Product1 distinct publisher3 min readPublished
Every headline number beat and the annual guide went up, yet the one line that would show AI workloads landing in production has held at 29% for three quarters. That flat line is the number that matters here.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Consumption pricing is what makes this quarter readable at all. Atlas bills on what clusters actually do, and most of MongoDB's revenue now comes from Atlas [10], so a retrieval feature that reached production in the spring either shows up in usage or it never left the demo. That is unusual. Most AI adoption claims come from a slide. This one comes from a meter that can be checked directly.
The meter said 29%, and it said something close to that in each of the two quarters before [6]. Read the order of the clauses in what CJ Desai said and the same picture appears: strength driven by core enterprise workloads, and early momentum with AI use cases [5]. Every quantified figure in the results is total revenue or Atlas revenue; the AI reference carries no number attached to it [19].
The raise is worth doing arithmetic on. The full-year revenue guide midpoint moved from about $2.94bn to about $3.01bn, roughly $70m [13]. The quarter itself landed about $36.8m above consensus [11]. So management pushed a little under twice the beat into the year [14], which leaves something near $33m of genuinely new expectation for the back half [14]. Meanwhile the current-quarter guide midpoint of $758.5m sits about $13.3m below the $771.8m just reported, a 1.7% sequential decline [15]. Sandbagging a guide is normal. It is still the company's own arithmetic for the period in which compounding AI workloads would surface first.
The market half is simpler than the product half. Mizuho Securities analyst Jordan Klein told clients that major hedge fund holders had privately wanted Atlas at 30% to 31%, and that a rally of more than 20% in the prior month left no tolerance for error [7]. The distance between the hope and the print was one to two percentage points [17]. It cost more than 14% after hours on top of 4% in the regular session, leaving the stock up 3% on the year [8].
Anyone defending an AI feature's cost internally can apply this same disclosure test to their own bill. Split the database spend the feature caused into the one-off embedding backfill and the steady-state query and index cost that persists after launch. If the only bump was the backfill, you shipped a migration rather than a workload, and the line will flatten the way Atlas's rate did. If steady-state consumption is still higher 90 days after launch, you have retention in the only form a database can express it, and you can ask for budget with a number rather than an adjective. Teams often assume that shipping vector search moves the database to the center of the AI stack; the meter here has shown 29%, three times running [6], with the enterprise workloads that were already there named first as the driver [5]. The same arithmetic applies to your own bill, not just MongoDB's.
Ranked by verification strength, evidence, and original report placement.
MongoDB's second-quarter revenue rose 30% to $771.8 million, above the analyst consensus estimate of $735 million.
MongoDB raised its fiscal 2027 outlook to earnings of $6.39 to $6.58 per share on revenue of $2.99 billion to $3.03 billion, up from an earlier range of $5.95 to $6.14 per share on revenue of $2.92 billion to $2.96 billion; consensus was $6.13 per share on $2.96 billion.
On the conference call Desai said Atlas revenue grew 29% year over year; some analysts were disappointed and asked how MongoDB could accelerate, given the service has grown at more or less the same pace for three straight quarters.
MongoDB reported earnings before certain costs such as stock compensation of $1.90 per share, versus the Street's target of $1.62.
MongoDB posted a net profit of $40.9 million in the quarter, compared with a net loss of $47 million in the same period a year earlier.
President and CEO CJ Desai said the 30% revenue growth rate was the company's highest in several years and that the performance reflected the mission-critical role of the platform, "with strength driven by core enterprise workloads and early momentum with AI use cases."
Distinct publishers with included, body-backed reporting in this cluster.
2 articles · September 1, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
build
Atlas App Connections hands AI clients your whole Atlas role, so fix the roles first1 distinct publisher
build
Your meter now runs on someone else's machine: signed receipts, fsync, and failing open1 distinct publisher
invest
Spark's $22M bet that the agent framework layer can stay independent1 distinct publisher
build
Three services you can delete: queue, cache and search in one Postgres1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Precise figures, one newsroom
The numbers are specific and internally consistent, and they all originate in one place: MongoDB's own results and call, as written up by SiliconANGLE and then published a second time in near-identical form. No filing is cited, no other outlet checks the print, and the most interesting non-company detail — what hedge funds privately wanted from Atlas — reaches the reader third-hand, through a characterization of a Mizuho client note.
Real consumption, unlabelled AI share
Money moving through the platform is not in doubt — $771.8 million in a quarter, Atlas carrying most of it, and a swing to GAAP profit. What cannot be measured from this reporting is how much of that consumption is AI work. The disclosure stops at total and Atlas revenue; the AI portion exists only as the word 'early' in a CEO quote, and Atlas's rate has not budged across three quarters.
AI credit ahead of the AI line
The overstatement is narrow but real. Beat, raise and guide are all as described, and the market's own verdict is reported rather than sanded down. The stretch is in attribution: vector search and 'early momentum with AI use cases' are offered as part of the growth story while the only line that would evidence them has held at 29% for three quarters, and the annual raise adds about $70 million of expectation against a current-quarter midpoint that steps 1.7% below the quarter just delivered.
Everyone speaking has a position
Three interested parties shape this story and none is neutral. Management sets the frame on results day, and 'highest growth in several years' is the frame that helps. Mizuho is explaining a drawdown to clients who are positioned in it, with unnamed hedge funds as the evidence. And SiliconANGLE's own pages close with an appeal to buy AWS services through its marketplace links and to join its alumni network — a vendor-adjacent publisher writing about a vendor.
Confident on the print, thin on the why
We would stand behind the arithmetic without hesitation: the beats, the midpoints, the roughly 18% combined decline all follow from figures stated twice in identical terms. Confidence drops on causation and on durability — one publisher, one analyst's read of why the stock broke, and no independent line on whether the Atlas plateau reflects saturation, macro, or AI simply not landing in production yet.