Q2 FY2027 is the quarter in which a filing, not the call, puts the first AI dollar on MongoDB’s income statement, and it is a cost. The 10-Q attributes a year-over-year increase in research and development software costs to increased use of AI tools, of that line’s increase, and the six-month comparison does the same for the half year. The call does not mention it. That makes AI tools the one channel here sized by the company; no saving is claimed against it, and the CFO credits revenue for the margin.
On revenue the words move again and stay words. Revenue was , up ; Atlas-related revenue , up ; Enterprise Advanced and other , up . The CFO says the company has started to see some benefit from AI and that it is small. Voyage customers roughly doubled for a second quarter, a majority of the new ones AI natives sent by coding agents; one frontier lab moved more inference-side workloads to Atlas; a record customers were added. The ledger’s sizes are still its own: enterprise AI workloads at . The AI-native cohort, measured this quarter only by counts of new customers and the word small, is left unsized.
A new priced channel opened: Search and Vector Search on the self-managed product, delivered 2026-06-30 and charged as an extra, which the CFO credits in part for the line’s growth and for raising its full-year outlook to about . The ledger leaves it unsized, since the CFO names the product's growing strategic importance beside the launch and the charge covers full-text search with vector search, and leaves the wider claim, that AI readiness is behind self-managed demand, unsized too: AI is named beside resilience, sovereignty and capacity, and the 10-Q explains the increase without naming AI.
Steps from Q1 FY2027: the AI tools bill moves from described to quantified and from inferred to reported; frontier lab revenue and self-managed demand attributed to AI move from described to directional; the add-on channel opens, unsized; the AI-native cohort is unsized on counts. Management describes, for the first time in its own words, AI natives that start on other databases through prompt-driven platforms and migrate later, the mechanism of the toll the 10-K named. Every revenue size in this exhibit remains the ledger’s inference.
Sized channels against the income statement, Q3 2026
9 of 15 channels sized
Each blue mark is one channel's dollars for the quarter; a bar is the range of an estimate. Grey marks are the company's reported lines. The distance between them is the point: how large the AI channel is next to the line it sits in.
Revenuereported line
$772mn
Total operating expensesreported line
$541mn
Sales and marketingreported line
$253mn
Research and developmentreported line
$214mn
Cost of revenuereported line
$202mn
General and administrativereported line
$74mn
Engineering spend on AI capabilitiesspend · our inference
Atlas consumption from established enterprises' AI workloads and AI-readiness modernizationrevenue in · our inference
Marketing and start-up programs aimed at AI-native companiesspend · our inference
AI tools used in research and development (software cost)spend · reported in the filing
Amortization and retention awards from the Voyage AI acquisitionspend · our inference
Engineering and operating cost displaced by the company's own use of AI toolscost displaced · our inference
Revenue from frontier model labsrevenue in · our inference
Compute for serving and training the Voyage models and AI featuresspend · our inference
Workloads lost when coding agents and prompt-driven platforms choose another databasetoll · our inference
$100k$1mn$10mn$100mn$1bn
AI channel, dollars for the quarter low to high of an estimate reported lineLog scale: each gridline is ten times the one before.
New money and old money, Q3 2026
7 new6 expanded2 relabelled
Each channel is tagged once for whether its money existed before language models, from the company's annual report and call at the start of the period. A bar splits one flow's sized dollars by that tag. The incremental total is the part that would not be there without the models: a new channel counts in full, an expanded one only for what AI added, a relabelled one at zero.
Flow, sized total
Split by novelty
Incremental total
Paid for AI$26mn to $90mn sized
new $7.7mn to $15mnexpanded $18mn to $74mn
Incremental total $13mn to $90mnpoint $17mn$32mn in 2 channels has no traced baseline
Cost displaced by AI$909k to $15mn sized
expanded $909k to $15mn
Incremental total $909k to $15mnpoint $3.6mn
Revenue arriving through AI$3.0mn to $43mn sized
new $500k to $18mnexpanded not sizedrelabelled $2.5mn to $25mn
Incremental total $500k to $18mnpoint $3.5mn
Cost imposed, or revenue lost, by others’ AI$0 to $2.8mn sized
new $0 to $2.8mn
Incremental total $0 to $2.8mnpoint $566k
The sized total counts every channel the company credits to AI, including relabelled money that existed before language models and the ledger's own estimates for it. The incremental total counts relabelled channels at zero. A flow is split by layer where its dollars sit at more than one: end use, compute sold to builders, and hardware. The same dollar can be a buyer's spend, a cloud's revenue and a chipmaker's revenue, so the layers are never added together.
Paid for AI5 channels · $26mn to $90mn sized · $13mn to $90mn incremental
other
Amortization and retention awards from the Voyage AI acquisition
0.67% to 1.1% of the quarter’s revenue
Incremental total: counts in full.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: quantified· motive: exploratory· before LLMs: expanded
The 10-Q’s intangibles note again gives the running cost: accumulated amortization of developed technology reached , more than a quarter earlier, and the release shows inside cost of revenue (claim c73). The restricted-stock table, which refers back to the annual report for the Voyage awards (claim c74), shows shares vesting in the quarter and still unvested at each, ; its six-month forfeiture count, , is the April 2026 forfeiture alone. The ledger spreads the awards still held over their remaining period, a quarter, in place of the straight-line of Q4 FY2026, which the forfeiture overtook. Neither is reported as a Voyage line or attributed to AI, so the size stays the ledger’s own, . The Voyage technology of has a two-year life from February 2025, so its amortization ends early in 2027.
Evidence: 5 quotes, 12 figures, 4 confounds, 1 from before coverage
The income-statement cost of buying Voyage AI in February 2025: amortization of its developed technology over two years, in cost of subscription revenue, and the restricted stock issued to its employees, expensed as stock-based compensation over their service period. About half of the restricted stock still unvested at 2026-01-31 was forfeited and reacquired in April 2026, so from the quarter ended 2026-04-30 the award part is sized from the shares still unvested in the 10-Q tables, not from the awards as issued. The purchase price itself, mostly paid in shares, is outside the income statement. Voyage is an AI company whose product sits in an AI channel, so its cost belongs here; the Clarity acquisition of May 2026, a federal services firm, is not an AI cost and is read as a confound. The payees are the former owners and staff of Voyage.
Why this motive
Carried from Q1 FY2027 with the product.
Before LLMs: expanded
Acquiring technology companies predates Voyage: the anchor records the 2023 purchase of Grainite, a stream processing company, for , and amortization of acquired intangible assets of in fiscal 2024. No Voyage cost was in this income statement before the acquisition closed on 2025-02-17. The size is the change AI made, not the whole line.
“On September 27, 2023, the Company acquired the assets of Grainite, Inc. (“Grainite”), for total cash consideration of $15.0 million. Grainite is a stream processing application company and the transaction is intended to accelerate the development of the Company’s stream processing offering.”
Developed technology amortization in the quarter, from the change in accumulated amortization since 2026-04-30 · 2026-CQ3
Developed technology, accumulated amortization at 2026-07-31 (negative: a contra balance) · as-of 2026-07-31
Amortization of intangible assets inside cost of revenue · 2026-CQ3
Voyage AI developed technology intangible asset, fair value at acquisition · as-of 2025-02-17
Restricted stock awards unvested at 2026-07-31, shares (all issued to Voyage AI employees) · as-of 2026-07-31
Weighted-average grant-date fair value per restricted stock award share, unvested at 2026-07-31 · as-of 2026-07-31
Grant-date value of the restricted stock unvested at 2026-07-31 · as-of 2026-07-31
Restricted stock awards forfeited and canceled, shares, cumulative for the six months ended 2026-07-31 (the same count as the three months ended 2026-04-30, so none in this quarter) · as-of 2026-07-31
Restricted stock awards vested in the quarter, shares: the change in unvested shares, with no forfeiture in the quarter · 2026-CQ3
Grant-date value of the restricted stock that vested in the quarter (a cross-check on the award expense) · 2026-CQ3
Voyage retention awards per quarter on the awards still held: restricted stock over its remaining period plus units at straight-line · 2026-CQ3
Voyage AI retention awards per quarter, straight-line over the stated service period · 2026-CQ1
Reported line it is matched to
Developed technology amortization was in the quarter by the change in the accumulated balance, and of intangible amortization sat in cost of revenue, the same as a year earlier now that the acquisition has passed its first anniversary. The filing names no AI cause.
2026-CQ3: 2026-CQ3: 2026-CQ3: 2025-CQ3:
What else could explain it
acquisition: Clarity, a federal services firm bought in the quarter for , added customer-relationship intangibles and staff; it is not an AI company and none of its cost is read here (claim c75).
line composition: Developed technology also holds a small older asset, and stock-based compensation is reported by function, not by award.
other: The 10-Q gives share counts, not the quarter’s award expense: the spread of the remaining stock over the 10-K’s remaining period is the ledger’s assumption.
other: Restricted stock units issued to Voyage employees sit in the company-wide unit table, so any forfeitures among them are not visible; the units stay at straight-line.
Quotes
“Voyage customer count nearly doubled quarter-over-quarter, and Atlas Vector Search adoption continues to outpace the growth of the rest of the company, showing our strong early momentum for AI workloads.”
“noted that MongoDB is investing heavily in AI research and strategic acquisitions to ensure its innovations consistently meet the evolving needs of AI.”
“Amortization expense of intangible assets was $3.8 million and $7.5 million during the three and six months ended July 31, 2026, and $3.8 million and $6.9 million during the three and six months ended July 31, 2025, respectively. Amortization expense for developed technology is included as cost of subscription revenue and research and development expense”
“Refer to Note 4, Business Combinations in the Notes to Consolidated Financial Statements included in Part II, Item 8 of the Company’s 2026 Form 10-K, for further details on the issuance of restricted stock awards in connection with the acquisition of Voyage AI.”
“On May 8, 2026, the Company acquired the outstanding shares of Clarity Business Holdings, LLC (“Clarity”), a services firm specializing in providing support and professional services to support modernizing legacy systems, scaling cloud-native development, and solving complex data challenges for workloads within the U.S. Government, for $16.7 million, which was accounted for as a business combination.”
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: described· motive: exploratory· before LLMs: expanded
The quarter’s launches are what the spend built: automated embeddings, new Voyage models, search and vector search for the self-managed product, a hosted MCP server (claims c11, c58). The CFO says product investment remains focused on AI and core database capabilities (claim c29). The 10-Q explains the research and development increase by of personnel cost, third-party infrastructure and AI tools, with no split of the work by purpose (claim c64). Research and development was , up . The size is the ledger’s own, , an assumed share of the line after the Voyage award expense, the AI tools cost and the compute term sized in their own channels are taken out.
Evidence: 9 quotes, 5 figures, 2 confounds, 2 from before coverage
What the company spends building its AI products: Vector Search, the Voyage models and their integration, agent memory and MCP tooling, and AI features for the self-managed product. Management lists these among the investments its margin guidance absorbs and has named a product chief for AI; the filing explains research and development by personnel cost and third-party infrastructure without splitting out AI work. The share is applied to research and development net of the three parts of that line other spend channels size (the Voyage retention-award expense, the AI tools software cost and the research and development compute term), so that no dollar is counted twice in the spend total; the base was narrowed rather than the channels marked as overlapping because each of those channels also holds dollars outside research and development. The payees are the company's own engineers and researchers.
Why this motive
Carried: AI capabilities remain an investment the margin outlook absorbs (claims c29, c28), funded in the CEO’s account by the profit of the self-managed product (claim c16): investing-ahead framing.
Before LLMs: expanded
Research and development was in fiscal 2024, with headcount up , and Vector Search was already among the capabilities it had produced. The anchor 10-K expected to commit significant resources to features that use AI and gave no amount. The budget existed; AI changed what part of it is pointed at, and no earlier size for that part can be traced. The size is the whole of an activity that existed before.
“During 2023, we added additional capabilities such as Atlas Vector Search and Atlas Stream Processing, as well as additional features for Atlas Search Nodes, which now provide dedicated infrastructure for search use cases so customers can scale independently of their database to manage their workloads with greater flexibility and operational efficiency.”
“For instance, with the development of next-generation solutions that utilize new and advanced features, including artificial intelligence (“AI”) and machine learning, we may be required to commit significant resources to developing new products, enhancements and developments.”
Research and development, year-over-year growth · 2026-CQ3
Research and development as a share of revenue · 2026-CQ3
Research and development as a share of revenue, prior-year quarter · 2025-CQ3
Research and development personnel cost and stock-based compensation, year-over-year increase · 2026-CQ3
Reported line it is matched to
Research and development rose , , to . The 10-Q attributes it to personnel cost and stock-based compensation, third-party infrastructure and software costs for AI tools, with no split by product.
2026-CQ3: 2025-CQ3: 2026-CQ3: 2026-CQ3:
What else could explain it
line composition: Research and development holds the core database, the self-managed product and every other program beside AI work.
other: The AI tools cost inside the line is the company using AI, read in its own channel; it is not spend on AI products. It is taken out of the base with the Voyage award expense and the research and development compute term, so that no dollar counts twice.
Quotes
“In August, we brought automated Voyage embeddings to Atlas for one-click vector search setup, launched Voyage Code 4, a model purpose-built for code, and shipped an upgraded reranking API”
“you can see in the first half fiscal 2027 results the leverage in the business model and the ability to drive incremental profitability while still investing in growth initiatives, specifically engineering and product innovation.”
“MongoDB announced the general availability of four capabilities that improve AI retrieval in MongoDB Atlas: Automated Embeddings powered by Voyage AI, the Atlas Embedding and Reranking API, voyage-code-4, and Vector Search in Atlas Stream Processing.”
“noted that MongoDB is investing heavily in AI research and strategic acquisitions to ensure its innovations consistently meet the evolving needs of AI.”
“The increase in research and development expense was primarily driven by a $16.7 million increase in personnel costs and stock-based compensation, an $8.1 million increase in third-party infrastructure expenses to support ongoing product development and testing activities and a $7.0 million increase in software costs due to an increase in the use of AI tools.”
“We are increasingly utilizing and building AI and ML capabilities into our business, as well as incorporating AI and ML into our internal operations.”
“In addition, we may incur significant costs and experience significant delays in developing new solutions and services or supporting or enhancing our product offerings to adapt to the changing AI and ML landscape”
Marketing and start-up programs aimed at AI-native companies
0.33% to 2% of the quarter’s revenue
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: described· motive: channel-defensive· before LLMs: expanded
The call keeps marketing programs and developer awareness among investments and says nothing new about programs for AI natives (claim c30); the release mentions a developer event in San Francisco only as the place a product was launched. The 10-Q reports a increase in events and digital marketing spend, the largest item in a small increase in sales and marketing (claim c66). The size is the ledger’s own, , on the assumptions of the two prior quarters.
Evidence: 2 quotes, 2 figures, 2 confounds, 2 from before coverage
Events, digital marketing and start-up programs the company says are aimed at AI-native companies and at awareness of its AI capabilities: the San Francisco developer conferences, the Bay Area campaign, and the start-up program with its credits. The filing reports the increase in spend on in-person events and digital marketing programs inside sales and marketing and does not say how much of it is aimed at AI natives.
Why this motive
Carried: marketing programs and developer awareness remain on the investment list (claim c30); nothing new is said about whom they are aimed at.
Before LLMs: expanded
Developer marketing through user conferences and community events predates AI natives: sales and marketing was in fiscal 2024, with marketing spend and travel up in the year, and the .local conferences were already among its largest events. The programs existed; AI changed whom they are aimed at, and no earlier size for that part can be traced. The size is the whole of an activity that existed before.
“Historically, we have invested in our community through active sponsorship of user groups, our user conferences, MongoDB University and other community-centered events.”
“Finally, we expect to see a significant sequential uptick in expenses since we have some of our largest sales and marketing events in Q2, most notably MongoDB.local in New York.”
“The increase in sales and marketing expense was primarily driven by a $5.2 million increase in spend on in-person events and digital marketing programs and a $4.6 million increase in commissions.”
Compute for serving and training the Voyage models and AI features
0.1% to 1.1% of the quarter’s revenue
Incremental total: counts in full.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: described· motive: exploratory· before LLMs: new
Still no named cost of serving or training models. Subscription gross margin rose, which the call and the 10-Q both attribute to revenue mix, and the cost increase is explained by more third-party cloud infrastructure for Atlas (claims c21, c63, c62). Research and development third-party infrastructure rose , described as product development and testing (claim c64). Serving load is rising: automated embeddings are generally available and the Voyage customer count roughly doubled again (claim c11). The size is the ledger’s own, . The counterparty is mixed: the cloud providers that host serving and training compute, and the third-party AI providers behind the assistants.
Evidence: 8 quotes, 4 figures, 2 confounds, 2 from before coverage
What the company pays to run its own models for customers and to train them: inference for the Voyage embedding and reranking APIs and automated embedding, third-party AI providers behind assistants in its tools, and training compute. It would sit in third-party cloud infrastructure inside cost of subscription revenue and in third-party infrastructure inside research and development; the filing explains both without naming AI. The payees are cloud providers and model providers, so the counterparty is mixed.
Why this motive
Carried as exploratory: the filing refers to third-party AI providers and to possible significant costs (claims c70, c71); whether serving cost is recovered in price is not said, though the models are now sold through a metered API inside Atlas.
Before LLMs: new
The anchor reports no cost of serving or training models: the company sold none. It refers to third-party AI providers it uses, without a cost. The line that would carry serving cost, subscription cost of revenue, was in fiscal 2024, with third-party cloud infrastructure up on the growth of Atlas.
“For example, if we or our third-party AI providers do not have sufficient rights to use the data or other material or content on which AI tools rely, or the output generated by our use of such AI tools, we may incur liability through the violation of such laws, third-party intellectual property, privacy or other rights, or contracts to which we are a party.”
“The increase in subscription cost of revenue was primarily due to a $43.5 million increase in third‑party cloud infrastructure costs, including costs associated with the growth of MongoDB Atlas.”
Research and development third-party infrastructure expense, year-over-year increase · 2026-CQ3
Reported line it is matched to
Subscription cost of revenue rose to , of which is third-party cloud infrastructure that the filing ties to the growth of Atlas. Model serving is not named.
2026-CQ3: 2025-CQ3: 2026-CQ3: 2026-CQ3:
What else could explain it
line composition: Third-party cloud infrastructure is chiefly the hosting of customers’ Atlas clusters.
mix shift: Margin rose on a shift toward the self-managed product, which carries no hosting cost.
Quotes
“In August, we brought automated Voyage embeddings to Atlas for one-click vector search setup, launched Voyage Code 4, a model purpose-built for code, and shipped an upgraded reranking API”
“MongoDB announced the general availability of four capabilities that improve AI retrieval in MongoDB Atlas: Automated Embeddings powered by Voyage AI, the Atlas Embedding and Reranking API, voyage-code-4, and Vector Search in Atlas Stream Processing.”
“The increase in subscription cost of revenue was primarily due to a $31.4 million increase in third‑party cloud infrastructure costs, including costs associated with the growth of Atlas and an increase of $2.3 million in personnel costs.”
“Our subscription gross margin increased to 77% primarily due to a shift in subscription revenue mix toward MongoDB Enterprise Advanced and other revenue products.”
“The increase in research and development expense was primarily driven by a $16.7 million increase in personnel costs and stock-based compensation, an $8.1 million increase in third-party infrastructure expenses to support ongoing product development and testing activities and a $7.0 million increase in software costs due to an increase in the use of AI tools.”
“For example, if we or our third-party AI providers do not have sufficient rights to use the data or other material or content on which AI tools rely, or the output generated by our use of such AI tools, we may incur liability through the violation of such laws, third-party intellectual property, privacy or other rights, or contracts to which we are a party.”
“In addition, we may incur significant costs and experience significant delays in developing new solutions and services or supporting or enhancing our product offerings to adapt to the changing AI and ML landscape”
AI tools used in research and development (software cost)
0.91% of the quarter’s revenue
Incremental total: counts in full.
reported in the filingdisclosure: quantified· motive: exploratory· before LLMs: new
The periodic report sizes this channel and names AI: research and development software costs rose over the prior-year quarter due to increased use of AI tools, of that line’s increase (claim c64), and the six-month comparison attributes the half-year’s increase to the same cause (claim c65). A new risk-factor passage says generative and agentic AI tools are increasingly deployed internally (claim c67). The reported figure is an increase over a year earlier and is the channel’s size by the ledger’s convention for filing-attributed figures; the whole bill would add whatever was paid then, which the ledger puts at as a comparison. Per employee the increase alone is for the quarter. The call says nothing about it.
Evidence: 5 quotes, 4 figures, 2 confounds, 2 from before coverage
The outside bill for AI tools the company's own staff use, chiefly AI coding tools in engineering. From the quarter ended 2026-07-31 the 10-Q reports the year-over-year increase in research and development software costs and attributes it to increased use of AI tools; before that the filings said only that generative AI is used in code development. In the quarters ended 2026-01-31 and 2026-04-30, where no filing attributes a figure, the channel is sized on the same increase-over-prior-year basis for continuity with the reported figure, a departure from the convention of the channel's own dollars; the whole bill is shown beside it as a comparison in the quarter ended 2026-07-31. No vendor is named. The payees are AI tool vendors and model providers, which the counterparty list has no member for.
Why this motive
Carried as exploratory: the filing now gives the cost and the uses, software development, security testing and other functions (claim c67), and still no reason or benefit beyond potential ones (claim c68). No displaced line is shown shrinking, the call does not mention the tools, and no source contradicts another.
Before LLMs: new
A bill for AI coding tools cannot exist without LLMs. The anchor 10-K already says generative AI is used in the company’s code development and gives no cost; the line that would carry one, research and development, was in fiscal 2024 and in its last quarter.
“Moreover, use of generative AI in our code development process, while offering various potential benefits, could also pose certain ownership and security risks with respect to our codebase, given the current legal uncertainties relating to ownership of AI-generated works and the potential for security flaws in output code.”
“For example, if we or our third-party AI providers do not have sufficient rights to use the data or other material or content on which AI tools rely, or the output generated by our use of such AI tools, we may incur liability through the violation of such laws, third-party intellectual property, privacy or other rights, or contracts to which we are a party.”
AI tools software cost increase as a share of the research and development increase · 2026-CQ3
AI tools software cost in research and development, whole bill for the quarter, ledger comparison · 2026-CQ3
AI tools cost increase per employee, using the head count at 2026-01-31 · 2026-CQ3
Research and development, year-over-year increase · 2026-CQ3
What else could explain it
line composition: The reported figure is a year-over-year increase in software costs, not the level of the AI tools bill, and covers research and development only; AI tools used in other functions sit in other lines.
other: No vendor, pricing basis or head count using the tools is given.
Quotes
“The increase in research and development expense was primarily driven by a $16.7 million increase in personnel costs and stock-based compensation, an $8.1 million increase in third-party infrastructure expenses to support ongoing product development and testing activities and a $7.0 million increase in software costs due to an increase in the use of AI tools.”
“The increase in research and development expense was primarily driven by a $40.2 million increase in personnel costs and stock-based compensation, a $12.0 million increase in third-party infrastructure expenses to support ongoing product development and testing activities and an $11.2 million increase in software costs due to an increase in the use of AI tools.”
“We are also increasingly deploying, and expect to continue to increasingly deploy, generative and agentic AI tools internally to support our own operations, including software development, security testing, and other business functions.”
“Moreover, use of AI in our code development process, while offering various potential benefits, could also pose certain ownership and security risks with respect to our codebase, given the current legal uncertainties relating to ownership of AI- or ML-generated works and the potential for security flaws in output code.”
“For example, if we or our third-party AI providers do not have sufficient rights to use the data or other material or content on which AI tools rely, or the output generated by our use of such AI tools, we may incur liability through the violation of such laws, third-party intellectual property, privacy or other rights, or contracts to which we are a party.”
Q3 2026quantified · reported in the filing · exploratory · $7.0mn
Cost displaced by AI1 channel · $909k to $15mn sized · $909k to $15mn incremental
engineering · cheap to verify
Engineering and operating cost displaced by the company's own use of AI tools
0.12% to 1.9% of the quarter’s revenue
Incremental total: counts in full.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: described· motive: exploratory· before LLMs: expanded
The cost of the tools is now reported, more than a year earlier, and what they displace is not: the 10-Q says generative and agentic AI tools are increasingly used in software development, security testing and other functions, and claims no saving (claim c67). Research and development personnel cost rose (claim c64). The CFO credits revenue and operating leverage for the margin (claims c22, c28). Research and development was of revenue against . The size is the ledger’s own, , on the assumptions of the two prior quarters.
Evidence: 7 quotes, 4 figures, 3 confounds, 2 from before coverage
Engineering time and other operating cost displaced by the generative and agentic AI tools the company deploys internally, which the filings say are used in software development, security testing and other business functions. Code with tests is cheap to check. Management claims no saving: the calls attribute margin gains to revenue, and the filing reports research and development personnel cost rising. The displaced cost would be the company's own staff time.
Why this motive
Carried as exploratory: the company now says where it deploys AI tools internally (claim c67) and still claims no saving; the CFO attributes the margin to revenue and to leverage in the business model (claims c22, c28).
Before LLMs: expanded
Engineering personnel cost sat inside research and development, in fiscal 2024, which the anchor 10-K reports as rising with headcount, up . The same filing already says generative AI is used in code development, with potential benefits and no measured saving. The size is the change AI made, not the whole line.
“Moreover, use of generative AI in our code development process, while offering various potential benefits, could also pose certain ownership and security risks with respect to our codebase, given the current legal uncertainties relating to ownership of AI-generated works and the potential for security flaws in output code.”
“The increase in research and development expense was primarily driven by a $83.9 million increase in personnel costs and stock-based compensation as we increased our research and development headcount by 19%.”
Research and development as a share of revenue · 2026-CQ3
Research and development as a share of revenue, prior-year quarter · 2025-CQ3
Research and development personnel cost and stock-based compensation, year-over-year increase · 2026-CQ3
Research and development software costs, year-over-year increase, attributed by the 10-Q to increased use of AI tools (an increase over the prior-year quarter, not the whole bill) · 2026-CQ3
Reported line it is matched to
Research and development rose and was of revenue against . Revenue grew ; the ratio fell because revenue outran cost, which is what the CFO says.
“you can see in the first half fiscal 2027 results the leverage in the business model and the ability to drive incremental profitability while still investing in growth initiatives, specifically engineering and product innovation.”
“The increase in research and development expense was primarily driven by a $16.7 million increase in personnel costs and stock-based compensation, an $8.1 million increase in third-party infrastructure expenses to support ongoing product development and testing activities and a $7.0 million increase in software costs due to an increase in the use of AI tools.”
“We are also increasingly deploying, and expect to continue to increasingly deploy, generative and agentic AI tools internally to support our own operations, including software development, security testing, and other business functions.”
“Moreover, use of AI in our code development process, while offering various potential benefits, could also pose certain ownership and security risks with respect to our codebase, given the current legal uncertainties relating to ownership of AI- or ML-generated works and the potential for security flaws in output code.”
“We are increasingly utilizing and building AI and ML capabilities into our business, as well as incorporating AI and ML into our internal operations.”
Revenue arriving through AI8 channels · $3.0mn to $43mn sized · $500k to $18mn incremental · 6 not sized
customer cohort
Revenue from AI-native customers
Not sized
The quarter's measures are counts and shares by count: a record net new customers of whom many are AI natives (claim c10), and a majority of new Voyage customers, , being AI natives (claim c12); the CFO's bound, small (claim c31), is a word the table never converts. Counts size nothing, so the channel reads directional and gets no ballpark (the methodology's count-only rule, rules sweep of 2026-10-06). The former estimate, mdb-2026-cq3-f39, was an assumed share of Atlas-related revenue. The prior quarters, which give words and no count, keep their ballparks; the step to unsized comes from this quarter's wording, not from a change at the company.
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: new
The bound is restated in the CFO’s own words: the company has started to see some benefit from AI and it is small (claim c31). Around it the detail thickens without a number: momentum in the AI-native cohort, a record net new customers of whom many are AI natives, and a majority of new Voyage customers, , being AI natives new to the company (claims c18, c10, c12). Atlas growth is still attributed primarily to the largest enterprise customers, on the call and in the 10-Q (claims c25, c61). The counts are quoted and the reading is unsized.
Evidence: 12 quotes, 5 figures, 3 confounds, 2 from before coverage
Atlas consumption, and model API usage counted with it, from companies management calls AI natives: start-ups whose product is an AI application or agent (ElevenLabs, Emergent, Fireflies, Eve, Endor Labs). They pay for the same database as every other customer and many arrive through the self-serve channel. Management names customers and says the group is not concentrated; it gives no revenue, share or growth contribution for it, and the dollars sit inside Atlas-related revenue.
Why this motive
Carried. The CFO says the benefit from AI has started and is small (claim c31), and the CEO that many of the quarter’s record new customers are AI natives (claim c10); a count of arrivals is not a measure of their spend, and converting Voyage arrivals into Atlas customers is called early (claim c35). No measured movement in the cohort’s revenue is given, so exploratory is kept.
Before LLMs: new
The payers exist because of LLMs, so the money is new by the first test. The anchor already shows the group forming: more than of the new Atlas customers in the first quarter of fiscal 2024 were AI or machine-learning companies, and the CEO said it was far too early to quantify their revenue. Atlas was of revenue in fiscal 2024, , with no part of it attributed to these customers.
“I think it's way too early, Tyler. I think it's also really tied to the, you know, the market and the product market fit of those customers' businesses, because, obviously, if those customers do well, then we're a beneficiary.”
Net new customers added in the quarter, a record · 2026-CQ3
Share of new Voyage customers that are AI natives with no prior relationship to MongoDB, from management’s words · 2026-CQ3
Atlas annual revenue run rate, approximate · as-of 2026-07-31
What else could explain it
line composition: Atlas-related revenue holds every customer; the 10-Q attributes its growth to large existing customers and names no AI-native part.
relabel: Voyage customers and revenue are counted inside Atlas, so a rising count of Voyage arrivals lifts the Atlas customer count and the record for new customers.
other: A count of new customers says little about revenue: the arrivals management describes are small accounts, many reached through a model API.
Quotes
“Atlas revenue grew approximately 29% year-over-year for the fifth straight quarter, driven by large enterprise customers and building AI momentum.”
“Some choose us from day one. Others start elsewhere, like prompt-driven development platforms, and migrate to us as they hit scaling limits and real usage arrives.”
“Some of our largest existing Atlas customers are beginning to adopt Voyage for AI use cases, while a large majority of new Voyage customers are AI natives and have no prior relationship to MongoDB.”
“We also continue to see momentum in the AI native cohort and across AI signals, including adoption of vector search, new Voyage customers, and a continued increase in clusters connecting through MCP.”
“Our growth to date has been driven primarily by continued strength with our largest enterprise customers, and we expect that to continue in the second half of fiscal 2027.”
“We have started to see some benefit from AI, even though it is small, but we are excited about the momentum, and we do expect consumption to continue to be consistent with what we have seen during the first half of the year.”
“Most importantly, there are some customers who come in as Voyage customers and they become Atlas customers. It is still early because, we just started making sure that we can now cross-sell, upsell, whatever the right term you want to use.”
“This performance reflects the mission-critical role our platform plays for customers, with strength driven by core enterprise workloads and early momentum with AI use cases.”
“Subscription revenue increased by $174.8 million primarily due to an increase in consumption of Atlas by our large existing customers and growth in MongoDB Enterprise Advanced, reflecting ongoing expansion within our customer base, as evidenced by our net ARR expansion rate of 122% as of July 31, 2026.”
Q2 2026direction only · our inference · exploratory · $5.1mn to $36mn
Q3 2026direction only · described, no size · exploratory
customer cohort
Revenue from frontier model labs
0.06% to 2.3% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: direction only· motive: exploratory· before LLMs: new
A direction for the first time: one lab that began in late 2025 with a single inference-side workload moved a few more to Atlas during the quarter, after leaving PostgreSQL over performance and outages (claims c45, c46, c5). Labs are also said to keep research data on the product, and the relationships are still called early (claim c6). No lab is named by management and no amount is given. The size is the ledger’s own, , against in Q1 FY2027; it sits inside the AI-native cohort estimate. The channel this reading was counted inside (ai-native-cohort-revenue) is unsized this quarter, so the overlap no longer holds and this reading counts in totals on its own evidence (overlap rule, 2026-10-06).
Evidence: 6 quotes, 2 confounds
Atlas consumption by the labs that build frontier models, which management says use the database for workloads critical to shipping their products: chat memory and inference-side workloads at one lab, and research stores for experiment results, evaluation data and training artifacts. Management speaks of the labs together with AI natives as one cohort, so this channel overlaps ai-native-cohort-revenue; it is kept apart because the payer is a lab funded by investor capital at a different scale. No lab is named by management and no revenue is given.
Why this motive
Carried: the relationships are called early and varying by lab (claim c6). One lab adding workloads (claim c46) is movement at a single customer with no amount; it does not reach the measured-movement tell.
Before LLMs: new
The anchor names no frontier lab as a customer. Its AI customers were start-ups such as Hugging Face, inside the more than AI or machine-learning companies among new Atlas customers in one quarter of 2023; foundation model providers appear in the anchor 10-K only as technology partners.
What else could explain it
other: Neither the number of labs nor any lab’s spend is disclosed; both are assumed, and the direction given is for one lab.
bundling: Labs are reported together with AI natives.
Quotes
“Multiple leading labs leverage Atlas for workloads that are mission-critical to how they ship their products. One lab uses us for inference and chat workloads after moving away from PostgreSQL due to performance lags and outages affecting user experience.”
“Beyond that, labs use us for research workloads to store experimental results, evaluation data, and training artifacts for model development. These relationships are still early, and engagement varies lab by lab”
“With one of the labs, they started towards the later half of last calendar year with one of the workloads that was running inference on MongoDB and used us as a memory layer.”
Q3 2026direction only · our inference · exploratory · $500k to $18mn
customer cohort
Atlas consumption from established enterprises' AI workloads and AI-readiness modernization
0.33% to 3.3% of the quarter’s revenue
Incremental total: counts at zero.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: described· motive: exploratory· before LLMs: relabelled
The bound holds: some benefit from AI, small (claim c31). The CEO attributes Atlas growth to large enterprise customers and building AI momentum, and the CFO puts early signs of AI workload adoption last in his list of what marks Atlas strength (claims c1, c26). The cases named are customer-facing and in production: a publisher serving over queries a day, a bank’s advisor chatbot, a media company’s agent (claims c4, c39, c42). The 10-Q again names consumption by large existing customers (claim c61). The size is the ledger’s own, , an assumed share of the increase in Atlas-related revenue.
Evidence: 14 quotes, 5 figures, 2 confounds, 4 from before coverage
The part of Atlas growth that management attributes to established enterprises building AI applications and agents on data already in the database (Adobe, Zomato, the Financial Times, a large technology customer), and to migrations it says are being done to get data ready for AI. The product is the ordinary database; the filings attribute Atlas growth to consumption by large existing customers and never name AI. No figure separates these workloads.
Why this motive
Carried: still early, with more workloads reaching production (claim c3) and adoption of AI workloads called early signs (claim c26). The CEO adds that small internal copilots are not where the product is chosen (claim c38). No measure of revenue from these workloads is given.
Before LLMs: relabelled
The same demand was on the books at the anchor under the name of new workloads and application modernization: Atlas was in fiscal 2024, of revenue, and the 10-K attributed subscription growth to direct sales customers. On the call of 2023-06-01 the CEO already described customers choosing Atlas for AI applications and existing customers raising AI use cases with the field, and said AI had not driven that quarter’s workload acquisition. Management gives no measure of the AI part in the coverage window, so the tag is relabelled.
“We are observing an emerging trend where customers are increasingly choosing Atlas as the platform to build and run new AI applications.”
“I think it's way too early, Tyler. I think it's also really tied to the, you know, the market and the product market fit of those customers' businesses, because, obviously, if those customers do well, then we're a beneficiary.”
Large Atlas customers using two or more platform features, share · as-of 2026-07-31
Large Atlas customers using two or more platform features, share a year earlier · as-of 2025-07-31
Daily queries a customer (the Financial Times) serves on a Voyage model (a floor) · 2026-CQ3
Reported line it is matched to
Atlas-related revenue rose , , to . Management credits the largest enterprise customers first and AI second; the filing names consumption by large existing customers and no AI part.
2026-CQ3: 2025-CQ3: 2026-CQ3: 2026-CQ3:
What else could explain it
line composition: The increase in Atlas-related revenue holds every driver of consumption, including the AI-native cohort.
other: The CEO credits the prior year’s performance release for healthy consumption and retention, a cause that needs no AI.
Quotes
“Atlas revenue grew approximately 29% year-over-year for the fifth straight quarter, driven by large enterprise customers and building AI momentum.”
“It is still early, but we are seeing more of these workloads reach production, such as the Financial Times, which leverages us to power AI-driven discovery, reaching millions of readers with interactive experiences at scale.”
“We are seeing this show up across industries in a range of use cases, whether it is retrieval of internal knowledge, customer-facing chatbots and agents, or fraud and identity workflows.”
“By indexing content with the high accuracy voyage-4 model and serving over 100,000 daily queries on the cost-efficient voyage-4-lite model, the Financial Times has significantly cut retrieval costs with minimal performance impact.”
“Our growth to date has been driven primarily by continued strength with our largest enterprise customers, and we expect that to continue in the second half of fiscal 2027.”
“The strength in Atlas is highlighted by the sixth straight quarter of increasing revenue dollar growth year- over- year, strong net ARR expansion rate, increasing multi-product penetration, and early signs of adoption of AI workloads.”
“We have started to see some benefit from AI, even though it is small, but we are excited about the momentum, and we do expect consumption to continue to be consistent with what we have seen during the first half of the year.”
“say you are a wealth manager at a bank and there are lots and lots of knowledge base articles that you want to vectorize, use our embeddings, and then use as a chatbot for folks that are doing wealth management and talking to clients real-time.”
“Even a large media company, which became one of our biggest vector search customer, that was driven by an agent trying to do the semantic query and figuring it out.”
“This performance reflects the mission-critical role our platform plays for customers, with strength driven by core enterprise workloads and early momentum with AI use cases.”
“Subscription revenue increased by $174.8 million primarily due to an increase in consumption of Atlas by our large existing customers and growth in MongoDB Enterprise Advanced, reflecting ongoing expansion within our customer base, as evidenced by our net ARR expansion rate of 122% as of July 31, 2026.”
The only measures are adoption outpacing the rest of the company, in words (claim c2), and the share of large Atlas customers on two or more features, , credited largely to vector and text search together (claim c24): a share by count of customers, joint with a feature that is not AI. A share by count sizes nothing, so the channel reads directional and gets no ballpark (the methodology's count-only rule, rules sweep of 2026-10-06). The former estimate, mdb-2026-cq3-f42, was an assumed share of Atlas-related revenue.
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: expanded
Direction without a level, for the third quarter: Vector Search adoption continues to outpace the rest of the company (claim c2). The share of large Atlas customers on two or more features is against a year earlier, credited largely to vector and text search (claim c24), and the CEO describes how adoption turns into consumption through dedicated search nodes (claim c43). No revenue or customer count is given. The share by count is quoted and the reading is unsized.
Evidence: 10 quotes, 2 figures, 2 confounds, 3 from before coverage
Consumption of Vector Search on Atlas, the retrieval feature customers use to ground AI applications in their own data, usually on dedicated search nodes that are billed as Atlas usage. Management reports the growth in customers using it and its part in the share of large Atlas customers on more than one platform feature; it gives no revenue. The buyers are AI natives and enterprises alike, so the channel shares its base with the two cohort channels.
Why this motive
Carried from Q1 FY2027: measured adoption credited largely to vector and text search (claims c2, c24, c43), while the CFO calls the benefit from AI small (claim c31); the less durable motive is kept.
Before LLMs: expanded
Vector Search was already on Atlas at the anchor, launched in 2023 beside the older full-text Atlas Search and the dedicated Search Nodes both run on, and described as a way to build generative AI applications. The anchor gives no revenue or customer count for it; it sat inside Atlas, in fiscal 2024. Vector similarity search can exist without an LLM and the anchor shows the feature, so the tag is expanded: what moved is the number of customers using it. The size is the whole of an activity that existed before.
“More recently, we launched MongoDB Atlas Vector Search, a capability that significantly simplifies an organization’s ability to use their proprietary data to build generative AI applications.”
“During 2023, we added additional capabilities such as Atlas Vector Search and Atlas Stream Processing, as well as additional features for Atlas Search Nodes, which now provide dedicated infrastructure for search use cases so customers can scale independently of their database to manage their workloads with greater flexibility and operational efficiency.”
“Over the years, we have introduced additional features and functionality, which have increased the capabilities of MongoDB Atlas and accelerated and expanded its adoption including Atlas Search, Atlas Device Sync, Atlas Data Federation and Atlas Charts.”
Large Atlas customers using two or more platform features, share · as-of 2026-07-31
Large Atlas customers using two or more platform features, share a year earlier · as-of 2025-07-31
What else could explain it
bundling: Vector Search is consumed as Atlas usage; nothing in the filings separates it from other consumption.
line composition: The multi-feature share is credited to vector and full-text search together, and the year-earlier share quoted this quarter is higher than the one quoted last quarter for the quarter before it.
Quotes
“Voyage customer count nearly doubled quarter-over-quarter, and Atlas Vector Search adoption continues to outpace the growth of the rest of the company, showing our strong early momentum for AI workloads.”
“By indexing content with the high accuracy voyage-4 model and serving over 100,000 daily queries on the cost-efficient voyage-4-lite model, the Financial Times has significantly cut retrieval costs with minimal performance impact.”
“We also continue to see momentum in the AI native cohort and across AI signals, including adoption of vector search, new Voyage customers, and a continued increase in clusters connecting through MCP.”
“Of our Atlas customers generating at least $100,000 in ARR, 48% are leveraging two or more features on our platform, which is up from 42% in the year-ago quarter, driven largely by vector and tech search adoption.”
“The strength in Atlas is highlighted by the sixth straight quarter of increasing revenue dollar growth year- over- year, strong net ARR expansion rate, increasing multi-product penetration, and early signs of adoption of AI workloads.”
“We have started to see some benefit from AI, even though it is small, but we are excited about the momentum, and we do expect consumption to continue to be consistent with what we have seen during the first half of the year.”
“say you are a wealth manager at a bank and there are lots and lots of knowledge base articles that you want to vectorize, use our embeddings, and then use as a chatbot for folks that are doing wealth management and talking to clients real-time.”
“Even a large media company, which became one of our biggest vector search customer, that was driven by an agent trying to do the semantic query and figuring it out.”
“MongoDB announced the general availability of four capabilities that improve AI retrieval in MongoDB Atlas: Automated Embeddings powered by Voyage AI, the Atlas Embedding and Reranking API, voyage-code-4, and Vector Search in Atlas Stream Processing.”
Q1 2026direction only · described, no size · exploratory
Q2 2026direction only · described, no size · exploratory
Q3 2026direction only · described, no size · exploratory
product revenue
Voyage AI embedding and reranking models
Not sized
The only measures are a customer-count multiple, Voyage customers roughly doubling again, times (claim c23), and the share of new Voyage customers that are AI natives, (claim c12), a share by count; no revenue is given, and the CFO's only size word for AI, small, is never converted. Counts size nothing, so the channel reads directional and gets no ballpark (the methodology's count-only rule, rules sweep of 2026-10-06). The former estimate, mdb-2026-cq3-f43, was one quarter of an assumed annual level.
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: new
The customer count nearly doubled on the quarter again, times, and management now says who the customers are and how they arrive: a majority of the new ones, , are AI natives with no prior relationship to the company, most referral traffic comes from coding agents, and few know the product is MongoDB’s (claims c23, c12, c36, c44). Some of the largest Atlas customers are starting to use it. No revenue is given; the CFO’s only size word for AI is small (claim c31). The counts are quoted and the reading is unsized.
Evidence: 19 quotes, 3 figures, 3 confounds
Revenue from the embedding and reranking models of Voyage AI, acquired in February 2025 and sold as a metered model API, now also from inside Atlas. Voyage is itself an AI company whose product is an AI product, so its revenue is an AI channel. From the quarter ended 2026-01-31 the company counts Voyage customers as Atlas customers and Voyage revenue inside Atlas-related revenue. Management reports growth in the number of Voyage customers and the 10-K says its revenue was not material; no revenue is given.
Why this motive
Carried from Q1 FY2027: customers roughly doubled for a second quarter (claim c23), and the CEO calls their conversion to Atlas early (claim c35); the less durable motive is kept.
Before LLMs: new
The anchor shows no model sold by the company: Vector Search stored and searched embeddings that customers produced with other providers’ models. Voyage AI was an independent company then, outside this income statement, and a metered model API cannot exist without the models; the baseline is the anchor’s silence.
Figures
Voyage customers, multiple of the prior quarter’s count, from management’s words · as-of 2026-07-31
Share of new Voyage customers that are AI natives with no prior relationship to MongoDB, from management’s words · 2026-CQ3
Daily queries a customer (the Financial Times) serves on a Voyage model (a floor) · 2026-CQ3
What else could explain it
relabel: Voyage customers and revenue are reported inside Atlas; the count that doubled is not published.
bundling: Automated embeddings are now generally available inside Atlas, so Voyage usage and Atlas usage are sold together.
other: Two doublings of a customer count from a small base say little about revenue; the new customers are small and many arrive on a recommendation from a coding agent.
Quotes
“Voyage customer count nearly doubled quarter-over-quarter, and Atlas Vector Search adoption continues to outpace the growth of the rest of the company, showing our strong early momentum for AI workloads.”
“By indexing content with the high accuracy voyage-4 model and serving over 100,000 daily queries on the cost-efficient voyage-4-lite model, the Financial Times has significantly cut retrieval costs with minimal performance impact.”
“The best AI applications need a strong database, which is why we have long pointed to developers building on Claude to MongoDB Voyage for embeddings.”
“In August, we brought automated Voyage embeddings to Atlas for one-click vector search setup, launched Voyage Code 4, a model purpose-built for code, and shipped an upgraded reranking API”
“Some of our largest existing Atlas customers are beginning to adopt Voyage for AI use cases, while a large majority of new Voyage customers are AI natives and have no prior relationship to MongoDB.”
“We also continue to see momentum in the AI native cohort and across AI signals, including adoption of vector search, new Voyage customers, and a continued increase in clusters connecting through MCP.”
“Within Atlas, Voyage customers roughly doubled quarter- over- quarter for the second consecutive quarter, continuing the encouraging signs of the demand for our AI embedding capabilities.”
“We have started to see some benefit from AI, even though it is small, but we are excited about the momentum, and we do expect consumption to continue to be consistent with what we have seen during the first half of the year.”
“we are really, really energized by the new customer count for MongoDB that is coming via Voyage. You are absolutely correct, and that is why those remarks were made explicitly, that many of them are actually not MongoDB customers.”
“Most importantly, there are some customers who come in as Voyage customers and they become Atlas customers. It is still early because, we just started making sure that we can now cross-sell, upsell, whatever the right term you want to use.”
“If EA gets to being AI-ready with search, vector search and so on, they are asking, "Hey, can we also make Voyage AI available in a self-managed type of an environment?”
“When the team did analysis on where is the referral for our Voyage is coming, as you would have imagined, most of this referral is coming via coding agents. Number one, Claude, and number two, Codex is driving most of the referral traffic for Voyage.”
“say you are a wealth manager at a bank and there are lots and lots of knowledge base articles that you want to vectorize, use our embeddings, and then use as a chatbot for folks that are doing wealth management and talking to clients real-time.”
“Looking ahead, we're emerging as the intelligent data platform for the AI era—launching powerful retrieval innovations, simplifying how developers connect coding agents to their operational data on MongoDB”
“MongoDB announced the general availability of four capabilities that improve AI retrieval in MongoDB Atlas: Automated Embeddings powered by Voyage AI, the Atlas Embedding and Reranking API, voyage-code-4, and Vector Search in Atlas Stream Processing.”
“Recently, we have introduced an application programming interface (“API”) within Atlas that natively provides access to Voyage AI’s embedding and reranking models.”
Q2 2026direction only · described, no size · exploratory
Q3 2026direction only · described, no size · exploratory
customer cohort
Enterprise Advanced demand attributed to AI in self-managed environments
Not sized
Management names AI readiness beside operational resilience, data sovereignty and public cloud capacity as reasons for self-managed demand (claims c40, c48), and gives no rate for the AI part; nothing separates it (the methodology rule for AI named beside another cause).
Matched line moved : the whole line, not this channel.
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: relabelled
Management now says in words that AI is pushing demand for the self-managed product as well as Atlas: regulated industries making operational data AI ready, AI on the self-managed product opening net new demand, the AI push showing in both products (claims c40, c15, c47). Those words, not a rate, carry the directional state. The line that holds this demand with every other, Enterprise Advanced and other revenue, grew , and full-year growth expected for it was raised to about (claim c27); no rate is given for the AI part. The other reasons given need no AI: resilience, sovereignty, capacity (claim c48). The 10-Q names growth in Enterprise Advanced and no cause (claim c61). The line rose . AI is named beside resilience, sovereignty and capacity and nothing separates its part, so the channel is left unsized.
Evidence: 14 quotes, 3 figures, 2 confounds, 2 from before coverage
The part of demand for the self-managed Enterprise Advanced product that management attributes to AI: regulated customers keeping critical data on their own infrastructure for reasons that now include AI, and extending those estates to AI applications. The dollars sit inside Enterprise Advanced and other subscription revenue, which the filings explain by multi-year deals and expansion in the customer base without naming AI.
Why this motive
Carried as exploratory: this channel is the demand for the self-managed product itself, to which no AI price or measure is attached; the priced feature now has its own channel. The reasons given mix AI readiness with resilience, sovereignty and cloud capacity (claims c40, c48), which is a joint cause, not a contradiction.
Before LLMs: relabelled
Enterprise Advanced was of subscription revenue in fiscal 2024, , and falling as a share. On the call of 2023-06-01 the finance chief explained its demand by the workloads customers still ran on premises, with no mention of AI. The product, the buyers and the licence are the same in the coverage window; only the reason given is new, and no measure of the AI part is disclosed.
“MongoDB Enterprise Advanced is our proprietary commercial database server offering for enterprise customers that can run in the cloud, on-premises or in a hybrid environment.”
Enterprise Advanced and other subscription revenue · 2026-CQ3
Enterprise Advanced and other subscription revenue, year-over-year growth · 2026-CQ3
Enterprise Advanced and other revenue growth expected for fiscal 2027, raised from mid-single digit, approximate · FY2027
Reported line it is matched to
Enterprise Advanced and other revenue rose , , to . The CFO credits the product’s strategic importance to large customers and early traction of search and vector search; the filing names growth in the product and no AI cause.
2026-CQ3: 2025-CQ3: 2026-CQ3: 2026-CQ3:
What else could explain it
one time item: Multi-year licences recognize part of their value up front; the CFO expects the line to be about flat in the second half and asks that it be read on the full year.
other: Operational resilience, data sovereignty and public cloud capacity are given as reasons for self-managed demand beside AI.
Quotes
“This quarter, we brought search and vector search to EA, closing a gap between our cloud and self-managed experiences. Demand came in immediately and across industries from customers looking to take a consolidated approach to building AI in their own governed, self-managed environments.”
“This quarter, that bank extended that same environment to GenAI and semantic search for employee advisors, chatbots, product search, and document intelligence.”
“Bringing AI self-managed opens net new demand for us, and hybrid deployment often means that the strong EA estate opens the door to net new Atlas conversations within the same customers.”
“Third, EA and other had an exceptional quarter, growing 36% year-over-year, driven by EA's growing strategic importance to many of our largest customers and early traction from our Q2 launch of search and vector search on EA.”
“This continued momentum highlights the strategic importance of EA as customers continue to expand their self-managed footprints to support both traditional and AI applications.”
“For EA and other, given the strength we saw in the first half, including the demand we are seeing for the search and vector search capabilities we launched on EA in Q2, we are raising our full-year expectations for EA and other revenue to approximately 11% growth in fiscal 2027, up from our prior guidance of mid-single digit growth.”
“customers asked us that we want to run for this large, massive workloads that they run on MongoDB EA. "CJ, we want to get this AI ready, hence the team should build a search and vector search on it.”
“when we introduce this functionality for search and vector search, which was driven by the AI demand. We are charging our customers extra for that feature set that we are providing in EA.”
“What happens is when somebody wants to run a neo cloud because of the capacity issues in a public cloud that they may have, they're saying, "Hey, can we run EA in that neo cloud?" Which goes to our run anywhere and driving demand.”
“MongoDB Search and Vector Search are now generally available in MongoDB Enterprise Advanced, bringing the retrieval capabilities MongoDB Atlas customers use to build AI applications in the cloud to self-managed, private cloud, and hybrid environments”
“Subscription revenue increased by $174.8 million primarily due to an increase in consumption of Atlas by our large existing customers and growth in MongoDB Enterprise Advanced, reflecting ongoing expansion within our customer base, as evidenced by our net ARR expansion rate of 122% as of July 31, 2026.”
“Our subscription gross margin increased to 77% primarily due to a shift in subscription revenue mix toward MongoDB Enterprise Advanced and other revenue products.”
Q1 2026described · described, no size · exploratory
Q2 2026described · described, no size · exploratory
Q3 2026direction only · described, no size · exploratory
pricing packaging
Search and Vector Search on Enterprise Advanced, charged as an extra
Not sized
The CFO credits the line's growth to Enterprise Advanced's growing strategic importance together with early traction of the launch (claim c17), and the extra charge covers full-text search, which is not AI, together with vector search; nothing separates AI's part, so the channel is left unsized with no ballpark (the methodology rule for AI named beside another cause). The ceiling is the year-over-year increase in Enterprise Advanced and other revenue, . The former estimate, mdb-2026-cq3-f45, was an assumed share of that increase.
described, no sizedisclosure: described· motive: offensive· before LLMs: expanded
The channel opens this quarter. Search and Vector Search for the self-managed product were delivered on 2026-06-30, the CEO says customers are charged extra for them and that customers had asked for them to make large self-managed workloads AI ready (claims c33, c41, c32). The CFO credits early traction of the launch as one of two reasons for the line’s growth and for raising its full-year outlook to about (claims c17, c27). No price, attach rate or revenue is given. The launch is named beside another cause and the charge covers a feature that is not AI, so the reading is unsized; the line's increase, , is its ceiling.
Evidence: 11 quotes, 2 figures, 4 confounds, 4 from before coverage
The extra charge for Search and Vector Search on the self-managed product, made generally available on 2026-06-30. The CEO says the feature set was built because of AI demand and that customers are charged extra for it. It sits inside Enterprise Advanced and other revenue; no price, attach rate or revenue is given. First read as part of ea-demand-from-ai, when it was a roadmap item.
Why this motive
A separately priced feature set: the CEO says it was built because of AI demand and that customers are charged extra for it (claim c41), with demand arriving at once (claim c13). The CFO calls the traction early (claim c17); the price in force decides.
Before LLMs: expanded
At the anchor Search and Vector Search existed on Atlas only, and Enterprise Advanced, in fiscal 2024, carried neither. The activity is in the anchor under the same names on the other product, and what AI changed is a price: an extra charge on the self-managed licence that did not exist before 2026-06-30. The size is that charge, so the basis is an increment; it also covers full-text search, which is not an AI feature. The size is the change AI made, not the whole line.
“More recently, we launched MongoDB Atlas Vector Search, a capability that significantly simplifies an organization’s ability to use their proprietary data to build generative AI applications.”
“During 2023, we added additional capabilities such as Atlas Vector Search and Atlas Stream Processing, as well as additional features for Atlas Search Nodes, which now provide dedicated infrastructure for search use cases so customers can scale independently of their database to manage their workloads with greater flexibility and operational efficiency.”
“Over the years, we have introduced additional features and functionality, which have increased the capabilities of MongoDB Atlas and accelerated and expanded its adoption including Atlas Search, Atlas Device Sync, Atlas Data Federation and Atlas Charts.”
“MongoDB Enterprise Advanced is our proprietary commercial database server offering for enterprise customers that can run in the cloud, on-premises or in a hybrid environment.”
Enterprise Advanced and other revenue growth expected for fiscal 2027, raised from mid-single digit, approximate · FY2027
Enterprise Advanced and other subscription revenue, year-over-year increase · 2026-CQ3
What else could explain it
bundling: The extra charge covers full-text search as well as vector search, and only the second is an AI feature.
one time item: A term licence recognizes part of its value on signing, so a feature available for one month can still show in the quarter.
other: Whether the charge is a separate line item or a higher licence price is not said.
other: The CFO names Enterprise Advanced's growing strategic importance beside the launch as a reason for the line's growth (claim c17).
Quotes
“This quarter, we brought search and vector search to EA, closing a gap between our cloud and self-managed experiences. Demand came in immediately and across industries from customers looking to take a consolidated approach to building AI in their own governed, self-managed environments.”
“This quarter, that bank extended that same environment to GenAI and semantic search for employee advisors, chatbots, product search, and document intelligence.”
“Bringing AI self-managed opens net new demand for us, and hybrid deployment often means that the strong EA estate opens the door to net new Atlas conversations within the same customers.”
“Third, EA and other had an exceptional quarter, growing 36% year-over-year, driven by EA's growing strategic importance to many of our largest customers and early traction from our Q2 launch of search and vector search on EA.”
“We saw early demand for the search and vector search capabilities we launched on EA in Q2, adding retrieval capabilities that enhance our ability to support AI workloads.”
“For EA and other, given the strength we saw in the first half, including the demand we are seeing for the search and vector search capabilities we launched on EA in Q2, we are raising our full-year expectations for EA and other revenue to approximately 11% growth in fiscal 2027, up from our prior guidance of mid-single digit growth.”
“customers asked us that we want to run for this large, massive workloads that they run on MongoDB EA. "CJ, we want to get this AI ready, hence the team should build a search and vector search on it.”
“when we introduce this functionality for search and vector search, which was driven by the AI demand. We are charging our customers extra for that feature set that we are providing in EA.”
“If EA gets to being AI-ready with search, vector search and so on, they are asking, "Hey, can we also make Voyage AI available in a self-managed type of an environment?”
“MongoDB Search and Vector Search are now generally available in MongoDB Enterprise Advanced, bringing the retrieval capabilities MongoDB Atlas customers use to build AI applications in the cloud to self-managed, private cloud, and hybrid environments”
Customers arriving through coding agents, MCP and agent frameworks
Not sized
The only measures are a share of referral traffic by source, most of it from coding agents (claim c36), and a rising count of clusters connecting through MCP (claim c18): shares and counts with no base and no revenue. They size nothing, so the channel reads directional and gets no ballpark (the methodology's count-only rule, rules sweep of 2026-10-06). The former estimate, mdb-2026-cq3-f46, was a multiple of the Voyage estimate, which is itself now unsized.
described, no sizedisclosure: direction only· motive: channel-defensive· before LLMs: new
The door is now described from the inside. An internal analysis found most referral traffic for Voyage comes from coding agents, Claude first and Codex second, and the CEO quotes a lab executive saying the lab points developers to Voyage (claims c36, c8). Clusters connecting through MCP keep increasing and a hosted MCP server was launched for coding agents (claims c18, c57). The share given is of referral traffic, with no base and no revenue, so the state stays directional and the reading is unsized.
Evidence: 9 quotes, 2 figures, 2 confounds, 1 from before coverage
Revenue that arrives because an AI platform put the product in front of a builder: coding agents such as Claude and Codex recommending Voyage and the database, the MCP server that connects agents to Atlas, the plugin on a coding agent's marketplace, and integrations with agent frameworks such as LangChain. Management reports where referrals come from and that MCP usage is rising; it gives no revenue arriving this way. The customers it brings are largely AI natives and Voyage users, so the channel overlaps those two.
Why this motive
Carried: a hosted MCP server for coding agents, described as staying embedded in how new applications get built (claim c7). Management now reports that coding agents recommend Voyage and bring customers (claim c37), which would read offensive; no revenue or conversion is measured, so the less durable motive is kept.
Before LLMs: new
Referrals from coding agents and connections through an agent protocol cannot exist without LLMs. The anchor shows the older form of the same door: developers finding the product through the free Community Server and the Atlas free tier, and cloud marketplaces as a partner channel. Its only mention of AI coding tools is the view that code assistants would raise demand for databases in general.
“In 2023, a number of companies launched code assistant tools, which leverage generative AI to help developers write and test their code faster, thereby accelerating application development. We believe these developments in coding assistant technology will further benefit the data management software market.”
Share of new Voyage customers that are AI natives with no prior relationship to MongoDB, from management’s words · 2026-CQ3
Net new customers added in the quarter, a record · 2026-CQ3
What else could explain it
other: Referral traffic is not revenue; the customers it brings are new and small.
line composition: Customers who arrive through an agent are counted as ordinary self-serve customers.
Quotes
“Just recently, we launched a fully managed MCP server, making it easier for developers and agents to connect directly to MongoDB when they are using Claude Code, Codex, and Grok Build, as well as popular coding tools like Cursor and Devin from Cognition. This is how we stay embedded in the AI supply chain for how new applications get built.”
“The best AI applications need a strong database, which is why we have long pointed to developers building on Claude to MongoDB Voyage for embeddings.”
“We also continue to see momentum in the AI native cohort and across AI signals, including adoption of vector search, new Voyage customers, and a continued increase in clusters connecting through MCP.”
“we are really, really energized by the new customer count for MongoDB that is coming via Voyage. You are absolutely correct, and that is why those remarks were made explicitly, that many of them are actually not MongoDB customers.”
“When the team did analysis on where is the referral for our Voyage is coming, as you would have imagined, most of this referral is coming via coding agents. Number one, Claude, and number two, Codex is driving most of the referral traffic for Voyage.”
“Looking ahead, we're emerging as the intelligent data platform for the AI era—launching powerful retrieval innovations, simplifying how developers connect coding agents to their operational data on MongoDB”
“MongoDB launched the Atlas Managed MCP Server at .local San Francisco Build Fest, a fully hosted service that connects coding agents—including Claude Code, Codex, Grok Build, and Devin by Cognition—to live MongoDB Atlas data without requiring customers to deploy or manage additional infrastructure.”
“In addition, the rapid adoption of generative and agentic AI tools may change how software developers design, build and deploy applications, including through increased automation of coding, infrastructure configuration and application architecture. Because a significant portion of our strategy is focused on being the database platform of choice for developers, changes in developer workflows, tooling or decision-making processes resulting from the adoption of AI-enabled development tools could affect how developers select databases or other data platforms.”
Q1 2026described · described, no size · exploratory
Q2 2026direction only · described, no size · channel-defensive
Q3 2026direction only · described, no size · channel-defensive
Cost imposed, or revenue lost, by others’ AI1 channel · $0 to $2.8mn sized · $0 to $2.8mn incremental
distribution
Workloads lost when coding agents and prompt-driven platforms choose another database
0% to 0.37% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: described· motive: imposed· before LLMs: new
Management now states the mechanism in its own words: some AI natives choose the product from the start, others start elsewhere, on prompt-driven development platforms, and migrate when they hit scaling limits (claim c9); a frontier lab is the quarter’s example of a move from PostgreSQL (claim c5). The 10-Q keeps the risk that AI development tools change how a database is chosen (claim c72). What is not said is how many never migrate. Net new customers were a record . The ledger’s range, , is the same allowance as before.
Evidence: 4 quotes, 1 figure, 1 confound, 1 from before coverage
Revenue the company does not receive because the choice of database is now often made by an AI coding agent or a prompt-driven development platform that defaults to another database, typically PostgreSQL. The 10-K names the risk; management's account is that such companies start elsewhere and migrate once they hit scaling limits. It is the other side of the door the coding-agent referral channel comes through. No figure exists for workloads that never arrive.
Why this motive
A toll channel: where a prompt-driven platform or a coding agent first puts an application’s data is decided outside the company. Imposed by construction.
Before LLMs: new
A database chosen by a coding agent cannot exist without LLMs. At the anchor the company read AI code assistants only as a benefit to database demand, named no risk from them, and counted more than AI or machine-learning companies among one quarter’s new Atlas customers.
“In 2023, a number of companies launched code assistant tools, which leverage generative AI to help developers write and test their code faster, thereby accelerating application development. We believe these developments in coding assistant technology will further benefit the data management software market.”
Net new customers added in the quarter, a record · 2026-CQ3
What else could explain it
other: Lost workloads leave no trace in a filing; the migrations management cites are the cases that came back.
Quotes
“Multiple leading labs leverage Atlas for workloads that are mission-critical to how they ship their products. One lab uses us for inference and chat workloads after moving away from PostgreSQL due to performance lags and outages affecting user experience.”
“Some choose us from day one. Others start elsewhere, like prompt-driven development platforms, and migrate to us as they hit scaling limits and real usage arrives.”
“In addition, the rapid adoption of generative and agentic AI tools may change how software developers design, build and deploy applications, including through increased automation of coding, infrastructure configuration and application architecture. Because a significant portion of our strategy is focused on being the database platform of choice for developers, changes in developer workflows, tooling or decision-making processes resulting from the adoption of AI-enabled development tools could affect how developers select databases or other data platforms.”
Q3 2026. Growing slower than revenue: cost of revenue (+17.8%), research and development (+17.7%), sales and marketing (+3.7%), general and administrative (+25.2%), total operating expenses (+11.6%). A displaced cost shows up as a line that stays under the dashed revenue line. These are the audited lines, as first reported; nothing here is attributed to AI by the filing.
Cost of revenueResearch and developmentSales and marketingGeneral and administrativeTotal operating expensesRevenue