation. Here is a rough way to think about it.
If you choose to run MongoDB on your own EC2 instances or in your own data center, the deployment, monitoring, alerting and backup work that a managed service would handle is yours to cover. Whaleal Platform is built for exactly this self-managed scenario, and it can be launched directly in your own AWS account through AWS Marketplace.
A cloud service turns operations into settings, but the responsibility for capacity, backup and security doesn't disappear — it just changes shape. See the whole ladder, and you can pick the right rung.