Every application needs a database, and the developer who builds it usually picks one. MongoDB's case to that developer is the document model: data stored as JSON-like documents shaped the way the code already uses it, with nested fields and arrays, no fixed schema, and no joins on the common path. A user profile, an order or a product listing is one document, read and written in one call. The database scales out by sharding collections across many servers and fails over automatically, so the same model works from a prototype to a global application.

It reaches developers for free first. Community Server has been downloaded over 700 million times, and Atlas has a free tier. Usage turns into revenue when a company runs an application in production: on MongoDB Atlas, the managed service, which runs in more than 130 regions on AWS, Google Cloud and Azure and is now most of the revenue, or on Enterprise Advanced, the commercial edition a company runs itself.

Atlas has grown into a platform around the database: full-text search, vector search for AI retrieval, stream processing, archival, and since the 2025 purchase of Voyage AI, embedding and reranking models. The idea is that an AI application can keep its business data, its embeddings and its search in one system instead of three.

Customers range from startups on the free tier to the largest banks, retailers and telecoms, over 65,200 paying in all. They use it for applications where data is nested, changes often or must scale across regions: catalogs, user data, payments, content, games, telemetry, and now AI features. MongoDB competes with the relational giants (Oracle, Microsoft SQL Server, IBM), with the clouds' own databases (DynamoDB, Cosmos DB, the MongoDB-compatible DocumentDB), and above all with PostgreSQL, the free default that many new projects choose. Chirantan Desai became chief executive in 2025.

A database is the hardest part of an application to change. Every query, index and data model in a MongoDB application is written for MongoDB's API and document structure; moving to PostgreSQL means redesigning the data as tables, rewriting the data access layer, migrating live data without downtime and retesting everything. Few teams do that to a working application, so a MongoDB application usually stays on MongoDB for its whole life.

The managed service adds to it. A team on Atlas has handed backups, scaling, upgrades and failover to MongoDB, and often its search and vector search too; leaving Atlas means rebuilding that operations work or finding another managed service.

The weak points are where a customer can leave without rewriting. The clouds' MongoDB-compatible services (Amazon DocumentDB, Azure Cosmos DB for MongoDB) speak enough of the API to take simpler workloads, often at a lower price and on the customer's existing cloud bill. A self-managed customer can move to the free Community Server or Percona. And the stickiness is per application: a company with ten applications on MongoDB can still build its eleventh on PostgreSQL. MongoDB's growth depends on winning those new applications, which is why it spends so much on reaching developers.

MongoDB's years end in January, so fiscal 2026 is the year to January 31, 2026.

Revenue was $2.46B in fiscal 2026, up 22.8%, after 31.1% in fiscal 2024 and 19.2% in fiscal 2025. Over the five years to fiscal 2026 it grew about 33.1% a year, from $114.8M in fiscal 2017. Because Atlas is billed by use, growth follows how much customers' applications are used, not only how many contracts are signed.

The shift to Atlas shows in the gross margin: 71.7% in fiscal 2026, down from 74.8% in fiscal 2024, because Atlas pays AWS, Google Cloud and Azure for the servers it runs on, where a self-managed licence costs almost nothing to deliver.

The operating margin has narrowed steadily, from negative 35.5% in fiscal 2021 and negative 27.0% in fiscal 2023 to negative 5.6% in fiscal 2026, as revenue grew faster than sales, marketing and research costs. The net loss was $71.2M.

It needs little capital of its own, since the clouds own the hardware: capital spending was $5.0M in fiscal 2026. Operating cash flow was $505.1M and free cash flow $500.2M, 20.3% of revenue. Equity pay is its largest single cost after salaries: $550.5M, 22.3% of revenue, and after it free cash flow was negative $50.3M. MongoDB ended fiscal 2026 with $1.08B of cash and $1.30B of short-term investments, and no convertible notes left on its balance sheet.

MongoDB is aiming to be the database for AI applications, the way it became a default database for web and mobile applications.

The argument is that AI applications need exactly what the document model handles well: messy, varied data (text, metadata, embeddings, model outputs) next to live operational records, retrieved by meaning and by filters in one query. Atlas Vector Search, the Voyage AI models, Automated Embedding and hybrid search that combines text and vector results are the pieces, and MongoDB markets them to AI-native startups and to enterprises through partners that include the three clouds, consultancies and model providers.

The second path is modernization: moving old applications off relational databases, which MongoDB is attacking with migration tools and AI-assisted code conversion, since the cost of rewriting is what has kept those applications where they are. A third is broadening Atlas into more of the data stack, stream processing and search among them, so a team needs fewer separate systems.

Underneath, the model stays the same: developers try it for free, applications go into production on Atlas, and usage grows with them. With operating losses narrowing toward break-even, the business's own figures depend on keeping Atlas growing while its cost of cloud servers keeps the gross margin lower than a licence business would have.

• PostgreSQL. The free, open-source default has added JSON, vector search (pgvector) and serverless hosting (Neon, Supabase), and many new projects, including AI ones, start there instead.
• The clouds. AWS, Google and Microsoft are MongoDB's hosts, partners and competitors at once, offering their own document databases and MongoDB-compatible services on the same bill.
• Winning AI applications. Dedicated vector databases, Postgres and the clouds all compete for the retrieval layer, and the AI tools developers use change every few months.
• Usage-based revenue. Atlas revenue rises and falls with customers' application usage, so a slowdown at customers, or a customer tuning its queries to cut its bill, shows up quickly.
• Legacy relational applications. The largest pool of databases is still on Oracle, SQL Server and Db2, and rewriting those applications for documents is slow even with AI-assisted tools.
• Gross margin. Running on other companies' clouds means paying them for every server, which keeps the gross margin below that of a licence business.
• Equity pay at 22.3% of revenue in fiscal 2026 turns positive free cash flow negative once it is counted.