The problem: seventeen kinds of database
MongoDB starts from a cost most engineering leaders recognize. Every application needs a database, and the company notes that a large organization can run tens of thousands of applications and the databases behind them. Choosing one is a strategic decision because it shapes how fast developers can work and how well the application performs. Over the years teams also added a separate system for search and now another for AI vectors, and every one of them has to be learned, synchronized, secured and paid for.
There’s no customer on this planet that wants to have 17 different databases.
Dev Ittycheria · Then President and CEO · Interview on reaching ten years as CEO · 2024-04-28
The company's answer has two parts. The first is the document model: each record is stored as one flexible document, close to the JSON that applications already pass around, so a team can add a field or a new kind of interaction without redesigning tables. The second is to put search, vector search and embeddings inside the same platform as the live data. In 2024 Ittycheria called standalone vector databases features masking as products, and in September 2026 his successor told investors that these capabilities are built in, not bolted on, so agents connect to one platform instead of many.
Why the bet looks stronger in 2026
Three changes made the bet more valuable. The first is the cloud. Atlas, the service in which MongoDB runs the database for the customer, launched in June 2016, was 2% of revenue at the 2017 IPO and 73% in the year ended 31 January 2026. MongoDB now earns most of its money operating the database, so it is paid when customers' applications grow.
The second is AI. Ittycheria argued in November 2025 that more software means more databases, and that a JSON database suits language models that read and write JSON. The company bought Voyage AI, a maker of embedding models, in February 2025 for $160.9M, and by March 2026 the number of customers using vector search had nearly doubled in a year.
Architecturally, we are well positioned for the AI era just by being a native JSON database, LLMs emit and consume JSON, MCP is built on JSON.
Dev Ittycheria · Then President and CEO · Investor call on his planned retirement, as reported · 2025-11-03
The third is migration. At its September 2026 Investor Day the company said AI tooling has changed the modernization process, so that Oracle, SQL and other relational workloads can move to MongoDB at record pace. It sizes that market at $10B, its own estimate, and sells it through two channels: directly to existing customers and through partners into new markets.
Following one application change through the platform
Take an illustrative case: a bank wants to add an AI assistant to an existing customer application. First the developers model the data, deciding which fields belong together in each document by how the application reads and writes, rather than copying the old tables one for one. Next platform and security engineers choose an Atlas deployment on the cloud the bank already uses and prepare access. Then the team indexes the same documents for full-text and vector search, so the assistant can find a customer's history by meaning without a second synchronized system.
If the application still lives on an old relational database, a migration comes first, and that is where people matter most. MongoDB and Cognition say their tools now automate much of the code work, but engineering teams keep the decisions that need human judgement: designing the target data model and sequencing the cutover. Before delivery starts, the customer and MongoDB agree a scope and its assumptions, so the project does not begin on a misunderstanding.
The pitch is simplification, and independent analysts accept part of it: fewer systems to stitch together means less code to maintain. They also note the cost. A unified platform can become lock-in, and native AI features help most when MongoDB is already the customer's main data store. Expect Korean buyers to raise both points.
Whichever Seoul seat you take, the daily work is to make this belief concrete for one Korean customer at a time: which application moves first, what it will cost, who will deliver it and how the result will be measured.
