MongoDB Grows 30% in the Quarter, Launches New Voyage AI, Yet Loses 14% on the Stock Market

Revenue Beat, Multiple Punishment

MongoDB reported revenue of $772 million in the second fiscal quarter of 2027, up 30% year-over-year, the fastest growth rate since fiscal year 2024. Atlas, the managed database offering that accounts for the bulk of revenue, accelerated to 29%, and the Net ARR Expansion Rate rose to 122%, up from 119% a year earlier. RPO grew by 91%, reaching $1.52 billion. The active Atlas customer base closed the quarter at 69,300, compared to 58,500 in the same period in 2025, with the company adding over 5,000 customers alone in the first half, a historical record.

Nevertheless, shares dropped approximately 13.7% during the session on September 1. The reason was not the past; rather, it was the future. The revenue guidance for the third quarter was set at $756 to $761 million, reflecting a 20% to 21% increase, which is a significant deceleration compared to the reported quarter. The annual revision was elevated to $2.99 to $3.03 billion (21% to 23%), but this does not offset the perception that the acceleration curve has already peaked.

Voyage AI Becomes a Product, Not Just Embedding

The most strategic aspect of the quarter is not in the revenue table. MongoDB announced the next generation of Voyage AI models, acquired in early 2025: the voyage-context-3, a contextual embedding that preserves the context of the entire document without requiring manual enrichment of metadata, and the rerank-2.5, the first market reranker capable of following instructions. According to benchmarks published by Voyage itself, the voyage-context-3 outperforms OpenAI text-embedding-3-large by 14.24% in chunk retrieval and 12.56% in document retrieval, and Cohere Embed v4 by 7.89% and 5.64%. The rerank-2.5 is 7.94% more accurate than Cohere Reranker v3.5 and operates with a 32,000 token window, eight times that of its competitor.

What this changes in the CIO's contract is the axis of the conversation around RAG. Until now, those building a retrieval pipeline decided separately on the database, embedding, and reranker. MongoDB is pushing all three into the same SKU, charging per token consumed in the embedding service rather than per allocated instance. The first 200 million tokens in the voyage-context-3 are free. This is the same strategy that Snowflake employed when packaging Cortex Search and Databricks did when integrating Mosaic Retrieval.

What the 14% Decline Is Indicating

The selling thesis is well-known. MongoDB is the backbone of operational data in thousands of AI applications in production, and the token curve of Voyage is growing faster than Atlas's instance curve. The opposing thesis, defended by part of the sell-side, is different. First, the consumption of Atlas has undergone a period of optimization in large accounts, and organic expansion has slowed in the commercial segment. Second, revenue from embedding and reranker is still small and faces structural price compression: OpenAI, Cohere, Voyage, and Google compete for fractions of a cent per token, and the incremental margin for the database seller is close to zero outside of data lock-in.

The third factor is regulatory. With the AI Act and the DSA in enforcement in Europe since August, large clients have begun to demand contracts with embedding portability clauses. This dynamic reduces the lock-in that MongoDB assumed as strategic collateral from the Voyage acquisition.

What Data Leaders Should Monitor Until December

Three indicators count towards the next story. The first is the conversion rate of Atlas accounts into contracts with Voyage AI: if it crosses 25% in the next quarter, the platform thesis stands. The second is the average contract length: if it increases from 24 to 36 months, the model behaves more like infrastructure software, not pure consumption. The third is the partner base: the expansion of the partnership with LangChain, announced alongside the results, needs to produce real references from substantial clients, not just technical integration.

In the United States, large banks and insurance companies are already operating RAG pipelines in production with multiple vendors, and the topic in RFPs this year is standardization by platform versus best-of-breed in each layer. In Germany and the UK, industrial and financial companies are demanding embedding portability as a contractual condition, aligned with the enforcement of the AI Act. In India, TCS, Infosys, and Wipro are pushing Voyage AI integrations in managed offerings for global clients. In Brazil, Itaú, Bradesco, and Nubank are evaluating Voyage AI for assisted consultation of regulatory documents, and the choice between a single platform (MongoDB plus Voyage) and a composed stack with Pinecone, Cohere, or OpenAI determines the real cost of the next wave of corporate RAG.