Arena: Market Conditions Before MongoDB
Web application developers · 2007 · Worldwide
Changing technical requirements
Web developers in 2007 wrote applications in PHP, Ruby, Python and Java, where a record such as a user profile was an object with nested parts: several addresses, a list of tags. Relational databases, designed decades earlier for accounting and banking, stored each object as rows spread across tables. Developers wrote or adopted object-relational mapping layers to translate between the two, joined tables back together on every read, and changed the database schema whenever the application changed. When a site grew, they split data across many databases by hand. Relational systems remained the right choice where one operation had to update several records at once, as in banking MDB3 MDB2.
How each step happened
Step 1 of 5 · 2007–09
Document model with drivers for every language
MongoDB began as one part of a bigger product. Its founders had run into the limits of relational databases while scaling DoubleClick's ad serving, and in 2007 their company, 10gen, set out to build an open-source platform-as-a-service with its own data layer MDB2 MDX-1. When Google released App Engine as a beta at about the same time, Dwight Merriman says the team threw the application layer away and kept the database MDX-1. Developers got one piece they could drop into any stack instead of a platform they had to move onto.
That piece had to fit how developers already wrote code. Eliot Horowitz described documents as close to the objects PHP, Ruby and Python programmers used: a user profile with an array of addresses is stored as one record, with no joins and no mapping layer, and a new field can be used without migrating the database MDB3. So the first job after the pivot was a driver, the library a program uses to talk to a database, for every popular language MDX-1. 10gen wrote the official drivers itself; by 2017 a dedicated team kept drivers in more than ten languages MDB3 MDX-2. The price was relational guarantees: MongoDB did not support transactions across several records, which left banking-style workloads to SQL MDB3. What it gained was a database any web developer could use in 2009, in their own language.
Rivals RethinkDB optimized for correctness, simplicity and consistency; it declared version 2.0 production-ready only in April 2015, and its founder later judged it three years behind the market RE1 RE2. MySQL and other relational databases kept the transactions MongoDB gave up MDB3.
Developer-First StrategyEase of UseFounder Domain Expertise
Step 2 of 5 · 2009–17
Free open source, seeded city by city
MongoDB gave the database away. Dev Ittycheria describes open source as a freemium strategy, a free product that leads to paid ones: easy for developers to try and quick to spread, while the AGPL licence limited others from selling the software as their own MDX-1 MDX-2. The free version, Community Server, needed no purchase.
Distribution went through the communities the drivers served. Merriman's team demonstrated MongoDB from Python at the Python meetup in one city, then at the same group in ten cities, for as many languages as it could, then ran its own conferences city by city MDX-1. Downloads rose from 2,563 in October 2009 to more than 30 million by 2017 MDX-3 MDX-2.
Paid business followed use. Many enterprise customers first met MongoDB through Community Server, and its developers became proponents inside their companies MDX-2. Chip Hazard, on the board from 2009, describes grassroots developer adoption turning into paying customers MDX-3. Accounts then grew with the applications: net ARR expansion, the growth in annual recurring revenue from existing customers, was over 120% for ten straight quarters, and the 2013 customer group spent 4.1 times as much four years later MDX-2 MDX-3.
Rivals RethinkDB was free and open source too; its founder later called open-source developer tools 'one of the worst markets one could possibly end up in' RE1 RE8. PostgreSQL and MySQL offered developers the same free start MDB1.
Freemium Model
Step 3 of 5 · 2013–17
The default document API for developers' code and skills
This step is the sum of the first two, as measured when MongoDB went public. DB-Engines ranked it the most popular modern database from 2013; more developers wanted to work with it than with any other database in Stack Overflow's 2017 survey; and more developers listed MongoDB as a LinkedIn skill than nearly all other non-relational databases combined MDX-2. MongoDB University, its free training, had over 730,000 course registrations MDX-2.
Popularity became a standard. Every application written against MongoDB's drivers and query language expected a MongoDB server at the other end, and every trained developer carried that expectation to the next project. For a new document database the question changed from which data model was better to whether it could work with code already written for MongoDB. The next two steps follow from that: MongoDB could sell the same interface as a service, and rivals had to copy it.
That shared API became more valuable with each application, driver and developer added to it, because each made the next project written against it easier to staff and support. Compatibility with it became what rivals sold: Microsoft added MongoDB driver support to its own document database in 2016, and AWS launched a service in 2019 that accepted 'the same MongoDB application code, drivers, and tools' MDX-8 MDX-5.
Rivals At RethinkDB, everyone asked how it differed from MongoDB, and its founder says hosted MongoDB services had one to two orders of magnitude more users than RethinkDB RE1.
Protocol network effects
Step 4 of 5 · 2016–
Atlas runs that API as a managed service
Free distribution left a gap. Many customers ran MongoDB on cloud infrastructure themselves, and once they were operating it, Ittycheria says, they had no ongoing relationship with MongoDB MDX-1. In December 2015 the board debated only whether to build a hosted service now or later; Atlas launched six months later, on June 28, 2016 MDX-1. Atlas runs the database for the customer, handling provisioning, availability, upgrades and scaling on the major public clouds MDB1.
Because Atlas ran the interface developers already used, it could turn Community Server users who would never buy the on-premises product into paying customers MDX-2. Pricing followed usage, so spending grew as a workload grew MDB1. MongoDB took about ten years to reach $100 million in revenue and five more to reach $1 billion, much of it from Atlas MDX-1. By fiscal 2025 Atlas was a majority of revenue, about 70% on Roelof Botha's figure MDB1 MDX-1. Customers then asked for adjacent services such as search, and Atlas added them MDX-1 MDB2.
Rivals RethinkDB decided plain hosting would not make a dent, and its Horizon Cloud was still a private beta in May 2016, months before the company shut down that October RE1 RE3 RE4. AWS and other cloud providers sell their own managed databases alongside the infrastructure Atlas runs on MDB1.
Managed ServiceMulti-Product
Step 5 of 5 · 2016–
Switching costs of MongoDB-API applications, fenced by the SSPL
The standard from step 3 is what holds customers, and it showed from 2016: rivals that wanted MongoDB workloads built MongoDB's interface rather than their own. Microsoft added MongoDB driver support to Azure DocumentDB in 2016 so developers could reuse their MongoDB skills and tools, and AWS launched Amazon DocumentDB in 2019 so they could keep 'the same MongoDB application code, drivers, and tools' MDX-8 MDX-5. In 2025 Microsoft, AWS and Google backed an open-source MongoDB-compatible API on PostgreSQL at the Linux Foundation MDB4. That choice shows the switching cost: a customer's applications are written to that API, and moving to anything else means rewriting them.
A licence fenced the copies. Merriman says cloud providers offering Atlas-like services built on MongoDB's own code would have been 'game over' MDX-1. In October 2018 MongoDB moved Community Server to the Server Side Public License (SSPL), which requires anyone offering MongoDB as a service to open-source the software behind that service MDX-4. AWS built DocumentDB from scratch to the older 3.6 API, with only the most widely used features at launch; Horowitz called it an imitation of 'the MongoDB code from two years ago' MDX-6 MDX-7. Ittycheria says Atlas only grew faster after the change MDX-1. The licence protects the current server, and Atlas, the service from step 4, is where customers get all of it.
Rivals PostgreSQL now stores JSON documents, and compatible APIs lower the cost of leaving MongoDB MDB4. RethinkDB's own installed base moved away: signageOS replaced it with MongoDB in 2019 RE9.
Switching costsIntellectual property