Welcome to MongoDB's 2026 Investor Day. Thank you so much for joining us, whether you are here in person with us at the Nasdaq market site or joining us online via the webcast. My name is Jess Lubert , Head of Investor Relations at MongoDB. We have a packed agenda. We are excited to talk about our market opportunity, some of our new innovations, hopefully you read about earlier today, as well as to provide you with the opportunity to hear directly from some of our senior leaders. Before we get started, a few quick housekeeping items. We expect today's program to run about four hours, concluding around 3:15 P.M. We will have two breaks. We will have a 30-minute lunch break around noon, and we will have a 10-minute bio break following our partnership session and before the financial update.
We are going to hold two Q&A sessions, one to address yesterday's announcement up front, and then we will do a longer session at the end following the presented content. Finally, the full presentation deck will be posted to the investor relations website shortly following our prepared remarks. Now for a brief legal disclosure. Our remarks today will include forward-looking statements for purposes of the safe harbor provision under the Private Securities Litigation Reform Act of 1995, including statements regarding our market opportunity, strategy, product innovation and financial targets. These statements reflect our views as of today only and are subject to risks and uncertainties that could cause actual results to differ materially. For a full discussion of these risks, please refer to our SEC filings, including our most recent Form 10-Q for the period ended July 31st, 2026. Additionally, we will discuss non-GAAP financial measures today.
A reconciliation of these measures to their most directly comparable GAAP metrics is included in the appendix of today's presentation on our IR website. Here is our schedule for today. We are going to kick it off by addressing what you read about yesterday, followed by a short Q&A session on the topic. After that, here is what you will hear. Dev Ittycheria, our interim President and CEO, is going to talk about our market and opportunity to drive durable growth. Ben Cefalo is going to talk about core products and some of the new innovations to Atlas and Enterprise Advanced you have hopefully read about today. Pablo Stern-Plaza is going to talk about our AI capabilities and Atlas Agent Engine. You are going to hear from Emergent Labs, which is one of our AI customers, about how they use our products. Ryan Mac Ban will talk about our go-to-market strategy.
Erica Volini will talk about partners as a scale engine, and then Mike Berry will come on the stage and tie it all together. I want to make a few acknowledgments. A tremendous amount of work went into preparing for today. We would like to extend a sincere thank you to all the internal teams who built today's content. You will also hear from some of our customers today, so we would like to thank our customers for taking time out of their schedule to help you understand how they use our products. Then we would like to thank the Nasdaq for hosting us here at this fantastic site. With that, I would like to have Mike Berry come on the stage for some opening remarks before we dive into the agenda. Mike, over to you.
Thank you, Jess. Great to see everybody. Thank you for joining us. Well, during my presentation, I will stand over here so you can see the slide. It just doesn't really matter, so I am going to stand in the middle. Before we get started on the event today, and hey, folks, we got a lot of great stuff for you to go through. We wanted to address the news that came out yesterday. I will do a few prepared remarks, and then I will ask Dev to join me, and we will do a short Q&A session focused on the transition. We really can't say anything more than is in the 8-K, and today we are focused on the plan going forward. As you know, CJ resigned last Friday, and Dev was named as the interim President and CEO. We wish CJ all the best in his new role at Meta.
He had a lot of great sayings, as most CEOs do, and one of his favorite was, "Don't look back, solve forward." That's what we're doing, folks. We are solving forward today. Going forward, the board has named Dev as interim President and CEO. Dev is certainly well-known to employees and the investment community, and he's uniquely qualified to lead MongoDB during this transition. He remained on the board over the past year, and I can say he was certainly very involved in the running of MongoDB during that time. The board has already launched a search for a new CEO, and we're very confident they will bring aboard a great new leader for the business. During the transition, folks, it is business as usual at MongoDB.
The business is doing very well, and we have a lot of exciting announcements to discuss today and tomorrow at our .local New York event. To the point of it's business as usual, it's important for you to know that the whole management team is here today, and we are fully aligned with the direction and plans that we're going to show you today. In addition, we have several members of our board of directors as well. I am not going to introduce the folks that are going to present. They're going to get enough of the spotlight. But I would love the management team of the non-presenters to stand up, if you would, please. Last year, you heard from Jim, our CTO, and May, our CMO.
In addition, you have Deepa, who's our Chief Information Officer, Andrew, who's our Chief Legal Officer, you have Harsha, who's our Chief People Officer, and you also have Ashish, right behind Andrew there, who's our Technical Fellow. So they are here as well as the folks presenting. I want to make sure and introduce you to them. Folks, this is a team event. We are all in front of you today. Then if I could ask the four board members to stand up, please, because it's super important they're here to support us as well. You have Roelof, Hope, Chip, and Tom. So they are here to support us as well. The goal, folks, is, and I've gone back and forth on this, and I'm just going to say it, this is not about one person. This is about 5,700 people that work every day to move MongoDB forward.
And what you are going to hear today is the plan of all of the employees. Our goal is to make sure the transition goes very smoothly. We as a team are going to continue to drive the business forward. We take our responsibilities to customers, employees, and shareholders very seriously, and you have all of our commitments, all of ours, that we will stay focused and drive execution. We have a great business and a lot of momentum, and our job is to keep it going. With that, I would like to invite our former and our new President and CEO, Dev Ittycheria, to join me for a short Q&A.
I will take a few questions here. Raise your hand when we call on you. Be sure to state your name and your firm's name. Keith, you want to kick it off here.
Hi. Keith Bachman from Bank of Montreal. Welcome back officially.
Thank you.
I wanted to ask a question on the transition. That is to say, Jess and I were talking to Hall. What do you think CJ did particularly well over the last nine months in a short period of time to help position MongoDB? When you think about your leadership, however long that may last, what are you really focused on in terms of priorities? The third part is, as you think about what you are looking for for your next leader, anything you want to call out that you think is particularly important, whether it is the R&D, go-to-market, et cetera?
Yeah. Thanks for the question. First I would say, I think what CJ did well was he really elevated the company to be able to sell at the C-level. He has a pretty big rolodex coming from ServiceNow, where he knew a lot of senior decision-makers. Historically, with infrastructure software, you have to sell both at the end-user level because you have to get people buying, because no one is going to try and shove infrastructure software down end users' throats, unlike application software. Then you have to still sell high to get the economic buyers bought into whatever the compelling reason is. I think he, especially in some key accounts where he had deep relationships, he helped us, I would say, accelerate deals.
What I will also tell you is that, and I have seen this myself, I show up at a C-level, we shake hands, a deal happens. It was not because of that meeting. It is all the groundwork that happened before that meeting to set the stage for us to consummate the deal. I do not want to suggest that was a one-man show. It was really the go-to-market team who really helped galvanize the momentum in those accounts. Clearly, CJ added some value on key accounts where he helped push the deals over the line. On the second question, what am I going to focus on? One is really driving momentum. I think we have a lot of momentum. I think you are going to be quite pleased in terms of what we are going to share. Sharpening and executing our priorities.
Obviously, a lot of focus on product and a lot of focus on go-to-market. Also just focusing on our people and our culture. I think one of our assets has always been our culture. That is something I am really focused on. Then in terms of, and I will speak more about this in my talk. In terms of who we are looking for, obviously, we want someone who can help scale it. We think the opportunity here is enormous. I did obviously resign because at some point, after 11 years, you feel like you want to hand the baton to someone else. But we want to find someone who is really passionate about where to take this business forward. I think this business has a chance to become a very, very large business, and let me just explain why.
With the move to AI and agentic workloads, as people get more sophisticated, agents need to understand what is happening, right? They need to be able to identify and find information, and that information has to be fresh, not stale. It cannot be from a week ago. It cannot be in from a few hours ago about what is going on. Then you need to be able to respond to that information and find the right relevant information, because you can have a lot of fresh information, but if you pick the wrong information, you can make a bad decision and reason about something poorly. Then you need to be able to scale elastically, because, by definition, these workloads are very unpredictable, and you could have agent people talking about agent swarms and stuff like that. The traffic could be very unpredictable. We have an architecture designed for that.
You also need to manage state. We will talk a lot about all this a little later, but I think we are really well-positioned to truly help customers through this agentic journey and be a beneficiary of that.
Thank you. Over here.
Sanjit Singh, Morgan Stanley. Dev, great to see you again, and so glad you are back to-
Thank you.
Head up Mongo on this transition. My main question is about when I look forward over the next several quarters, just the magnitude of change. In CJ's tenure, he did make a lot of changes. You mentioned the go-to-market changes. Product-wise, there was a double down on Enterprise Advanced. A lot of people left MongoDB that formerly worked here. A lot of new people have come in. When we think about the next couple of quarters, there will be a new CEO at some point as well, and that also represents change, right?
Yeah.
In terms of managing all this organizational change, what's your strategy for that? The newer team members that have come over in the last year, are you reaching out to them to make sure they stay on board? How are you just thinking about managing that process?
Yeah. Thanks for the question, Sanjit. What I would tell you is for the board members who've been here a long time, like Chip and Tom and Roelof, and obviously Hope has also been here a very long time, we've seen a lot of executive changes. We went through a lot of people, and one of the things we realized is as the business scales, not every executive scales at the same rate. I would suggest that the change we've seen over the last year is something more profound. The organization's gone through many changes in the past. In fact, we replaced our CRO right after we went public. I was temporary CRO for, if you people remember, for almost a year before we promoted someone in that role. I think the organization's ability to handle change is quite good.
I would also say that change is a constant. We are in an environment where things are changing so rapidly. I think it is just par for the course now to deal with change. I would say to the third point about someone new coming, I would say, clearly, we would want someone who really understands the business and not just make change for the sake of change. Some of the changes that CJ made, people were just tired. Some of my team who I was working with me, they had also been on the journey for a long time. It was not like they suddenly need a new skill set. They were just tired. It was a long time, so they want to take a break.
I am pretty confident we are going to navigate through this well. I would just say that change is something that we are very used to.
Hey, Dev. Rishi Jaluria, RBC. Good to have you back. Maybe one that I would love to continue on that thought process, one of the big changes that we saw under CJ was this refocus on Enterprise Advanced, bringing some of what was Atlas-only features into Enterprise Advanced, especially for these large enterprise AI workloads. Can you maybe talk about that strategy, kind of some business continuity on that? It seems to be reflected in the updated targets of not just an Atlas growth but an an overall growth number. Maybe what more can you do to build on that success that we have seen with Enterprise Advanced, especially for AI workloads in large regulated industries? Thanks.
Yeah. I did not touch on that, and Sanjit asked the same question. What I will tell you is those decisions that were made to invest in Enterprise Advanced happened when I was here. It was not like all of a sudden we suddenly decided to double down. You are going to hear from Ben and Pablo later, but we were making a concerted effort because we had heard from customers, a lot of organizations you represent, a lot of large financial institutions still want to run workloads on-prem. Now obviously what we are seeing is the market is shifting and buying behavior is shifting, especially with AI, that people do not necessarily want to run everything in a hyperscaler. They may want to run on-prem, they want to run in Neocloud, and they are also very concerned about IP protection, so they may want to use open source models.
Having that flexibility really strengthens our run anywhere story. We had currently put a lot of momentum around Atlas, and obviously, that showed up in our results. But we heard loud and clear from a lot of large customers, and financial services represents a large Enterprise Advanced estate for us that, hey, they want to see more investment. We had already started that work before I had left.
All right, we will take our last question on this session of the day here. Alex?
Hey, Dev. Alex Zukin from Wolfe Research. Glad to have you back, and I apologize. This is probably an unfair question, but I think it is important. It is basically, if you had to think about, given the rate of change and the pace of change, how long do you anticipate this process to take to find a new leader? What is the prototypical leader in your mind? Is this somebody that is highly technical? Is it more sales-oriented? Is it more West Coast-based to Reclaim the Bay? Just give us kind of your wish list.
Yeah, I want to be clear, it is not going to be just solely my decision. The board will be involved, and obviously, I suspect many members of the executive management team will be part of the process to evaluate candidates. What I would say, and again, it is obviously only been a couple of days, but what I would say is that I do not think we are looking for one particular domain expertise. I think we want someone who has experience building software companies and also understands both building and go to market because we have a large organization. We serve so many different organizations. We have a fairly sophisticated go-to-market organization with an enterprise sales force, a commercial organization, a self-serve business, and a partner channel, and you will hear all about that a little later today. We want someone who really understands.
My preference is, in fact, I wrote this when I just recently joined Sequoia. I like people who can understand technology but equally understand what it takes to acquire customers and keep them, right? I think you need to have a mix. You cannot just be purely one-dimensional. We are going to look at a lot of candidates. My phone is already been ringing off the hook. I do not doubt that we will have a lot of good candidates to choose from.
All right.
Okay. Thank you for that. Thank you for keeping it short. What does Ed Sullivan say? On to the show. Now we are going to walk you through the story. Hey, folks, we got a lot of great things. Thank you for joining us. It is going to be a long day. It will get a little bit hot in here, but we will take breaks. Dev, on behalf of everyone at MongoDB, welcome back. Thank you. I do want to say, I just want to double click what Mike just said. This is a great team here. Mike is a world-class CFO, and this business was not run by one person. There was a real team here. Not just the team that reported to CJ, but the team under them.
I am privileged to represent and support that team, but they are the ones really running the business. Okay. I will try and cover a few things quickly. Can we get the slides going? First and foremost, what we are doing is we are essentially trying to become an intelligent data platform. I just want to go back. When I joined MongoDB, the challenge then was could we prove ourselves to be viable for mission-critical use cases? Could customers really conceive of using a document database to run truly mission-critical workloads? We did that. The second job was could we build a cloud business, because people were fairly skeptical that we would go head to head with people like Amazon and Google and GCP, and win. We actually proved that.
The third leg of the stool now is really expanding the breadth of capabilities we can offer to be able to essentially address a broader set of customer needs. We came out with things like Atlas Vector Search. We came out with just full text search, and then with the Voyage AI acquisition, we do things like embeddings and re-rankings. What we've also seen is that our business has grown very quickly. The market is massive. This is one of the largest markets in enterprise software. This market is going to essentially double from 2024 in six years to nearly $200 billion in terms of size. If you look at our business, we've grown the business 5x since 2020, and we expect the business, based on our guidance, to hit $3 billion by the end of this year.
The key point is our market is large and growing, and we're growing much faster than market because of our strong execution. A couple of things we've done is also, since we've grown our revenue 5x, our customer count has grown by 3x. What that says is not only are we acquiring lots of customers, but we're also expanding more deeply inside those organizations to acquire more workloads and address a broader set of use cases. We serve some of the most demanding and sophisticated customers across nearly every vertical industry, every geography, and for almost every conceivable use case. When it comes to AI, I think it's very important for all of you to understand why AI is a tailwind for three basic reasons. One, more workloads. With AI, you can produce far more software far more quickly.
By definition, more software applications require more databases and more data persistent stores. That's point number one, and you're going to hear a lot more about this. Two, new workloads. We're going to address a broader set of workloads around agents, memory, knowledge work. The breadth of use cases expands not just from software, but taking over knowledge work that people were doing manually. The third is a brand new set of customers. We've already expanded into the frontier labs. We're seeing a lot of AI native companies starting to use MongoDB. So it's a whole ecosystem, a new genre of customers who are not just customers of ours, but are starting to become larger and larger customers. What do these customers really need?
Whether you're a Fortune 500 company or an AI native company or a frontier lab, there's some things that are common across all these customers. One, you need real-time context. AI, as I mentioned earlier, is only as good as the data it uses. If it reasons on bad data or stale data, you're going to have bad answers. Two, even if you have fresh data, you need to precisely retrieve that data accurately and succinctly. Third, you need to be super elastic and highly scalable, right? Because the unpredictable nature of these agentic workloads is something people have not seen before, and traditional SQL databases are not designed to scale well. The fourth thing is that you also need trust.
No one's going to trust an AI system or an AI application if they, one, can't guarantee the quality of the answers, or that potentially something else may go wrong. It's not reliable, it's not available, it's not secure. Security, governance, availability, and reliability are super important. Last but not least, and I touched on this, you need to be able to run anywhere. I think it's very clear with AI that customers are taking a very different approach than, say, with the cloud and anyone who was rushing to the cloud. They're starting to be much more discriminating about which workloads they may run in the cloud, which workloads they may run on-prem. Sometimes they may consider a hybrid environment for and it's all about IP protection, data residency and sovereignty, and a bunch of other reasons.
What that says is that the more options customer has, the better it is for MongoDB, because we can run anywhere, and you don't have to change a line of code. What's interesting, and you're going to hear from this from Ben, but even with Postgres, if you go from AlloyDB to an AWS offering like Amazon Aurora, you actually have to change a lot of code. That's a misconception a lot of people have is that Postgres is this portable technology that can be run anywhere. On top of that, you're going to hear a lot about this today, so I'm not going to really belabor this, but as I said, the docu model and the capabilities are coming out with agent memory is going to be really interesting. MongoDB 9.0 and Atlas Infinite is something that I'll touch on a little bit.
We've really doubled down on Queryable Encryption, something no one else has in the industry, and the fact that we can run anywhere. Let me just give you a couple of customer examples that I think you'll find interesting. This is one of the largest healthcare providers in the world today. Their business has grown with us by 22x in 3 and a half years. One of the big use cases they used us for was to modernize their entire claims platform, which is probably their most mission-critical platform. It handles 15 million transactions per day at under 200 milliseconds, so super low latency. On top of that, once we laid that foundation, they said, "Hey, we want to offer a chatbot for our customers to answer questions rather than trying to search for documents," so forth.
They built an AI chatbot on top of MongoDB, and with that, they've lowered their TCO by almost 80% and increased their accuracy of answers by almost 60%. They're a very happy customer. Another customer is a very large media company. They've run their entire streaming business on top of MongoDB. That ARR went up by 3x in just 12 months. They started with one app and now they've grown to 20 different streaming services and streaming applications. Their user preference of record, because they have different brands, so users can pick their preferences, and it's portable across all the streaming services that this company offers. All that was built on MongoDB. Now they're offering very sophisticated search, where people can search on videos, and not just titles, but also scenes.
They are really trying to amp up the ability to search almost any content that they offer for people to use that in different ways. Last but not least, is a large European retailer. They have been up 7x in the last three years. What they did was they moved a very complex ERP application off Oracle to MongoDB. Then on top of that, they replaced Elastic with Vector Search and basically are running their production system on MongoDB. Then on top of that, they built an AI chatbot to help customers address specific questions about, are items available, the variations that they could offer, and other details about their inventory. What you see is that once you lay the data foundation for customers, it is very easy for customers to build upon that to offer new AI capabilities. Excuse me.
This is just a small subset of some of the AI natives who are already customers of ours. We cannot mention the labs for all the obvious reasons, but you can probably predict who some of them are. What is interesting is the frontier labs are using MongoDB for critical use cases like model training, managing their chat system, high velocity product delivery like feature testing, experimentation, analytics, et cetera. What is also interesting is many of these AI natives are not just using one capability. Two-thirds of them are using actually more than one capability. So they are really using the broad platform. We find that super exciting. Then these frontier labs are not just customers, but they are also becoming distribution partners. One of the interesting things we did a little over a month ago was launched an MCP managed connector.
We always had an open source MCP connector, but it required a lot more work on the customer to set it up and configure for them to use it on MongoDB. We came out with a managed connector Atlas, and we have seen interest spike 8x over the last six or seven weeks. What it allows customers to do is with one button, have a one button connection to Claude Code, Codex, Grok Build, et cetera, and so they can connect to Atlas to either instantiate a new instance or be able to query the data as they are building applications. We feel like we are just getting started. The other interesting thing is a lot of you have asked, how can you tell if an end user is using Atlas versus an agent using Atlas?
Not every agent use the MCP connector today because people sometimes still direct connect to Atlas. This gives us a lot of visibility about agent traffic coming into Atlas and will allow us to get a sense of how we are doing in this new genre of workloads. Again, you are going to hear a lot more about this later, but we have made many meaningful investments over the last few years. Things like, as I said, Vector Search, 8.0, stream processing, and Voyage. We also announced a bunch of new capabilities earlier this year, both to deepen and broaden the platform, as well as announce some new AI capabilities. What we are really excited about is that you are going to hear some really interesting stuff. First and foremost, 9.0.
The performance of 9.0 is the best we have ever delivered. The throughput is 2x faster, and you can do reads and writes 30%-35% faster. We know that 8.0 was a step function in performance over the prior releases. We believe we are delivering the most performant platform in the industry. Second is Infinite, and this may be the most profound innovation we have launched since, I would say, asset transaction support, which is very complex in a distributed database or even Atlas itself. What we have done is decouple storage and compute, so you can scale storage and compute independently. You could have very storage intensive use cases or very compute intensive use cases.
What it does, it allows you to scale up very quickly, allow for agentic workloads, get better economics, and you can tune the performance based on the type of the workload. You are going to get a lot more detail on this when Ben comes up here. Last but not least, we have built an agent memory runtime and governance stack for running agents on production. Remember I said, once you lay the data foundation, it becomes easier for people to build agentic applications, and you are going to hear a lot about how we are priming this stack to enable customers to get started. The advantage we offer, unlike some other people, is one, we are open to any model. If you are working with a model provider, you cannot do that. We are open to any framework, and we are open to any cloud or open source model that you want to run on-prem.
It gives customers extreme optionality. Like software, there is really two engines of the plane that really drive a software business. One is the product innovation we just talked about, but the other is the world-class go-to-market. Even my short time when I left MongoDB day to day and then started working with VCs and more recently with Sequoia Capital, I will tell you that MongoDB's reputation from a go-to-market point of view is second to none. If anyone knows my history, we have produced some world-class go-to-market leaders. This is something of much thought and effort we put into product, we also put into distribution. We are spending a lot of time and a lot of investments in moving up the market to serving our enterprise accounts better and driving more market share inside those accounts. We are investing a lot in partnerships.
Ryan is going to talk about sales organization, Erica is going to talk about partner organization, and we are also making a lot of investments in specific vertical organizations. You are going to hear a lot from our go-to-market leaders about how we are really investing and growing our go-to-market team to capture the big opportunity we see in front of us. To kind of summarize where I am focused, one is really to drive momentum. This is not a company that needs a lot of fixing or problem-solving. The company is on strong footing. You will see it when the other people present, and then you will see it in the numbers when Mike presents. Two is to really execute sharply on product and go to market, and three is to provide a seamless transition while I am here.
That's what I'm focused on. I'm going to hand it off now to Ben, who's our Chief Product Officer for Core, to talk about the innovations there. Thank you.
Thanks, Dev. Hi, everyone. Nice to see everybody. I'm Ben. I lead the core products across our core platform, including the database, our Atlas managed service, and our community and our Enterprise self-managed products. Today I'm going to, as well as my co-CPO, Pablo, we'll tell you all about why we have the right product and strategy to win now. I've been with MongoDB for nearly a decade, and in that time I've had countless customer conversations. Lately, they all come back to the same theme. Customers are building applications that are more demanding than we've ever seen before, with real-time operational data playing a crucial role. Our platform is built for this moment.
That's because of a decision our founders made at the very beginning when they built MongoDB, and they created the first database to make JSON in the document model truly native. They saw that relational databases were holding companies back and relational schemas are inflexible by design. Every new field, every structural change takes a significant amount of effort to migrate. They didn't try to make a better relational database, they built something better. Here's why that matters right now. AI agents need to work with complex and evolving data on the fly, and JSON is perfect for this. It maps naturally to how applications and now agents want to model data, and that's why JSON is the default language of AI. A question I hear constantly, though, and a lot from this community, is how does native JSON support stack up against JSONB in Postgres?
I get why. On paper, JSON's in the name, right? But having native support versus a bolt-on hack is fundamentally different. The difference should show up in multiple ways in how we handle large documents more effectively. Updates are faster because we're not rewriting the entire JSON blob on every change. Our query language is designed for JSON, making it easier on developers and also now agents. At the end of the day, if you're using Postgres and JSONB, you are forcing flexible data into a rigid table, canceling out all of the flexibility benefits customers need from JSON. This kind of inflexibility isn't a nice-to-have in the AI era. It's a prerequisite, and it's our durable differentiator, and it's more valuable than ever. We are delivering on innovations built around the document model that strengthens this position.
I am going to focus on three key areas today. First, we are doubling down on our Enterprise Advanced and our Run Anywhere strategy. Regulation, data sovereignty, AI intellectual property concerns means customers can't accept limits on where to run. Second, we are hardening our foundation with 9.0, so it is always ready for mission-critical workloads. Third, we are delivering extreme elasticity with Atlas Infinite, so customers can scale rapidly for unpredictable events and agent-driven demands. Okay, so let's get to it. First, we are going to dig into Run Anywhere. In nearly every enterprise conversation I have, customers tell me they need the option to run across multiple cloud environments and cloud and self-managed environments. They need the control over where workloads run and where data lives for three reasons. First, regulation. In many industries and countries dictate how data can be stored and where.
Enterprises want control and are concerned about intellectual property. They need to know exactly where their data lives and how it is handled. Now with AI in the mix, they don't want their proprietary data exposed to risks around AI access, training, or ownership. Plus, business continuity and resiliency means enterprises need a safety valve. They can't be locked into a single environment. If there is an outage at a data center or they want to change vendors, they need their applications to be truly portable. We are seeing these reasons play out with our customers today. For example, Nationwide U.K., the second-largest provider of household mortgages and savings in the U.K., runs across both EA and Atlas to meet their regulatory requirements while giving members real-time access to their accounts and transactional data. A major U.S.
bank uses EA to serve more than 100 production applications across payments, fraud detection, and account services. By self-managing, they keep the sensitive customer data and conversational data inside of their own environment. Lastly, Zoom runs EA deployed across dozens of environments globally to deliver low latency, highly available communications at scale and wherever their business needs dictate. This need for flexibility is exactly what our Run Anywhere strategy is built for. What makes Run Anywhere even more powerful is having EA as our self-managed option alongside of Atlas in the cloud. Last month marked the 10-year anniversary of MongoDB Atlas. Back in 2016, we were convinced that the cloud would fundamentally change how organizations build and use the database. So we built Atlas, a fully managed database as a service with all the benefits of cloud-native infrastructure.
Since then, we have delivered a steady stream of innovation. From our initial launch in support for the three major hyperscalers to client-side field-level encryption to advanced networking options, we stayed ahead of the curve and defined what a world-class managed database should be. Today, Atlas can run across AWS, Azure, and GCP in 130 cloud regions. Customers can even run a single cluster across multiple clouds simultaneously. Atlas is also available in all the cloud marketplaces. So regardless of what hyperscaler our customers choose, they can transact through their cloud provider with ease. Let's not us forget that this is the same database that you can run in any environment. This is not something you can easily do with Postgres. Since there are so many flavors out there, and each behaves just a little differently.
As Dev said, Amazon Aurora is not the same as Neon, which is not the same as Supabase, which is not the same as open source Postgres. You think you're betting on one database, but you're really limiting yourself to that specific flavor. This means applications are not portable. With MongoDB, our database can run on a developer's laptop, in self-managed enterprise environments, and even in air-gapped deployments for the most sensitive workloads like public sector or regulated industries. This is the definition of run anywhere. Because of this innovation and flexibility, Atlas has been used by nearly 20 million users and processes more than 3 trillion queries each day. To put that into a little perspective, that's 35 million queries a second and about 340 queries for every man, woman, and child on Earth per day. Let's leave that for a second.
It's more than double what it was just a few years ago. But let's not forget, Atlas is also a great way for builders to get started. There's this perception out there, and you all ask me about it every time we talk, but the data tells a different story. Not everyone starts with Postgres. On average, we see 500,000 free tier clusters each quarter, and this has been growing steadily each year. And when you combine that with 150 million community downloads, that's a huge number of builders that begin their journey with us. Putting it all together, we believe that our customers deserve the same capabilities no matter where they choose to run, and it's this portability that really matters. Run Anywhere gives our customers unique, flexible advantages and is a huge competitive moat for us. That is why we're also continuing to invest in EA.
In June, we GA'd Search and Vector for self-managed customers. As a reminder, we introduced Vector Search in 2023 in Atlas, which was actually before the ChatGPT launch. We were in private preview with our Vector search. Today we see customer workloads with billions of vectors. Vector search is a core building block for AI applications, and customers told us clearly they needed it to run in their own self-managed environments, not just in our cloud. And with that, teams end up stitching together, without that, excuse me, teams would end up stitching together a database in a vector store themselves. This is what they have to do today with relational databases, which adds complexity and an infrastructure tax teams cannot afford when trying to ship AI applications fast.
By making search available for Enterprise Advanced and customers get the core database and Vector Search running anywhere they need without the complexity. The demand came in immediately and across different use cases. A large multinational bank is simplifying its AI architecture by bringing together operational data, vector embeddings, and full text search on top of Enterprise Advanced. Another large multinational financial services company wants MongoDB to serve as a system of record for all the customer information using our search capabilities. Enterprise choice matters more than ever. Our customers want to know we will keep investing in our self-managed options, which we've made that commitment, and we don't plan to stop there. Now, all of that only matters if the core database can carry the load, which brings us to MongoDB 9.0.
Since our inception, we've never taken the foot off the gas when it comes to the database innovation. As we've moved deeper into the enterprise, the applications we power grew more demanding. We matched that growth with a constant drumbeat of innovation, staying ahead of the customer's needs so they could keep building on us at any scale. With that innovation, I just want to remind you that we were one of the first distributed databases to support transactions and that we natively supported horizontal scaling, which is how you distribute data across machines to support a larger scale from the very beginning. Our 8.0 release went GA two years ago right around now, which delivered meaningful performance gains over the previous release. This release has experienced faster adoption, with nearly 80% of the Atlas fleet is already on eight.
Since launch, we're also seeing higher customer satisfaction and improvements in customer retention. Clearly, our customers have been pleased with what 8.0 has delivered, and now this brings us to today with the release of 9.0. 9.0 is the best release of MongoDB we've ever shipped. We continue to harden the core to give the most demanding enterprises a strong foundation. With 9.0, we're setting a new bar for security, durability, availability, and performance. We've made a significant improvement in the way we look at performance and scale. Up to 20% higher transactional throughput, 35% faster reads, 30% faster updates, and 2x more throughput on larger instances. To put it a little bit into perspective is 9.0 is faster than 8.0, which is faster than 7.0. If you used a release before 2023, we're about 100% faster than we were back then.
The customers have also noticed this. This has made MongoDB a better fit for more opportunities. All of this adds up to greater efficiency, which means greater consumption. We saw this pattern with 8.0. Better performance and scale means customers serve more users, transactions, and AI-driven activity. In other words, it's Jevons paradox in action. When we introduce better price performance, we've seen stronger growth in subsequent periods. Our customer, Coinbase, has been putting 9.0 through its paces at the real production scale, so let's take a look at their video.
Coinbase is the largest crypto exchange in the U.S. In addition to trading tools, we also provide a developer platform to empower agents to discover and buy things with crypto.
Back early in 2024, we saw an opportunity to revive the HTTP 402 standard. However, we realized that we needed to build a way for agents to be able to autonomously discover all of the different services that existed on the Internet, and that is why we decided to build the Bazaarvoice.
One of the advantages of running our XHR2 facilitation and Bazaar discovery on MongoDB's multi-region architecture is that it allows us to have low latency as well as high availability across regions while still keeping our core functionality available. This enables us to have more flexibility and observability into how we are generating our search results without having to maintain several separate search infrastructures on our own.
MongoDB 9.0 is a great step forward for us as we think about how internet-native payments are going to scale to billions of transactions over time. There is going to be parallel searches, all of that being handled and abstracted away behind the scenes is yet another example of how our team can stay lean, focused on building, and have all of the implementation details and nitty-gritty of our database handled for us. I can say with confidence, we do not have to worry about details around our latency, uptime, operational SLAs, and SLOs, and all of that back-end work is going to be handled seamlessly. Working with MongoDB means that our team can focus on the core competency of bringing agentic payments to the mainstream, supporting hundreds of thousands, if not millions, of AI agents transacting seamlessly and frictionlessly across the Internet.
Thanks to Coinbase for that awesome collaboration. Can we switch the screens in the back, please? Thank you. Performance and scale are non-negotiables for their business, and that is exactly what 9.0 has been built to deliver. In 9.0, we have also enhanced our industry first and industry-only Queryable Encryption capability. Last year at our Investor Day, our CTO, Jim, talked about Queryable Encryption, which is a powerful security capability that only MongoDB has. In most databases, the data remains in the clear so it can be queried. Queryable Encryption lets customers keep the data encrypted and still query it. Let us say a customer wants to find a name that starts with the beginning parts, or find a credit card number that ends with six, seven, eight, nine, or a medical record note that contains fever.
They can run these queries while the data stays encrypted, and the query itself is also encrypted. This is why Queryable Encryption makes MongoDB really suitable for applications with super highly sensitive data. For example, the France's internal digital force manages and secures digital communication systems for law enforcement. They needed a defense-in-depth strategy, and Queryable Encryption let them encrypt the data knowing that they are the only people that could ever decrypt it. To meet regulatory requirements, a leading financial institution needs to retain all the customer notifications about their financial products. They leverage Queryable Encryption to protect the PII data contained in those communications, which helps safeguard their customer identities. Like everything else we do, we always are thinking about Run Anywhere. Again, you can use Queryable Encryption no matter where you are running. 9.0 is the best release we have ever delivered for our customers.
The demands of modern and AI applications require us to keep building the most performant and fastest database we can. There is another level of demand that is on the horizon, and we are skating to that puck today. We are bringing even greater elasticity and scale to our platform with Atlas Infinite, making it easier to handle the bursty nature of AI agents. Infinite is Atlas's biggest architectural innovation since its inception in 2016. We just heard from Coinbase about how important scale is as they bring more and more agents online. What makes agent scale requirements different from those we have seen in the past? Take a look. User-driven applications can be more predictable based on timing or seasonality. Their intensity also scales with the number of users in the system. Think about Cyber Monday. You know the traffic is coming, you can plan for it.
On the other hand, agents do not generate steady, predictable traffic. They can sit idle for long stretches and then suddenly introduce massive, unpredictable demand that hits the database all at once. Even with users interacting with the application, agents will also make purchases, adjust prices, manage inventory, and handle fulfillment on that same Cyber Monday. So you need a database designed for this unpredictable and intense agentic activity. This is what Infinite delivers, the ability to handle both of these workload shapes simultaneously. It is built for extreme elasticity. Infinite is designed to scale up and scale down rapidly. When agents run in an autonomous loop or fan out and spawn additional agents, the database is able to respond. This is the exact shape of workload Infinite was designed for from the ground up.
As an application scales, customers want more performance per dollar and pay only for what they use, so it offers tunable performance. To be clear, this is not about spending less. It is about making sure businesses are not paying for idle capacity that could be used elsewhere. Plus, it removes all the guesswork out of capacity planning by scaling up and down to meet the customer's demands in real time. To be very clear, this is not a whole new database to migrate to. It provides the same Atlas experiences customers already love. Infinite is a new deployment option within Atlas that current customers can adopt seamlessly. No application code changes are required. The current edition of Atlas database is not going anywhere. We are renaming it to Atlas Core Database, and we are going to continue investing in it as well.
Infinite delivers all of this with a very high security posture. At launch, we already have SOC 2, ISO, and PCI, and more to come later on. Architecturally, we are encrypting the data before it leaves the compute layer, and that compute always runs in a secure environment dedicated to each customer. Neither the cloud providers nor we can see any data that a customer stores in Atlas. We have been running a private preview for Infinite with some of our largest customers, and the results have been excellent. One of our preview customers, Icon Solutions, processed more transactions per second, and another one, PicPay, ran 4x its normal peak traffic. Tomorrow, Okta is going to join me on stage at our local event to share their experience with Infinite as well. Right now, let us hear what Carrier had to say.
Go ahead and play the video, please.
Carrier is one of the biggest HVAC companies in the world. We build HVAC equipment for stadiums, schools, commercial buildings. We have had a digital platform initiative, so we get a large amount of telemetry data coming in from our chillers into our customers' specific application environments. It means a lot of data. Prior to Atlas Infinite, we were not able to separate our compute-based scaling from our storage-based scaling. If it cannot scale, if it suffers outages, then that has a direct impact on our application, which has a direct impact on our customers that use that application. We started with other vendors, and we ran into problems when the tooling does not necessarily fit the problem. MongoDB ended up being really the right solution in a few different areas for us.
Since Atlas Infinite is based on a zero-waste model, it helps us to feel a lot more comfortable that we can continue to scale without hitting unexpected storage or compute costs. It almost sounds unbelievable, but MongoDB, compared to our prior tooling, showed that it was at least 10x faster in terms of building and at least 10x more efficient in terms of resources needed in order to deliver the product. Having that relationship with MongoDB gave us confidence in that when we have a complex problem, there is enough thought put into the product to provide a robust solution. We feel pretty confident that it will help us solve that problem in the future.
Our thanks to Glenn and the team at Carrier for their partnership. As Glenn said, efficient scaling directly impacts his customers and the Carrier business. Atlas Infinite can deliver elastic scaling because we have separated the compute from storage. Let us take a couple moments, and we will walk through how that works. When tuning for performance and scale, you can adjust compute and storage in multiple ways. First, there is vertical scaling. This is when you just buy a bigger box. Then there is horizontal scaling. When you add more servers into the system. In both approaches, compute and storage are coupled. Scaling up often means adding both compute and storage together, whether you need both or not. How does Atlas Infinite get extreme elasticity? This is where separating the compute from the storage comes in. This architecture unlocks the next level of performance.
When an application suddenly needs more compute to process eventic activity, the database can respond in an instant. Extreme elasticity and tunable performance at scale are what we hear from our customers, and they need it and expect it from a database. It is exactly what Infinite delivers. Changes like this do not happen overnight. You cannot easily implement this architecture in an OLTP database. With our large enterprise customer base, you need to do it in a way that guarantees minimal disruption without lowering the bar for security and durability. That is why we have been working on Atlas Infinite for over two years. This expands the set of opportunities that we can compete for, especially ones where scaling compute and storage separately are critical, such as bursty, unpredictable nature of agents. We have been putting Atlas Infinite through its paces, and the results have been pretty impressive.
These gains help customers scale faster and support a higher ceiling as more and more agents come online. As I mentioned earlier about Jevons paradox, we expect better price performance to translate into greater usage as our customers build bigger and more ambitious applications. Atlas Infinite is a major innovation in our core database, and we are perfectly suited for the eventic applications that require high performance with rapid scaling. It also puts us in a position to win even more customers where compute and storage have to scale independently. This means that Atlas Infinite will be a win for workloads that drive more growth for MongoDB. Let me bring this back to where I started. Our founders bet on JSON because they saw that the real world does not fit neatly into rigid tables, and that bet has proven to be a great one over time.
When we have the opportunity to educate customers on the benefits of MongoDB, they choose us over relational alternatives. Everything I have walked through today builds on that original advantage. Run Anywhere means we can compete for more because customers can choose where to run. 9.0 means our foundation is strong and earns enterprise trust because our security, durability, availability, and performance keep getting stronger release after release. Infinite means we will win when the demand is unpredictable, and to be clear, with agents, it will be, because our platform can respond in an instant. Individually, each of these is a meaningful investment, but together, they are the reason we believe MongoDB is the ideal platform for the AI era.
With that, I thank you. I guess I will say lunch is going to be in the foyer, and we are going to be back here in 30 minutes, right?
That is right.
Sweet. Thank you all for paying attention.
Please take your seats. The program will resume soon. Please welcome to the stage, Pablo Stern-Plaza.
Hey, everybody. It's great to see you. I'm Pablo Stern-Plaza, and I run AI and Emerging Products. It's really good to see a bunch of familiar faces here in the audience. Today, I'll cover why MongoDB will be a winner in AI. I want to start with what we're seeing and hearing from customers. We have lots of customers that are building on us, from AI natives to enterprises, and I've met with nearly 100 of them in the six months that I've been here. One of the clear messages is, in order to deliver AI outcomes at production scale, you need the right stack. We've been talking about the latent power of data for decades, and AI is changing that. It's providing insight, and AI needs real-time data to be performing. MongoDB is that platform.
As you'll hear in a bit, Emergent Labs, an AI native, picked MongoDB over Postgres as they look for an adaptive and flexible model to be able to build and scale their data. One of the U.S.'s leading hiring platforms is switching from Postgres to MongoDB because of its struggles with nested JSON. To show you how MongoDB wins in AI, I'm going to walk you through this in three parts. First, I'm going to show you how we're showing up for builders. I'm going to talk a little bit about how our platform is perfect for AI and building AI workloads. I'm going to talk about some of the stuff we covered a little bit earlier today in terms of how we're bringing customers into production faster for agentic workloads. Let's start with the builders.
The way builders find and choose technology has changed a lot. Builders are learning about exploring and choosing technology in wildly different ways than they have in the past, from answer engines to coding agents. We show up in many different places, and the tooling matters underneath them. We're investing to make sure that MongoDB shows up well across all of these. For answer engines, as they are the conversational tools like a chatbot, our focus has been twofold. First, we're showing up in the answer, and secondly, making sure that you're driving accuracy. We've invested heavily, and we're seeing very strong results here. Over the past year, we've improved accuracy by over 50% in prompts that test high fit scenarios from MongoDB, such as if you're looking for a managed database service. It's important to show up and to show up accurately.
Now, think about the AI frameworks. This is really about having the right integrations to the development frameworks that you're using to build code on top of. It's important to be in the right place for these because those end up being the selection for the rest of the stack as you build new workloads. Here we have deep integrations with the popular frameworks like LangChain and LlamaIndex, and we're continuing to add new ones as new emerge and gain popularity. You'll hear a bit more about this from Erica in a little bit. Our coding agents. This is an important one. They're critical to how applications are getting built and how they're going to be built in the future.
To make it easier for coding agents, we introduced Frontier Marketplace plugin into Codex, Claude, Grok Build, Devin, Cursor at our build test in mid-August, and we are seeing really good traction on those. Lastly, AI platforms. These are the prompt engineering platforms that more and more developers are choosing to build on. Earlier this month, we announced integration to Vercel v0. Tomorrow we are going to talk about Monk, Retool and Emergent as new integrations, enabling developers and builders to build AI apps using Atlas. As you heard from Dev, a big unlock for many of these capabilities was our Managed MCP. As Dev said, we have a local MCP that we released last year, and that has seen good traction, but it requires overhead to manage, operate, and install. With Managed MCP, it is one click. We remove the friction.
That is important for builders and developers as they are trying to get and build those applications quickly. Because of all this innovation, we have seen tremendous traction since launching this just a few weeks ago. The second thing I want to hit on is why we are perfect for AI. I will start building on some of what Ben laid out before me. We are built on the JSON document model. As Ben said, this is how AI thinks, it is how it interacts. It uses a document model. On top of that document model, we have native retrieval capabilities. That is really critical because without accuracy, there is no trust. Without trust, you are not going to put stuff into production. Our AI retrieval capabilities, full text search, vector search, embeddings and re-ranking are what drives that accuracy. Let me take these in turn.
As you know, many people start with full text search, the exact keyword match. It is fast, it is precise, it gets you the answer for what you are looking for. Vector search took that a step further because it understands user intent. It can find information based on semantically similar meanings for terms. For example, if I am looking for car repair, then the exact words match the query. Those two things are really important and drive value. The unlock really is around getting into embeddings and re-ranking, which is part of our Voyage AI acquisition from February 2025. This is about driving the quality of those results. Vector searches are only as good as the embeddings behind them. Our Voyage models consistently rank at the top of Hugging Face's retrieval embedding benchmarks, consistently beating models from Google, Cohere, and OpenAI.
Our re-rankers take those results and improve the accuracy and relevance. If you think about it, if you are in a trawler in the sea and you are fishing, you grab a whole host of fish. That is where the embedding helps you out. Picking out the best fish out of that net is what the re-rankers do. Because of our investments in research and in our vector search capabilities, we have seen a 4x increase in the customer count using vector search in the past two years. Our Voyage customer base is doubling quarter- over- quarter and has for the last two periods. We are seeing that momentum with customers. Take Financial Times, one of the market leaders in journalism. They use MongoDB as their content management system.
They have taken our Atlas Vector Search and the automated embeddings that we just released seven weeks ago, and they are using it to find the right articles at the right time for their AI searches. It is not just enterprises. Take AI natives too. Gravity AI, one of the AI native ad networks that is helping drive relevant ads in AI responses. They picked Voyage AI to surface the most relevant responses quickly for any marketing campaign. What the CTO told me was, their internal benchmarking showed when they tested Voyage AI against other models, we were more accurate than the others, and that was their key selection criteria. What I love about these capabilities is that they allow us to have new conversations with AI customers. It creates this virtuous flywheel. Existing database customers extend into Atlas Vector Search and Voyage AI.
Voyage AI customers need Atlas Vector Search and then look at the database, and enable us to have database conversations with customers. The thing that is really exciting about all these capabilities on the platform is that they are also the natural home for memory. First, let me do a little grounding on how agents think and how memory works. On their own, agents are like goldfish. You ask them something, next time you ask them, they are back to basic. They do not remember anything. What changes that is memory. When an agent perceives what is going on, it does so in real-time, and it shapes what it sees. Those perceptions inform a plan. Then the agent takes that plan and acts on it.
The outcomes of the plan and the actions then are fed back into memory to build a knowledge base, and does so, and it can think better the next time. The process starts over and over again. Perceive, plan, act, repeat. Creates an agentic loop. Memory is what makes that loop coherent across sessions, time, and agents. It is memory that enables agents and agentic systems to make smarter decisions and take better action over time. Memory is powerful, and it needs to live close to the data. It is an extension of everything that you heard about from Ben and from me. It changes often, so it is perfect for JSON and document models. Compared to PostgreSQL and JSONB, as you heard, we perform better in memory cases around large documents and we are much easier to work with.
The right memory requires the right moment with the right accuracy and precision, which is truly about retrieval. As an example, Dev talked about one media company that is using us. Another large U.S. media company that is using us for memory is taking their entire catalog of movies, series, and is embedding them using MongoDB, and is using that to then go and find and quickly surface the right and relevant use cases where something appears in one of those videos they can then use either for an ad placement or for doing some research. Or Metal, an AI native that many in the audience may be familiar with, that is building context layer for investors, uses us for memory context, including all past messages, all user preferences, and all context traces. You see, one thing that technology-forward companies are realizing is that agent memory is critical.
It's going to become one of the core systems of record of the enterprise. It's a core source of knowledge, and it's evolving source of knowledge. That positions us perfectly because memory itself is evolving. We're staying ahead of the curve, and we're introducing some new capabilities for agents. This is where I'm going to spend the rest of the presentation. Today, we announced Atlas Agent Engine. This is one of the largest expansions of our platform in company history. Agent Engine is our open, secure, governed agentic infrastructure built for builders and designed to run anywhere. When we started talking to customers about building our agentic stack, we heard three major unsolved challenges in the market. Agent Engine is specifically designed to address those.
First, as I discussed, memory is critical, and having an operational enterprise-grade system that can store the data is crucial and a gap in many of the solutions out there today. Our solution automatically enables long and short-term memory to bring the right context at the right time. Today, if you look into the market, many customers need to build the memory logic themselves or connect multiple building blocks in order to enable that. We give them a system that does it all. They own their data, and they have full visibility into it as well. It truly enables them to build knowledge and a system of record. Second, from a security perspective, one of the things that we heard consistently was many of the systems today don't have the right level of security isolation for their workloads and the data.
This puts a lot of onus on the developers and risk into the company in terms of ensuring that you're building the security correctly. Atlas Agent Engine uses Atlas's enterprise-grade security framework as a starting point that is already trusted by the world's largest enterprises. We've extended it to include governance and guardrails built-in on first principles as a design point into the system, so it can run safely and fully auditably. Thirdly, as Dev said at the beginning, many systems are closed because they focus on their cloud, their LLM, their framework. We've taken a completely open approach. We enable customers to run across any cloud or self-manage.
We let customers choose which model they want to use, and we are enabling them to use any framework or integrate into any observability provider or solution that they want and back into any data source that they want to use. To summarize, Atlas Agent Engine is a complete enterprise-grade data governance and runtime to bring AI agents into production. It has persistent memory with context, built-in security guardrails from the start, and it lets the builders choose their LLM, their cloud, and their data sources. As I'm on stage here today, we've already had many customers in our design partnership across multiple different industries giving us feedback, and the feedback has been tremendous. Take Paysafe. They're streamlining their payment processing business with agents that investigate payment surfacing errors and bring them to analysts to explore.
Or CDK, an automotive retail software company that is cutting their reliance on third-party data sources by having the agents extract rates, rebates, lease terms to be able to calculate car deal payments directly. So far, the feedback in the preview in the design partnership has been great and tremendous. We are also building alongside our partners. We have global systems integrators that are coming to our customers and bringing customers to us as we design this and as we bring it out into the market. One of our big partners is Accenture. Let us hear from them.
At Accenture, we do multiple modernization journeys for our customers. MongoDB is established enough to be trusted for mission-critical transformation and innovative enough to stay ahead where enterprise AI is going. That balance matters particularly for Accenture because our delivery model has been multi-cloud, multi-platform by design. Our clients get genuine architectural freedom, not a trade of one lock-in for another. So when I say MongoDB is the right partner, that is a conviction we have already acted on commercially, not a view we are still forming. If you look at large transformations, the reason they take so long and carry so much execution risk is not usually the technology. It is getting the right information to the right people at the right moment across programs that spans years and hundreds of work streams simultaneously. The context of our agents is the most important part in this equation, right?
Surfacing relevant project knowledge, connecting the integration decisions. Atlas Agent Engine brings all these together in a genuine enterprise-ready stack. It has built all the required constraints that makes agents work in an enterprise, providing the right context. So that is how I am seeing Atlas Agent Engine accelerate a massive transformation we are driving day in and day out for our customers in the world of AI. The good news is this is already running live in a multi-year transformation engagement. Atlas Agent Engine changed the transformation sequence entirely. Intelligents can now operate where enterprise knowledge already lives. The value starts not in years, but in weeks. This is just not about the technology alone, but the governance, the architecture, the process, and the industry and function knowledge which we bring in.
When we come to a customer together, we have never let the customer down, and that is the Accenture's principle, right? Delivery first, and that is where MongoDB's culture and Accenture's culture fits in really well.
We really thank Accenture for their partnership. You heard those three themes come out in terms of being able to have that memory and context, the security guardrails and governance, and then being able to run anywhere. We work with many other partners, including Infosys. Infosys actually brought one of their customers to us as a design partner for Agent Engine, and they are joining us live today. We have Citizens Bank, one of the largest financial institutions in the U.S., who is using Atlas Agent Engine to be able to drive outcomes. To talk about it, I want to welcome to the stage Vikas Agarwal, SVP and head of Data Platform. Vikas, welcome.
Thank you. Thanks for having me here.
Let us start off. Why don't you introduce yourself and the bank?
I am Vikas Agarwal. I head up data platform and streaming services for the bank. I always say this, that the data flows through me, basically. My responsibility is to provide a faster and better platform, so that we can justify all the requirements coming for the data, and we provide the data as faster and accurate. Like Dev said at the beginning, the most key important thing is the quality of the data and how we make sure that we provide that quality of data. MongoDB is a great partner of us. We are in the fifth year of the partnership, so we finished five years with MongoDB. Personally, I am working with MongoDB from 10 plus years.
I love MongoDB as a database. We are using MongoDB Atlas. For me, MongoDB is not just a database. When I use MongoDB Atlas, it is a complete framework for a faster delivery and better resiliency, and also best performance provided database in the system. We build our technology around it. We are using it in so many places. The database itself is used for many tier-1 applications in the bank, so pretty amazing.
Yeah. I think one thing that, a lot of what you hit on is what Ben talked about in terms of that foundation that we built on top of as we thought about Atlas Agent Engine. Talk a little bit, you guys are a tech-forward bank, and you are constantly looking at the market, so can we talk a little bit about how you guys are looking at the agentic landscape?
Yeah. Like other banks, we are also going very deep into this agentic landscape. We are building lots of agents for whether it is for our customers or for internal use. For me, the primary goal is to build the agents for my platform so that I can deliver faster and provide better support for our end users and the business. We are building so many solutions, which is going for patent as well into that space. I can talk, I think it is already out for the patent, so I think I can talk about it as well.
We are building a multi-agent framework. We are kind of the first bank who are building that multi-agent framework where agents are talking to each other internally. What they are doing is, if they are catching all the events in the platform, then multiple agents will work together to identify the problem, and they will go and provide the solution to a human, and then human can go and fix the problem pretty fast. That way we are putting more resiliency into the platform. We saw that it reduces our overall incidents and things are moving faster. Anybody coming up with a change, it is going faster because of those agents. Those kind of use cases is going on, and that is where we are with the agents.
When you and I first talked about Atlas Agent Engine, you were really excited.
Yeah.
You want to share what excited you, given the fact that you have been looking at this landscape for some time?
Yeah. Since we are building or, since I or my team is building the agents, and the bank is going on from quite long time, but me and my team had start looking into that direction in last one year. The one thing which we realized, if you think about the agents, it evolved too much in last one year. Even last night, I was sitting with some people, we were talking about it. Three months back, I was sitting with the same people, we were talking about that Claude is the best LLM. Today my statement will be like, Codex is the best LLM. It is always going back and forth between these LLMs. Everybody, what they are doing in the technology is basically they want to build something which is LLM agnostic.
You can switch the LLM depend on what you need and what LLM works best for your need, whether it is Claude or OpenAI or Copilot, the Meta is coming out of this news as well. Based on that, we are also trying to do the same thing, on top of it, we build some SLMs as well on top of it so that we can use our internal models for a small task or the task which we try to do with, where we have knowledge and we know what we have to do instead of going to the LLM all the time. When I was looking into that Atlas Agent, I was still in touch with Magenta.
Yeah. I was going to come with the name, it's fine.
Yeah. When I start looking into it, the one thing resonate with me was the memory feature, which we are also using in our other system with MongoDB. I think we built something similar four years back with MongoDB, and when it came out with the LLM, it becomes really useful because what I feel is, like I said, the knowledge is there. We have those sort of knowledge. We no need to go to LLM all the time. But the things which we can do within the memory, that works perfectly well for us. And that's where I think it provides more value because the performance is pretty good because when you go and hit the LLM, it always takes some time to come back with the answers. But when you go and directly hit to the memory layer, it's pretty faster.
It's quite faster than hitting to the actual LLMs. I think that makes perfect sense to use for that memory layer. The second thing is it's platform agnostic. Since everybody have this question in their mind, I think people sitting here also have this question, that every single tool outside or the software company outside in the market have something similar, right? Or they are saying that they are building something similar. So how this is different? The difference is it's not tied to the MongoDB. It's not like you have to use MongoDB. It's a completely agnostic framework where you are using a vector search of MongoDB as a store. And no matter what tool you use today or what AI platform you use today, you need to build something as a vector store or the store where you can store your data.
Even we as a bank, we also use MongoDB for vector store for our existing ones. I think using that agent-
Agent Engine lets you enable that.
Enable that Magenta for me, yeah. Enable it.
It lets you get across multiple different data stores, be able to have that memory and context.
Exactly
That comes straight. I think the other thing that you mentioned also is the platform-agnostic pieces. You can run air-gapped, you can run it in the cloud.
The containment of the agent is pretty much required, the security, the guardrails. We are using three layers of security when we build the agent, so it will not touch any of my production data. It will not go and do anything in my production data. There are three different set of guardrails. That's why reaching to that point is always tough for the agent. But when it goes in the air-gap memory layer, which we built with MongoDB, going in there, getting the answer will become more easier and quicker in that way. It's become more easy. Also the heterogeneous platform side, MongoDB itself can go and connect and collect the information, what we need, then it can feed into the agent if required.
And maybe one last question for you. Infosys introduced us. Can you talk a little bit about how you see the partnership between MongoDB, Citizens Bank, and your partner Infosys?
Yeah. Infosys is a great partner, and our partnership is going almost 10 years plus with the bank, and they are our strategic and implementation partner in the bank. We use them for any new thing we onboard, and we do the partnership together. It's our tight partnership that provide a lot of value in it. Because for us, we don't want to spend a lot of time into the new tech because there is a lot of new tech coming to the market. First, they help us to identify whether that tech is useful for us or not. Instead of we go and find out whether that tech is useful for us or not, they do their own research. They can always come back with the great research and tell us that whether we should invest in this tech or not.
That's how the partnership goes.
Yeah, and it's a great plug for Erica's section coming up in a little bit in terms of what we're doing with partners as well. Vikas.
Thank you so much.
Thank you so much. Thanks for the partnership. We're really excited to work together to build our agentic future.
Yes, absolutely. Thank you.
Thank you. It is great to hear the value that Agent Engine is bringing to customers. We are just getting started. Tomorrow, at .local, you will hear from CDK. You will hear from partners like Infosys and our ISV partners as well, like Fireworks in a fireside chat. Right now you are probably thinking two things. How is this going to contribute to growth? How does it stack up against the alternatives? The beauty of Agent Engine is that all existing Atlas customers are already entitled to it. No new contract, no new sales motion. All Atlas builders can start using it today. That is tens of thousands of Atlas customers who just got access through our public preview. It is not just one feature. It is a stack of agentic capabilities and components. That means that it enables you to build, deploy, and run these agents.
As each component solves a problem, it also creates a monetization opportunity. We have new opportunities to monetize our runtime and our memory. It also pulls through our data foundation, which will drive consumption of those products as well. There are, of course, many technology solutions out in the market. However, we believe Agent Engine is different because we have looked at the market, and we built it on first principles from what we saw. First, I want to emphasize, we are building for our core persona, the builder. As you heard from Vikas, memory is critical, and an operational data layer, OLTP, is truly critical path for agentic applications. Agent data grows, and agents need to access their data in real time and at scale. That is why they need a proven and mature OLTP database right at the center.
Layered on top, our advanced retrieval, powered by Voyage, enables that memory and context that you heard Vikas talk about. Security was also a first principle for us. Many systems today do not have the right guardrails for both agents and the data that they access. Agent Engine addresses that. As you heard, the AI landscape is moving incredibly fast, and lock-in is a risk. Anything you get from a hyperscaler will be tied to their cloud. From an LLM provider will be tied to their models. We do not lock you into any cloud, any model, or any environment. We truly let you run anywhere. We believe Atlas Agent Engine becomes the agentic runtime for building, deploying, and scaling AI agents in production. Let me close where I started. Why MongoDB wins in the AI world? It is three things. One, we are investing to show up for builders.
We have always been a builder platform, and we are going to continue to invest to ensure builders love us. Second, we have the perfect platform for AI. From JSON and the document model, our run-anywhere flexibility, the real-time operational core that we have, and built-in retrieval that is proven from AI natives to enterprises. Third, as you heard, we are expanding our agentic stack with Agent Engine, combining memory with full agentic infrastructure, including security, governance, and an openness for enterprises and AI natives to put agents into production. Thank you. With that, I am going to welcome Ben to the stage to have a little customer conversation.
Thanks, Pablo. I think we have the slide. Yes, we do. Cool. We love talking with our customers. We have another customer here from Emergent. Madhav, please come up on stage and let's have a talk. Thank you so much. Before we get into the questions, why don't you take a moment, introduce yourself to our friends here?
Yeah. Hi, I'm Madhav. I'm the co-founder and CTO of Emergent. At Emergent, we are a platform which lets non-technical users build production-ready full-stack applications, both mobile applications and web applications. Even if you don't know programming, you can come and build apps on Emergent and deploy it and then run your business on it.
Yeah, that's great. You've seen tremendous growth in the last few months. Why don't you tell us a little bit about that?
Yeah. Just to give you some scale. We have 15 million users. 18 million apps have been built on our platform. We use MongoDB. All the apps that get deployed on Emergent uses the MongoDB. Around 150K active paid deployed apps on Emergent right now.
That's amazing.
Yes.
Before you arrived at MongoDB, even before we talked about how you looked at your database selection or even your entire stack selection, what was your approach when you originally founded the company? What did you think, and how did that change over time?
Yeah. I think a lot of the session in this room had the word AI in it, and you're probably already tired of listening to AI. But the thing is AI doesn't do anything for you if it's not friendly, right? Imagine somebody who's a domain expert but doesn't know any programming and they still want to build an app. How do you help that person? How do you help this small business owner? So our approach was, hey, how can we make a platform where users do not have to be expert in coding, they have to be expert in their own domain and still build their business-critical apps. We abstracted a lot of these things from them.
One of the design decisions we took, we said that, "Hey, we're going to use MongoDB for our database," because many of these users who come to our platform, they do not have a well-formed idea what they want to build. As they are building, they evolve their idea, they decide what new features to add, and MongoDB is very flexible in that, right? Your schema can evolve with your new features, and so it evolves with your idea. It's very friendly for this kind of coding, and then it also scales well. Once you deploy to the platform, it scales well. So it's best of both worlds.
Well, we obviously thank you for being a customer. We love working with you. You didn't start out on MongoDB, right? When you first started the company, what did you start out on, and then what did you learn and made you come to MongoDB?
Yeah. Remember, we started as a research lab two years back. Our product launched one year back. As a research lab, we were experimenting with building our own coding agent harness from the ground up. We started with typical codes we had set up. What we were realizing was that it becomes very difficult for agent to sort of like. This is Claude 3.5. Intelligence hasn't yet caught up. Claude 3.5, back in the day, if you gave a lot of the migrations-related problem, you'll run into these kind of migration issues where it has to kind of maintain the database as you go along, right? It's not just a programmer that you have to bring in. You have to bring in a DB admin. You have to encode all those persona in it.
What we realized very early on that a lot of those headache goes away if you switch to a document-based database. Part of the reason is that conceptually, an app very much is like these JSONs, right? If you're familiar with technologies like React, they talk and think in terms of JSON, right? The translation there is very simple. Whatever your app functionality is in thinking in terms of JSON, the same data representation sits in your database. It's also very friendly for backups, if you want to backup, if you want to bring it back in. So we are a very eval-driven company. We ran our evals, we figured out, you know what? MongoDB is the way to go. That's how we took the decision.
Yeah, I think there's a perception out there that agents, because they're just agents, don't necessarily care about complexities of stacks. I think you've proven that wrong.
Yeah. I think a lot of people miss that in this world, agents is definitely important, but there's a human component to it as well, right? So a programmer using an agent is different from a non-programmer using an agent, right? So a programmer might be able to say, "Hey, you know what? That migration did not go through. Can you roll it back?" But a non-programmer will not be able to even articulate these things, right? So keeping the audience in mind who are non-technical users, they are building these agents. How do we make it friendly for them? That is where this design decision came for us.
Yeah. Well, a lot of your customers aren't developers, right?
Yes.
They're typical people that are just saying, "I have this idea.
Yeah.
It's great. But that creates all that inflexibility you talked about before. But then you also have customers that have 300,000 lines of code, like big actual enterprise apps.
Absolutely.
How are you approaching just your tech stack decisions or even managing the database now because you have to think about the $2 customer all the way up to the million-dollar customer?
Absolutely. I was just asking one of my engineers what are some of the scalable apps built on our platform. There are users who are building apps which is scaling like 5 GB per week on MongoDB. They have 15 million documents. We have users who have built pretty scalable apps on MongoDB. The whole idea is that it has to scale with your ambition. We build Emergent with this concept in mind that it's not just for UI prototyping, it's not just for building toy apps and demo apps, it's for building production-ready business critical apps that you would actually ship to your end users. That's where the enterprise nature of MongoDB comes in. We have the great fortune of collaborating with MongoDB on various platform features.
For example, how do you have APIs to spin up a database or to bring them down? With those platform features, we are able to scale our platform.
That's amazing. As you think about agents and better memory and retrieval inside of the apps that people are building, how has your collaboration with MongoDB been able to help with that? Are you seeing a dramatic difference from what the market sees from a retrieval perspective?
Yeah. I think fundamentally, it is like when people start building these things, one of the design principles that we have is we want to avoid complexity.
You do not want to bring in five different technologies to accomplish the same thing. If your database already has things like text search, vector DB search, et cetera, then you do not have to bring in a third integration to make that happen. This simplicity allows for agents to work well, the humans to sort of keep everything in their head and then scale as well. We are literally making use of all the new features that are coming on MongoDB and basically making them available for our customers.
No, that is great. We are super excited about that as well. As your deployment rates have been increasing, and you start looking at what next year is going to bring for you, what do you need from an infrastructure partner or a vendor? What is the most important thing to you?
No, I think one of the things that is very important for us is just the platform capabilities of meeting the uptime, meeting the scale requirements. The cost is also important for our customers. Some of them want to make sure that you can scale down to zero if it is not being used and things like that. I feel like with the partnership with MongoDB that we have, we are kind of pushing each other. We have some requests that we bring it to you, and you have some requests that you bring to us. We are very fortunate to be in a situation where we can push each other in this AI native world.
Yeah, I agree. We're super lucky to have you. Last question. AI agents are writing more and more of the world's software. What does that actually mean for the database from your vantage point? You're in a unique position. You see all this inbound, you see all this activity. Are you prepared for what's coming next and-
I think we are very excited. I think anybody who is in this room, they should just go and maybe sign up for Emergent and just build an app. You get this kind of not just see it in theory, but see it in practice. It's kind of mind-boggling the kind of apps that people are building. We have people who are building mobile apps, which they've never written a single line of code, and just prompting and building mobile apps, shipping production-ready apps. Very excited to be in this world.
Well, that's so great, and thank you for sharing your story with us, and appreciate you coming.
Thank you.
Thank you.
Yeah.
Now I believe we are going to welcome our CRO, Ryan, up to the stage. Ryan.
Thank you, Ben. It is so great to be here this afternoon and also to see some familiar faces. For those of you who I have not met before, I have spent my career building and scaling enterprise go-to-market systems across consumption and cloud-native platforms alike. The real question I am sure you have for me today is why MongoDB and why now? There are really three key reasons behind my decision to join MongoDB. It is people, tech, and timing. On the people front, we have one of the strongest operating benches in enterprise software today. Everyone on this leadership team has built and scaled global platform companies. Together, we have over 100 years of experience. On the tech side, not only does our JSON document-based model align well with the frontier labs, but specifically the way that we store and retrieve application context, it closely aligns with how agents consume information.
Last, we are trusted for mission-critical workloads across financial services, healthcare, public sector, and AI digital natives. Whether it is JPMC, Wells Fargo, or Santander for payments or delivering global services for governments around the world, we are the trusted platform of choice. The last is timing. AI tooling with partners like Cognition unlocks the modernization and migration of legacy Oracle, SQL, and relational database workloads. We can do this now in days, not weeks and years. A year ago, these capabilities were just getting started, and now they are real. I have to tell you, they are very real. Not only are we winning new AI workloads, but AI opens up a new TAM to over $10 billion to our teams to go after.
For the reasons above, I truly believe this is the best time to not only be a MongoDB customer, as you heard on stage here, but equally be a leader at MongoDB. We are driving parity between Atlas and EA. We are delivering more core platform capabilities and product, and doing so with greater velocity and greater reliability than ever before. As many of you know, I was also the CRO at Confluent previous to the IBM acquisition. While there are many similarities, specifically around consumption and the equivalent of an EA or self-managed capability, workloads on this side, streaming projects on that side, this is a much larger opportunity. Without question, the market is coming to us, especially with AI after a decade. Before we talk about the investments for the future, let's quickly take a look at the progress we have made to date.
With Q2 earnings behind us, we have now delivered six quarters of consecutive Atlas growth at 29% year-over-year. Our self-serve continues to thrive, adding 2,500 - 3,000 customers a quarter. We had a record of net new customers in Q2, totaling over 70,000 customers at MongoDB. I never had this at Confluent. 17% year-over-year growth in our $100,000-plus ARR customers and our $1 million-plus ARR customers. I want to be super clear. This is a story of strength and momentum. We have seen tremendous traction over the previous six quarters, and now it is time to build on that momentum the team has created. To drive the next wave of durable revenue growth, we are making three strategic investments. We will focus our go-to-market efforts on the enterprise and executive buyer and our AI and digital natives.
We are going to evolve our go-to-market system as we extend our capabilities and our platform with new products by adding product specialization, all those resources across all three global regions. We will be doubling down on our application modernization platform. With our AI partners you heard me mention, we will continue to identify and alleviate the modernization and migration pain that has plagued all of our customers for years. The path to $5 billion-plus is through the enterprise executive suite. We are now evolving beyond a developer-led bottoms-up motion, and I want to be super clear here as well. This is an and, not a replacement. By adding equal focus on the enterprise suite that is driving enterprise-wide AI buying decisions. We will drive greater executive alignment by being laser-focused on our segmentation.
We will drive executive briefings and events, increase our forward invest for some of the new capabilities you heard about today, as well as we will increase our incentives to our sellers to drive larger, more complex deals. We will go deeper in our accounts with our vertical depth, financial services, healthcare, public sector, and there again, a special focus on AI and digital natives. Regionally, we will be investing in U.S. Fed. We are actively pursuing FedRAMP High, and we will have some announcements shortly. At only 10% penetration today, we have nothing but headroom to grow. We will be hiring a new U.S. Fed general manager to accelerate the recent acquisition of Clarity. We will be making strategic geo-investments in APJ and doubling down in Japan.
We are establishing a new entity and hiring a new Japanese country leader, fourth-largest economy, and it's been somewhat untapped by MongoDB, although you would be surprised at the current ARR without even having feet on the street. We expect this to be accelerated by a strong partner motion with my partner, Erica, as well. We also need to build our go-to-market muscle around product specialization as we extend our capabilities of the platform with the addition of new products. Specialization will allow us to go from incubation to scale with every new product without slowing our acceleration that we've had of Atlas and EA. We will maintain a strong core sales motion while driving the sale and adoption of new platform capabilities and products with this specialized team.
We have developed a model now with specialists focusing on products like Atlas Vector Search, embeddings, and re-ranking that will allow us to repeat this and scale with every new launch. As you heard today with the announcement of Atlas Agent Engine, we will also increase our investment here as well. There's more to this than just adding resources. Specialization allows us to fine-tune our value messaging, our demos, our pricing and packaging, and equally to ensure we have a fast feedback loop to our product engineering teams as we compete. Here, there's no better prospect than an existing customer, and with over 70,000-plus customers, we have an enormous upsell, cross-sell opportunity for us to capitalize on. Lastly, modernization and migration. This represents a massive opportunity for this team. AI tooling lowers the cost and risk of legacy migration.
We can do this now, as I mentioned, in weeks and not years, and it doesn't require armies of bodies as it has in the past. Equally, this modernization challenge has plagued our enterprise customers, as I mentioned, for years. We are actively migrating Oracle SQL and other relational workloads to MongoDB via our MongoDB Application Modernization Platform and with partners like Cognition. Our path to driving durable and consistent growth of $5 billion and beyond is clear. Engaging with the executive buyer, leaning into our vertical strength and geo-specific investments. We'll be strengthening our commercial muscle through platform specialization and capturing the modernization migration wave all of our customers have been waiting for. We have the right leadership, now the right go-to-market system to drive the next wave of durable growth here at MongoDB. Before I turn it over to Erica, Dev is correct.
We have the world's best go-to-market machine, and I have absolutely no question in our ability to drive and capture the next wave of growth for MongoDB. I'm now going to turn it over to Erica to go over our strategy with our partners.
Okay, Ryan. Thank you. I am thrilled to be here with all of you at Nasdaq. For those who I have not met, my name is Erica Volini, and I am the Chief Customer Officer here at MongoDB. In this role, I am responsible for ensuring an outstanding customer experience, from technical support to customer success, to professional services, to our partner ecosystem, and that is going to be my focus for today. By way of background, I spent over two decades at Deloitte running one of their five global consulting businesses. In that role, I had a front-row seat to understanding what it takes for customers to truly drive digital and business transformation.
After Deloitte, I pivoted from consulting into the world of tech, joining ServiceNow, where I was responsible for scaling the partner ecosystem and ultimately their global go-to-market at a very strategic inflection point as they went from $5 billion- $10 billion. The through line of both of these experiences has been a deep focus on why and how customers make the decisions that they do. That is something that I hold front and center as I focus on ensuring that we deliver an outstanding customer experience for our over 70,000 customers. Part of bringing that remit to life is building a strong, trustworthy, and innovative partner ecosystem because we know that ultimately no one can win the AI era alone.
We need MongoDB in every room where enterprise decisions are being made, no matter where a customer is on their agentic journey, whether they are modernizing their legacy systems and data, whether they are building agentic features into their applications, integrating agents, data, and systems so that they can act in concert, or scaling that capability across the business. Partners are guiding their customers at every step in the journey, and our goal is to be the platform they consistently rely upon. To do so means intimately understanding the expectations of our customers, expectations that have fundamentally shifted in the last few years. Let us start with values. Customers no longer measure a company by the outputs they deliver. Today, customers measure us by how we impact their business outcomes. That means being deeply ingrained in their business, in their technology landscape, and being core to driving their AI transformations.
Customers value business outcomes in large part because their AI initiatives are not pilots anymore. In production, AI has to work. It has to be secure, durable, available, and performant, delivering the business outcomes our customers are after. That is what our platform was built for. Our differentiators, though, they cannot just be about technology, because transformation today means more than a new system or a new platform. It means an entirely new way of working. To navigate this landscape, customers cannot just rely on a single platform or service provider. They need an ecosystem working together to help them succeed. That is why a strong partner motion here at MongoDB is non-negotiable. It is not because we cannot grow alone, we have proven we can, but because the fastest path to scale runs through partners enterprises already trust. We will see partners influence our revenue, our reach, and our innovation through four primary levers.
First, embedding our platform in every aspect of the AI ecosystem. Second, becoming the default embedded data layer inside the agentic stacks of the leading system integrators. Third, co-innovating with hyperscalers and Neoclouds. Fourth, becoming the intelligent data platform for industry ISV solutions. No matter what partner our customers choose, we want MongoDB to be the platform that powers their success. Our first lever is integrating across the full AI ecosystem. This is about making MongoDB easily and readily available within the companies that are redefining today's technology landscape. As you heard from Dev earlier, we look at the AI ecosystem across four major categories: coding agents, AI platforms, answer engines, and AI frameworks. Across them all, we are invested in showing up wherever a builder needs us, whether that's an agent or a human, like our connection with Devin, the AI coding agent from Cognition.
With Devin, a team staring down a decade of accumulated legacy code doesn't spend the next six months manually mapping every table and dependency. Instead, Devin leverages MongoDB to work against live data instantly to transform the code in days or weeks, not months or years. Vercel is one of the largest distribution channels in the industry for how applications get built. Millions of developers use Vercel's v0 today. MongoDB Atlas is now a native option in that flow, thanks to our managed MCP provision in a single click. So the agent building the app can query live data directly as it works. Mastra, fastest-growing TypeScript framework for AI agents, now runs all of their data retrieval and memory on a single MongoDB Atlas cluster. That means a team can ship a production-ready agent without assembling and maintaining a patchwork of infrastructure.
Each of these partnerships helps us embed MongoDB as the preferred data platform from day one. LangChain is a perfect example of how this motion translates into both revenue and reach. Our customers can build generative AI and RAG applications on MongoDB through LangChain today. The partnership is still relatively young, but the revenue tied to LangChain is up more than 460% year- over- year. Builders are reaching for MongoDB inside LangChain projects 220,000x each week. As important as each of these partnerships are individually, their true strength comes from the ecosystems that each of them has built, including our next lever, system integrators. All of the leading SIs, as you know, have built their own agentic stacks, often positioned as frameworks that incorporate the best-of-breed technologies required to drive a customer's transformation end to end.
So when MongoDB is embedded as the data layer in these stacks, we get invited to transformations driven by the top of the house. Together, these top six systems integrators touch nearly every Fortune 2000 company on the planet. Gartner estimates that enterprise AI engagements with SIs will increase by 15% by 2029. That's the access that these partnerships hand us, and that is why I am so excited that all six of these organizations have not only named MongoDB as a strategic partner for agentic delivery, but have already started to deploy Atlas Agent Engine internally. The results are showing. One of these partners has already seen 984% higher query speeds and a 63% latency drop when testing Agent Engine against sovereign AI offerings. Those are metrics that matter. But these partners don't just see the value of MongoDB for their internal use.
They have also integrated Agent Engine into their agentic stacks that they're bringing to their customers, and they have identified use cases that span the entire enterprise, from an agentic assistant for the oil and gas industry, to an agentic business loan platform, to agentic ticket management and technical support, and agentic knowledge hubs that can search across policies, procedures, and even medical documentation autonomously. Listen, I could go on, but what we know about these use cases is they span in all industries, and they're going to play a critical role in helping us drive that vertical depth and going deeper and wider across the enterprise. Two of our go-to-market priorities that Ryan just spoke about. If you want to know why are these partners believing that MongoDB needs to be an essential part of their agentic stack? Please look no further than the quotes right here.
These are quotes not just from individuals at these partners, but the leaders of their global AI practices. They have become not just business partners but close collaborators as we build and scale for this next era. That leads me to our third lever, hyperscalers and Neo clouds, the compute power that this next era relies upon. Nearly every workload and enterprise runs, every model they train, every transaction that they process sits on someone's cloud. Ben talked earlier about our Run Anywhere strategy, and that's what gives customers the flexibility to control where their workloads run and where their data lives, which means we need strong relationships across cloud providers of all types. With so many customers running Atlas on hyperscaler infrastructure, our teams are constantly collaborating on joint solutions and products.
Roughly half of all MongoDB revenue is supported by a hyperscaler partner, and we're seeing marketplace revenue growing faster than overall revenue year- over- year. That success shows up in the recognition we've earned. We've recently been named a partner of the year by AWS, GCP, and Microsoft. At GCP specifically, we received partner of the year recognition for the past seven consecutive years, and we're one of the top three Azure partners. I'm committed to continuing and expanding this streak, but as we all know, hyperscalers are not the only cloud business in town. The cloud provider market has expanded exponentially. We already work with many of the names you're seeing behind me, and we are actively pursuing relationships with others to ensure that wherever a customer's capacity lives, MongoDB meets them there.
Specialty in Neo clouds are reshaping the economics of AI infrastructure and moving up the stack. At MongoDB, we are uniquely positioned to support the scale, the speed, and the data requirements of AI workloads. Our strategy is not only to launch new NeoCloud relationships, but to evolve with existing partners as their platforms mature. As DigitalOcean, for example, expands from GPU infrastructure into an integrated AI-native cloud, spanning compute, inference, data, and agents, MongoDB can act as the foundational data layer for these emerging AI applications. Our goal with hyperscalers and NeoClouds is not just to integrate with them, but it's to co-build and co-innovate alongside each other, like with AWS, our largest cloud partner today. We're building with AWS and our own application modernization platform, known as AMP, to make legacy modernization dramatically simpler for enterprises stuck on decades-old systems.
We are also, and I am very excited about this, in discussion with both AWS and Microsoft about leveraging the memory layer that Pablo talked about of Atlas Agent Engine directly into their AI platforms. Having this type of robust set of cloud partnerships, this really comes to life when we see it at our customers. Ben, I do not know where he is, talked earlier about Nationwide and their regulatory requirements, but what I want to talk about with this customer is what was happening behind the solution that got them where they needed to be. Nationwide needed to modernize their payments with zero downtime risk. Working at this scale, it does not happen with a single vendor. They needed the ecosystem, MongoDB as the data platform, Accenture as the systems integrator, and multiple cloud providers running a single stretched MongoDB cluster across both simultaneously.
Nationwide now moves GBP 1.5 billion in transaction volume every 24 hours, handling 3 million payments at peak, and we went from zero code to fully in production in under eight months. That is the nature of an ecosystem win. We make each other stronger and drive better customer outcomes than any of us could do alone. As much focus as there is on developers building applications, we also know that many enterprises out there are buying applications to meet their needs. That is why the fourth and final lever is becoming the default data platform for industry ISV solutions. Because getting embedded in industry-specific solutions means that a customer that is won by the ISV is a customer won for MongoDB, and our install base grows even further. Let us talk about what that looks like in practice.
A security team leveraging BigID does not start their morning worrying where sensitive data might be hiding across a decade of acquisitions and half-documented systems. With MongoDB embedded, they run one search and see all of it, structured and unstructured, before an auditor or an attacker finds it for them. A medical technician using software from one of the industry's best healthcare technology providers no longer stitches together one system to run a test and then another to manage the data behind it. It is one platform, powered by MongoDB, as the data foundation to keep it real-time, accurate, and of course, safe. A customer messaging into an Ada-powered support line has no idea what is happening in the background. They just get a better answer than they did last week.
Behind the scenes, the model is learning from real conversations in real-time, and because of our Queryable Encryption, we never actually see what was said. So we are able to keep customer privacy paramount. Someone with no engineering background opens Emergent Labs, who you just heard from. They describe the app that they want to build in really plain English language, and they get a working product back, grounded in, you guessed it, MongoDB's data platform. In four months, that has happened nearly 2 million times, with more than 50,000 of these apps now in production. This is what winning through the ISVs looks like. We are the batteries powering the flashlight. Partners with deep industry expertise embed MongoDB in their solutions, so every customer bought into them is bought into us.
We have talked about these four distinct levers and how each one works to expand our reach and expand our revenue. But to me, the story really is not about each of these levers individually. It is about how they come together into one ecosystem. A customer on their agentic journey needs a full team behind them. They need a systems integrator. They need cloud providers. They need unique industry-specific solutions. And they need the best data platform in the world to underpin it all. Partners have the relationships, the trust, the industry expertise.
But what powers all of it, that is ours. We are not just building the products enterprises want. We are becoming the platform that partners build their own AI and modernization strategies on. That is the bet we are making on partners, and it is the exact same bet I made when I took the role of Chief Customer Officer.
We do not scale by doing more ourselves. We scale by leveraging the best people, products, and partners in the market to deliver what our customers need when they need it. Hopefully, that all resonated because we are going to take a quick 10-minute break, and when you return, Mike is going to talk about what all this goodness looks like in the numbers. Thank you.
We will resume at 1:50 P.M. Please make your way back to the room then. We will resume at 1:50 P.M. Thank you. Please take your seats. We are just about to resume.
Okay. Thank you for your patience, putting up with the warm room a little bit. I had to take my jacket off. It got a little too hot. Before I get started on the safe harbor and go through some stuff, just two things. One is, you know I like the pace, so if I am in the way of the slide, just say, "Get out of the way," and I will try to move. Jess did talk about, hey, these will get posted right when we are done, so you do not have to take a picture. You will have access to it. The other thing is, Jess said it at the beginning, I just want to reiterate. Folks, this is a big event for us, and it takes a lot of time, people, and effort.
It truly takes a village, to use that term, from Rita and Megan at the front that you saw there, to the finance team that does all kinds of great work, to the legal team that puts up with all of us. "Oh, no, we can say that," right? To the marketing team who helps put this on, and then, of course, Jess and Will and the investor team. Thank you very much. It's truly a huge effort. Before we get into the financial update, once again, just on the safe harbor, as we said before, we will be making forward-looking statements during today's presentation. These statements reflect our views only as of today and are subject to a variety of risks and uncertainties.
Please refer to our Form 10-Q for the quarter ended July 31st, 2026, and other filings that we may make with the SEC for more details on the risks. Thank you for listening to that. Hey, I'm going to key off what Dev and everybody else said, and hopefully you heard a lot of great things today. I know you did. Let me highlight a few of the things that you heard from the team. Dev's presentation talked about we compete in a very large, growing market, and we have, bless you, true discernible differences in the JSON native document model and our run-anywhere strategy. We talked to you folks a lot about that. That's what really wins us business when we're in there, especially with the enterprises.
Ben and Pablo did a great job sharing the exciting updates for our new innovations around Atlas, EA, and AI. Ryan and Erica gave a great overview of the go-to-market priorities as well as our plans to significantly expand our partner ecosystem. My job is to pull it together. What does it mean for financial metrics? We could possibly talk about the long-term model as well. As I go through my presentation today, I'm going to break it into five sections. I try to do three, but I had to do five, so you'll have to bear with me. We're going to talk about the strong track record of results. Then we're going to shift to our growth, why we believe it is real and durable.
I'm the CFO, so of course, I'm going to talk about operating margin and leverage, which is mainly driven by revenue growth. Then I'm going to talk about our AI focus and how we expect those product innovations to reflect in the financial model. Then after three and a half hours, we'll talk to you about our financial targets. Hitting on our commitments. Hey, folks, we talked about it at the beginning. We, as a management team, take this very seriously. Credibility is built one quarter at a time, and every quarter matters. We take our commitments very seriously. What we are going to show you is what we showed you last year, and then how did we perform against those. Many of you are going to recognize this slide.
We have included the commitments from last year as well as how we did for 2026 and then what the guidance is for fiscal 2027. Last year, we talked about Atlas growth. 20% plus in fiscal 2026, rounded up, we grew by about 29%, and our guide for this year is approximately 27%. While we did not specifically guide for Enterprise Advanced and other, the implied growth in the guidance that we gave was low to mid-single digits. In 2026, we grew at 7%, and in 2027 guide, we have now increased that to 11%. We noted on the earnings call, this is the first time in three years that we are projecting double-digit growth for Enterprise Advanced. That equated the total revenue growth commitment in the high teens. For fiscal 2026, 22.8%, and the guide for fiscal 2027 is 21%-23%.
This will be the second straight year of acceleration of total revenue to come in at the high end. Then of course, my next two favorite, operating margin commitment, and we will talk about cash as well. Our commitment was, on average, 100-200 basis points a year. We went from 14.9% in 2025 to 18.5% in 2026, and we are guiding to approximately 21%. That was 360 basis points growth from 2025 to 2026 and 250 for this year. We were super clear last year, the 100-200 was not a ceiling. We have talked a lot about the thing that I loved about Ryan's presentation is, man, there was a lot of investment in that discussion. It is going to result in great revenue at some point. We are continuing to invest in the business, and I will walk you through a slide on that.
We have a great business model, and when we drive revenue growth and still invest, it all falls to the bottom line. Then finally, free cash flow conversion. Jason, you have known me for a long time. Kirk, as you. I love cash. It is my favorite. Cash flow conversion has been exceptionally good the last couple of years. We talked about 80%-100%. Keep in mind that this is the business model is dependent on it in terms of how you collect from your customers. There is also this thing called working capital, and the team has done a great job managing both of those pieces. So fiscal 2026, the cash conversion, which is non-GAAP operating income, or I am sorry, free cash flow divided by non-GAAP operating income, 108%. Year- to- date in 2027, 108%, and we guided to the upper end of the range.
As you can see, sorry, I missed that one. We have hit every single one of our commitments. Again, we take this very seriously as a team to make sure that we do that. In addition to hitting our commitments, we also listened. I think in the first two months of my role here, we spent a lot of time on calls with you folks saying, "What are the things that you would like to see better from MongoDB?" Two things came up. "Please give us a view of what you are guiding Atlas when you do your guidance. Do not make us do the math." Starting in Q2 of last year, we started to give that view. The other ask was, "Please share the ARR growth for Enterprise Advanced and other.
Because of the multi-year revenue, it jumps all over and please share that. We started that in Q2 of last year, those two pieces. Then, of course, we introduced the long-term model, which hopefully has helped, especially you long-only investors, get a view of where we think the business is headed over the long- term. Okay, let's shift to growth. Something we all love. Dev had this same slide that he showed earlier, and we showed it last year. It's a very large market. IDC estimates it's going to reach $200 billion by the year 2030, and it's growing at a compound annual growth rate per IDC of 12%. There are very few markets that are this big that grow this fast. It was one of the things that I loved when I had my first conversation with Dev about joining MongoDB. It is a huge market.
Both of these, it includes both OLTP, which is obviously where we focus, and OLAP. We look at really the OLTP market mainly, but we also look at OLAP as potential areas to invest in the future. We still have a very small share of the TAM today. Based on our guide, it's about 2.2%. There is a ton of room to run in this market. As Dev, Ben, and Pablo discussed, we believe we have the best data platform for the AI era, and we're very confident that we have a long runway ahead of us in this market. Okay, let's shift to Mongo. This is the last several years of our revenue broken out between Atlas and EA. Over the last five years, total revenue compounded growth rate 28%.
On a market that's growing 12%, we are clearly gaining share, and that's a big part of our strategy. As you can see, the revenue growth has been strong and consistent. One of the great results of this is a metric we talk a lot about, which is the 122% net dollar retention rate. Again, that's total company across both of these products. Our view is we have two independent, healthy growth engines. Atlas, which has grown at a compound growth rate of 36% over this time frame, and EA and other, and we added services in there to get the full picture, growing at a compound growth rate of 14%. Both of those over the last five years have grown double-digit growth. It is not a single growth engine story, and we feel it is durable because we have two strong growth drivers.
This is one of my favorite charts after cash. We talked about this last year. Which is, even though Atlas in FY 2025 was growing very nicely, 27% year-over-year, we had flat lined as it relates to growing incremental dollars year-over-year. For me, when somebody says the word acceleration, this is the slide I think about. We have now, for six straight quarters, added more dollars year-over-year, each quarter. That is a great testament to not only the sales team, to the customer success team, to the technical services team, and I'm going to show you a lot of stats on our customer growth that plays directly to this. This growth has been across multiple vectors in our business.
One is, we do not believe it has been negatively impacted by the growth of Enterprise Advanced, and we've been super clear about that. We've also seen growth across both self-serve, our wonderful product-led go-to-market motion, as well as our sales sold. As we've discussed, the majority of the growth in Atlas in the last several quarters has been driven by the larger enterprises, specifically in the U.S. and EMEA. Erica touched on this. The marketplace offerings are a big part of our growth as well. All three of the hyperscalers are growing very nicely, and it now represents approximately half of the total revenue of MongoDB. This is a critical point because we get this question a lot. How do you compete against the hyperscalers? When someone deploys in the marketplace, everybody wins.
We win if we get a customer, the hyperscaler gets consumption, and the customers are able to burn down their commits. That's one of the reasons why it continues to grow. Atlas is the biggest driver of growth. Guess what? Enterprise Advanced also transacts in the marketplace as well. So it's a big driver of growth. As you look at that slide, those are the things continuing to push that up to the left. Okay, let's shift to Enterprise Advanced and other, our self-managed product. This is over the last 10 quarters, the same time frame I just showed you on Atlas. As you can see, we were really averaging mid-single digit growth until we saw an inflection in Q3 2026, and it's gone nothing but up at that point. Dev talked a lot about the industry drivers. Ben talked a lot about the product enhancements that we've made.
Very importantly, all the new stuff we just built, it just rolled out a couple of months ago, so it has not really been a big driver of this. This has always been a strong product in regulated industries, but now we're starting to see other segments, especially technology, as they're trying to deal with all the shifts in the industry, that they want to run it on their own and self-manage it. The ARR growth has now been double digit for the last three consecutive quarters, and again, that was before we added search and vector search in Q2 of this fiscal year. We get this question a lot, so let's just get it out there. There's also a significant overlap with Atlas. If you look at Enterprise Advanced customers with greater than $100,000 in ARR across MongoDB, approximately two-thirds of them also use Atlas.
This issue of, hey, there's a cannibalization. We talked about it on the last earnings call, we see that cohort actually growing faster than the total. So more MongoDB is better across either Enterprise Advanced or Atlas. Okay? So now let's take a look at some customer metrics, and I'm going to move over so you can see these. Again, folks, you'll get a view of these. This is our total customer growth. Compound annual growth rate, 23%, driven mainly by Atlas growth, over 70,000 customers. Now let's break that down into the specific cohorts. If you look at customers above $100,000 in ARR, it's grown largely in line with the total and a little bit more, but not much. Now let's go to the larger customers.
If you look at customers, and Ryan talked about this as well, with greater than $1 million in ARR, you can see that that cohort's actually growing much faster at 34%. I like to do this because I'm going to give you a couple of new pieces of information, and this is one of them. This is the growth of the cohorts above $5 million in ARR, and it's growing at 57% compound annual growth rate. When I first started, this was the move that Dev and Cedric made to move upmarket to drive more growth, and we'll talk about it. Folks, we get a lot of questions about, "Hey, can you walk us through the workloads and how they're scaling?" My answer to that is typically, that doesn't matter, that does. Because they may deploy a workload.
Our job is to cross-sell, up-sell, and expand that customer, and this is what the move-up market has done. Now, you say, "Great, Mike. Hey, that's customer numbers. What does it mean for revenue?" This is another new one. I'm going to let you look at it. This is ARR starting in Q2 of fiscal 2018 to the quarter we just finished, and it shows you the growth in ARR by customer size. You can see the durable growth, especially in the larger customers, continue to drive the revenue growth. This is the move-up market. This is the platform play. This is AI starting to take shape in enterprises that's driven such a big growth. Our job is to continue to get them to expand with MongoDB, even if their early workloads start to plateau. This is a customer play, not a workload by workload play. Okay?
I just need to get a drink. Hold on a second. It's a great chart, isn't it? We had a bunch of different colors on there, and Mae said, "No, Mike, they're going to throw up if you throw purple on there." I put purple on there because it's my granddaughter's favorite color, and she said, "Hey, Grandpa, you're going to put purple on there." I'm like, "No sorry, I got to go with green." Okay. We have a lot of conversations with you folks. The number one, actually number two, ask for metric, give us total ARR. You tell us what Enterprise Advanced is. I've said repeatedly, Atlas growth largely mimics revenue growth, mimics ARR growth, and the ask was very simple. I cover a lot of companies. Don't make me do the math. We did the math for you.
This is subscription ARR ending Q2 2025, 2026, and 2027. Okay? As you can tell, it has accelerated each year, and they're both accelerating. Atlas growth consistent with revenue growth. We've given you this number before from a growth rate perspective, but there are the dollars. Spoiler alert, you're not going to see this every quarter. You're going to see it every year. Take a hard look at it. We'll talk about growth in revenue, but again, folks, Atlas revenue growth largely consistent with ARR growth, and we give you the Enterprise Advanced number as well, but those are the dollars. Hopefully that helps. For us, this is a big part of the Run Anywhere story and the importance of self-manage, because it is another growth driver, and it helps us drive total revenue growth, which is what I care about because it funds everything that we do.
Okay. What does that mean for margin? This is another great slide. This is the revenue growth versus the operating margin. During this time, we've increased, and we talked about it, 360 basis points last year. At the upper end of the guide, 250 basis points growth. That revenue growth drives all of the operating growth. At the same time, we've invested in the business. Last year, OpEx was up, call it mid-teens. This year in the guidance, it's almost 20% growth, and R&D is 30% in our guidance. The result of that 30% is what you just heard from Ben and Pablo in terms of the product innovation. Over time, we'll talk about this when we guide 2028.
There are areas of optimization, AI, offshoring, and other things, but our goal is to continue to invest in growth because revenue drives operating margin expansion. What are those areas we're going to invest in? Ryan went through a bunch of these in his, and you heard this as well from Ben and Pablo. Most of it has been on the core database and meeting our customers where they are as they make their AI journey. No surprises here in terms of what we're investing in. We haven't talked a lot about Voyage AI, but we're also obviously investing in Voyage AI as well. This is rolling out 9.0. It's Atlas Infinite. It is our MCP, managed MCP server, and then, of course, new products like Atlas Agent Engine. I get See, I'm so used to saying EA, now I have to say AE.
So Atlas Agent Engine. That's a big part of the investment. On the go-to-market side, Ryan went through a good bit of this. Adding quota-carrying reps, especially in the enterprises, to drive the data I just showed you. A stronger muscle around specialization for new products. Expansion into specific geos. Asia-Pacific and U.S. Fed are the two we've called out. Investing in the partner ecosystem. Also, and super importantly, folks, we continue to invest in developer awareness and supporting our AI initiatives. That, go back to Reclaim the Bay, is a big piece, and we're going to continue to invest in that as well. A lot of this is in the All of this is in the 100-200 basis points growth that I showed you about 10 slides ago. So it's all baked into that. Okay.
A small thing, but super important for us is, hey, you want to be a business? In my mind, you need to be GAAP profitable. The company has invested for all the right reasons. We guided to be GAAP profitable $0.77 at the high end of our guide. It's a part of it. A lot of this growth has been driven by the revenue growth, but also we've gotten a lot better about managing our stock-based comp, and I know a lot of you care about that as well. 30% five years ago. Year- to- date, 20% as a percent of revenue. We're not committing to a range other than to tell you we know it's important. We will continue to manage dilution, and we expect this to go down.
It may bump around a little bit if we do acquisitions or we want to invest everywhere else. We want the flexibility to drive growth. Everybody knows, especially for engineers, this is a big part of their comp. We want that flexibility, but we are mindful of the dilutive aspect of equity plans. Okay. Now my favorite chart, cash flow. When I started, I looked at MongoDB, and I struggled as to why the cash conversion was only about 50%. You generate all this profit. Why is not it going to cash? There were two big drivers. One is the business model and product mix normalized. As we went from Enterprise Advanced to Atlas, the payment cycles there are just different. You saw that in the deferred revenue. Look at the cash flow. That was a negative impact to operating cash. That is largely modulated.
As that has normalized. The other part is what we can all, and I give all 5,700 employees at MongoDB a lot of credit, we can manage our working capital a lot better. I never get a vendor anymore that says, "I am going to pay you in net 30." No, we are going to pay you a lot longer, and it is just like everybody else. It has been the normalization of the business as well as managing working capital. Folks, we generated almost $500 million of cash last year, and our goal and our desire is to continue to generate meaningful cash flow. It gives us a lot of flexibility around the share buyback, as well as using our cash to grow. So huge slide for us. Okay, before I get to the financial targets, I do want to hit AI and new product innovations.
We have had a lot of debate about this, and we get a lot of questions from you folks. Tell us how big AI is in your business. This is a point in time. We will define for you exactly how we define AI natives and frontier labs. I am almost assured when we stand up here or wherever next year, that definition will change. The market is moving too fast. So it is a point in time. Please do not expect we are going to update it every quarter, but we want to give you a view of how big it is now with the full proviso, which is we make no promises to keep the same definition. Do not yell at me if we move it around. The market is moving too fast. How do we define AI natives and frontier labs?
Those companies that model R&D, whose core product is their own foundation, i.e., frontier labs. That is one. AI infrastructure companies whose core product is the physical or software infrastructure that AI models run on. Number three, AI native software companies whose software application would not exist without generative or deep learning AI. Those are the three. We do our best to audit it. Everybody says they are an AI company. Fantastic. We use outside services as well. This is our best guess as we sit here today. All these metrics key off of that definition. Okay, let us jump into it. On the left you see, and we have talked about this, what is the percent of customers that are using multi-product? All these products are focused almost always on an AI use case, search or vector search.
We have talked about that 34%-48%, basically from a third to a half. These are for customers with more than $100,000 of ARR. These are the biggest customers. You look at Atlas Vector Search, which is a big piece of it, and Pablo talked about this as well. Customers that use Atlas Vector Search in the last two years have grown by 4x . It is a huge growth in the customer base. Atlas Vector Search, as you know, powers AI in agentic use cases such as RAG, semantic search, and agent memory is a big piece of it. So lots of good signs in the customer base. Last year at Investor Day, we talked about what percent of Atlas ARR customers have at least one AI use case. We define this. Are they using vector?
Have they raised their hand and said, "We want to be part of your programs"? Last year it was 30%, this year it is 40%. This is when we talk to you folks about, hey, we are starting to see enterprises deploy AI, at least internally, and these are some of the use cases that we see, and it is across a mix of enterprise and AI-native customers. We get this question a lot. What are the top use cases you see in your customer base deploying AI? I will give you the three. Giving their systems accurate real-time context from their company's own data, making sure the data can support their AI initiatives. Number two is running autonomous AI agents, and number three is AI-powered customer support. Those are the three big use cases that we see today.
That is going to change over time, but that is where we are today, specifically focused around enterprise customers. All right. I am going to give you a second to stare at this chart. On the left, your left, is the number of customer count in the AI-native cohort that I just defined for you. On the right is the revenue for Atlas and the percent growth. As we sit here today, AI natives make up less than 5% of Atlas revenue. I am just going to cut the question off. When you think of that, think of 3%-5% as we sit here today. Okay? Look at that incredible growth, 65, 129, 165. We talk a lot about this, which is it is still small, but it is growing very quickly, and this is why we are comfortable with this continuing to drive growth in the future.
It is off of a small base, but it is super powerful. We showed you the names. You know who some of the customers are. A lot of our product innovations are designed to drive that number up. Total customer count doubling in the last two years. Revenue for Atlas growing at 165% year- over- year, and it is about 3%-5%. I never have to answer It is just a small number. There it is. Okay. We are spending a lot of time and effort extending our lead in the AI universe. Okay. We introduced three new big things today. New features on Enterprise Advanced, Atlas Infinite, and Atlas Agent Engine. We have talked a lot about Enterprise Advanced features, and how we monetize that is incremental Enterprise Advanced monetization through new search and Atlas Vector Search licenses. If you want that functionality, you need to basically sign up for it.
We'll give it to you, but you're going to pay for it. Early revenue contributions should start in FY 2027, but build much more in FY 2028. On Atlas Infinite, it is so well-suited to agentic workloads, and as Ben shared, hey, these are spiky, bursty workloads, and Infinite is a great solution for that. It delivers extreme elasticity, which, again, is perfect for those workloads. It just went into public preview. We expect at some point in 2028 or in early 2028 that it goes GA, but it's going to build throughout 2028. There's a lot of development that still needs to happen, so don't look for meaningful revenue in 2028, but certainly in 2029. With all of these new products, we hope we go faster, but that's the base case today.
Atlas Agent Engine, and Pablo talked a lot about this, we feel really good about the product positioning. To reiterate some of those points, it delivers the capabilities that we think is needed in the market. It provides agent memory so agents can learn. It has the governance and guardrails that keep agents safe and compliant, and it's open and flexible. It's for builders, which is who we market to. This is a new component. The monetization of this is it's going to drive more Atlas consumption. As Pablo talked about it, everybody can use it in the public preview now. The expectation is that GAs either later this fiscal year or early next fiscal year, and we should start to see some momentum in FY 2028. So those are the three new products. That's how we view those today. We have a guide at 2028.
We'll certainly let you know how these play in. I think the number you'll see move the most is EA, and then the other two following as we go through 2028. You've been super patient. Let's jump into it. These are the long-term targets that we showed last time, and let me just run through them. Importantly, folks, we're at a three to five-year timeframe. This was our first long-term model. This is from last year. This is not new. We were making sure to give you a view of where we thought the business was moving, and we picked three to five years. Again, total revenue growth, high teens, Atlas revenue growth, 20% plus, EA and other implied mid-single digit, and then operating margin of 100-200 basis points, not a ceiling, and then 80%-100% free cash flow.
We did not talk about GAAP profitability, but we will today. In the new model, we have shortened the time to three years. Folks, five years is a long, long time. This gives us the capability to be a little bit more pragmatic about what we think the next three years look like, and we don't have to have the hedging of five years is way too long. So that's important because it narrows the timeframe. Let's start at the top. Our new long-term target for total revenue growth is 20% plus. We feel very confident that we have two growth drivers, and we'll talk about each one of them, Atlas and EA. We are very comfortable saying that we expect to grow total revenue north of 20%. For Atlas, we have increased our long-term target to mid 20%. Okay?
This is an important metric, so let's talk about this a little bit. Our old target was 20% plus. Our new target is mid-20s. Keep in mind, we are using the same guidance framework that we've talked about the last three quarters. If we've been clear about one thing, it's the guidance framework that we have for Atlas and for EA, and we've not changed that. We continue to feel very good about Atlas. It is a consumption product. It will always be prudent in our projections, and we've told you that. It's a little bit shorter timeframe, but it is a meaningful jump in terms of how we think Atlas should grow. As we've said, folks, we're going to be prudent in our forecasting and guidance. That being said, we get a lot of questions about how can Atlas grow faster.
What we've listed here are what those targeted sources of upside are. We've talked a lot about this. Better enterprise AI adoption is probably number one. AI native contribution is three to five now. Boy, we would love that to be much, much higher. The new products, if they get going faster and we start to see better traction, that will help. App modernization is the goldmine we're all chasing, and now with our new app product, we feel much better about that. That is an upside. Then all the great partnership motions that Erica talked about as well. So in our base case, these are all here, but they're all If they do better, then hopefully we can grow Atlas faster. Okay? So that's Atlas. Let's shift back and talk about EA and other. We all know the nuances of the multi-year revenue impact to EA.
I've said from the time I started, I'm not going to lean over and bet on a multi-year deal because the minute you do that, you don't get it, and I look like a fool in front of all of you. We are basing our long-term targets on ARR because it mitigates the multi-year impact. What we're committing to is 10% plus growth in EA. That is not a ceiling either. We expect double-digit growth if all the new products pick up and the world continues to move. Hopefully, that's even higher. But at a minimum, we expect to grow ARR over the three-year timeframe at 10% plus. Operating margin is the same at 100 - 200 basis points, and it is not a ceiling, folks. What I want to make sure you realize is we've walked you through the business model.
We want to invest to drive growth in product innovation and in go-to-market. I don't think we have a margin problem. We want to drive better growth. We expect the business model to continue to drive up margins, but we also want to be able to invest in the business. So we're going to leave it at 100- 200. The last year, two years ago, it was 350, 260. It's not a ceiling. The model has great leverage. The same thing for cash flow. We're going to leave it at the same 80%-100%, because again, so much of this is driven by the pricing and the collections from customers. Folks, we don't want to go down that path. We feel good about where we are in our pricing, and we expect the business model and the product mix to be relatively normalized.
And then very importantly, we fully expect to be GAAP profitable in 2027 and in the future. You add all that up. This year on the rule of 40, we have guided to the rule of 44. Here is what we say. If we do what we say we are going to do and maybe a little bit better, gosh darn, we get pretty close to the rule of 50 exiting the three-year. That is our goal. I am not going to commit to it because, again, I want the flexibility for those things to move around a little bit. We feel really good about this. Get a lot of questions, how do you feel about the business? We feel really good about the business, and that is a commitment that we are making over the next three years.
Okay, before we go to Q&A, let me talk about capital allocation, just to close this off. We continue to expect to drive material free cash flow. That then we will use to help manage dilution from the equity plans. It is not going to completely offset it, but we do expect to manage dilution through the cash flow. And then we have over $2 billion of cash on the balance sheet. It is not burning a hole in our pocket. We will look to do targeted acquisitions that are focused around product roadmap or open market adjacencies. And I missed one thing, which is I am very happy to announce that the board of directors has approved another $1 billion in our repurchase plan. Thank you for that. That is all part of the stock buyback, and we are announcing that today.
We will target disciplined acquisitions that are either tech and talent or bring new adjacencies to the business like we did with Voyage AI a couple of years ago. Okay, key takeaways from the day, from my perspective, it is a huge market opportunity. We have strong momentum in a large and growing market, and we still have a very small share. Our core products, we are confident in the durability of growth of both of them. AI is early, new innovations are early, but momentum is building, and the new innovations strengthen our right to win. The business model continues to generate operating margin expansion and fully supports the new long-term targets. And potentially most important, after all the questions in the last two days, folks, we know we have the team to execute. We have the right team and the right place to hit our commitments.
We take very seriously our commitments to 70,000 employees, 80 million plus shares outstanding, and our awesome 5,700 employees. We are proud to stand in front of you today and tell you the plan for all of those folks. We are excited about MongoDB. And at that, it is time for Q&A, and I am going to ask Jess Lubert to moderate. If I could ask the management team to come up, that would be great. Thank you.
We have got about 45 minutes for Q&A here. I will just run a show here while the team comes up. Direct your questions to either Dev or Mike, and then they will chunk them up with respect to the answers. Jason, you want to kick us off here? State your name and firm name.
All right. Jason Ader with William Blair. I wanted to ask about the Atlas Agent Engine, and in particular, what does this compete with today? Is it replacing something? Maybe talk about what people are doing today in terms of that agent narrative.
Yeah. What we see when we talk to a lot of customers, especially in the early days, is they are using a host of different solutions. I mentioned some of them on the slide. You will see hyperscaler SaaS solutions, some of the LLM provider SaaS solutions, and then there are also some that are verticalized around analytics or SaaS. What we have heard from customers is in a lot of these different solutions, the building blocks. If you are using a hyperscaler as example who provides this, they provide you with all the building blocks, but you have to stitch this all together and put a lot of onus on the builders around some of the requirements for security, for being open, which a lot of the hyperscalers and LLM providers do not provide that flexibility.
What we have seen is it really comes down to three things from a competitive landscape where we think we have a differentiated solution. As you heard from Vikas, memory and getting memory to context, which we do natively and configurably for our customers, is number one. The second one is we do not lock you into a model or we do not lock you into a cloud. We really do provide that open framework. Because we had a chance to listen to customers who were playing around with different solutions early on as we were going through design, we took security as the first principle and actually built it into the system. You do not have to actually build security on top or bolt it on top.
It is actually part of the system, and that is really important for customers, especially a customer like Citizens Bank that is in a regulated industry.
Hey, thanks so much. Tyler Radke from Citi. Great presentation and appreciate, Dev, you being here in person. Maybe I will direct this at you, and it probably applies to the broader group, but one of the things that we have heard you guys talk about as it relates to AI and why it is maybe a little bit earlier in the journey for MongoDB versus some of your peers is many of the AI applications built were more employee facing, back office facing. I wanted to ask you now that we have seen a lot of momentum, Meta Muse, some of these consumer facing and customer facing agents really rise into production, how do you see MongoDB's role within those use cases? I know Meta, for instance, uses PostgreSQL, but is that an opportunity to potentially win that in some type of partnership? Just how do you think about that opportunity broadly?
Thank you.
Thanks, Tyler, and I will start and I will ask the team to jump in if I miss anything. I think I said this right at the beginning. I think, one, when you use an agent to reason about a question or a decision, they need to have access to fresh data. So the freshest data in your system is live operational data, information about your business, what your transactions you are executing, what orders you are taking, what sensor information you are getting about your business. And so one is you need to have fresh data. MongoDB is an OLTP data store. Two, just because you have fresh data does not mean you can find the right information because you can have a lot of data. So being able to precisely retrieve that information based on what is being asked is super important.
The fact that we have a very sophisticated retrieval mechanism with Voyage and using embeddings and re-rankings to basically think of it as like if you told someone, "I want you to find information about something, and go read the library." If they read every book in the library, that would be a very expensive process. But if you tell them, "Go into this section, it is about chemistry, it is about organic chemistry, and I want you to find this particular topic," all of a sudden, very quickly you find the relevant information. So finding information quickly and precisely is super important. We are world-class at that today. Three is scale. As we talked about and as Ben talked about, the unpredictability of these workloads is very high. Because now you do not know when the agents work, obviously for users, after everyone goes home, you see traffic come down.
You have a good pattern of traffic. But agents can be working all night in an unpredictable manner. So you need a platform that can elastically scale. With 9.0 and Infinite, we think we are the most performant scalable OLTP in the marketplace. Then last but not least, you also need to know state because you need to understand what happened and when, and keep understanding the history and what you are supposed to do next. If you do not understand state, things could go awry, and obviously with the memory capabilities we are rolling out. So we have all the ingredients to really position ourselves well to go after this market. Again, all these things have come out recently, but we think there is going to be a ton of other exciting, potentially new AI workloads that we can really go after.
I will just say two things. One, with some of the frontier labs. One we have seen as they get into production scale, needing both the operational real-time data and also being able to drive it at scale, moving from solutions like PostgreSQL onto us, both across pre-training and post-training. So I think there is a journey for them. The second thing I will say, which we hear from customers, is a lot of customers look at these pieces as composable blocks. So they may use a cloud agent from one of the frontier labs or from a hyperscaler, but we are providing all of Atlas as composable components. So if you want to own your data, if it is core to your knowledge, you can actually put it into one of those systems and we will just work natively within that cloud. So it gives customers some flexibility.
Hey, Rishi Jaluria, RBC. Thanks so much for this. This is super helpful with all the information. Mike, in your remarks when you were pointing at the TAM slide, you hinted about maybe wanting to invest potentially more into OLAP as we think about the TAM. I know in the past you had Atlas Data Lake, I think that is now called Atlas Data Federation, or been folded into there. Maybe can you talk about what could that look like? Is that a demand that is starting to come from customers? The flip side of that question that is always going to get asked is we have seen some of the OLAP vendors try to get into OLTP via M&A. Not too much has happened out of that, but maybe just how should we be thinking about both of those forces? Thanks.
Yeah. Thanks, Rishi. I will start and I will ask my friend Ben to jump in as well. We are focused on OLTP. That is in building out that platform. That is the majority of the R&D. That is what we are focused on. When we look at M&A for tech and talent, it is largely roadmap accelerators in there.
That being said, there is a big market there. There is now starting to be a little bit of a gray area in that. We will look at it, but we also do not want to get diffused outside of our core focus. We think the market is so big in our core market, and the growth is there. So we will look at it. We will be disciplined. But that would be a big step for us to step outside that. Again, we have got great sellers. We want to keep them focused, and that is the big piece. But you want to talk about our friends coming into our group?
Yeah. For everyone who has listened to me talk the last 12 months, you know I describe this as cute, that OLAP players think they can be credible in the OLTP space, and it took us a lot of years to do that. I think if anything, then we have been saying from the very beginning of the AI journey that OLTP was the higher ground for AI, then getting into OLTP is just asserting what we knew from the very beginning.
Hey. Ryan MacWilliams, Wells Fargo. Good to see another Ryan MacWilliams. Really great to see the growth, since from Mike in the AI customer cohort, in Atlas, really strong growth in that segment. What are you seeing in both your AI Atlas customers and your core Atlas customer segment? Those two segments that help give you confidence in the new multi-year Atlas market.
Yeah. As we've said it for the last several quarters, we feel really good about the Atlas business. It's growing at 29% consistently, and a lot of what we've seen is not only the incremental adds. It really goes back, Ryan, to the customer piece that we talked about in terms of the growth within our existing customer base, the expansion there. We get a lot of questions about workloads, and yes, workloads will plateau, but Erica's team and Ryan's team have done such a wonderful job of cross-selling and upselling. We're adding to the platform less still of half of our greater than $100,000 ARR customers who use a multi-product. We've talked about this. There's a lot of product adoption. There's not a ton of revenue generated yet because it's going to be lower than that. That is an upside as well.
All the new products we talked about, even though we have marketplace in Japan, we should be a lot bigger. U.S. Fed has a huge growth opportunity. You look across that whole area and Voyage, and that's why we have confidence in the multi-year outlook. The other piece is, and I just want to address this. We didn't have a CRO for six months, and now we do. CJ, bless his heart, really had to jump into that void. Now we've got him. The great part about that is Dev has the same relationships, awesome, but now we have a guy to execute on it. That's the other thing. We feel much better about where go-to-market is sitting here today than we did last year because of the coming transition.
Siti Panigrahi from Mizuho. Congratulations on that Atlas Infinite launch. It was great to see the architecture, how you decoupled memory and storage. The question is, how are you going to position? There are certain application can be both agentic and user workloads. How are you going to differentiate that positioning? As customers do some kind of spend optimization, do you expect some of the Atlas Core customer switching to the Infinite?
Yeah, I think from on the positioning side, first of all, remember what we announced today is preview, right? Just on AWS. We have a ways to go. We're going to add the other clouds and everything. It's still early. As I said, Infinite is definitely designed to handle the bursty and unpredictable nature of agents and scale. That doesn't mean that it's only for that. We see different levels of scale all the time, depending on customers' workloads. Sometimes it's a steadier peak. It all kind of depends. My expectation is that, one, because it's preview, and two, since Atlas Dedicated clusters has been out now in the market for 10 plus years, I don't see a mass movement of existing customers moving to Atlas Infinite. They're happy where they are. It's a predictable pricing model. There's no complaint, right?
As they start working with data in agentic workloads, my guess is those workloads will start out earlier on Infinite. Because we're giving the customers flexibility of going back and forth however they want to. Remember, guys, portability. Number one thing I've been saying, go on-prem, go into self-managed, go into Atlas. There's going to be movements, and that's what the flexibility you're providing. I don't think this is going to be like, "Oh, shit, I can optimize. I'm just going to go fast-forward there." I think it's going to be slow, and I think it's going to be allow customers to prototype more and allow them to move data around.
Okay.
Ivan Feinseth
Ivan Feinseth , Tigress Financial Partners. What do you see as driving your biggest competitive wins? Is it new application development? Is it data migration, data consolidation? What do you think are your advantages that are driving that? What do you think is on the horizon that's the next big thing that also you're best positioned to benefit from?
Look, we've done a really good job of getting into one line of business with a customer, and now we have a seat at the table at a more executive level, and they're looking at it across now multiple lines of business. I would say that's a huge strength for us to go upsell and cross-sell. As you heard from Mike, too, and the team, adding new platform capabilities.
They are looking at our roadmap and saying, "Wow, okay, this is where I want to be." I also mentioned I have been doing this long enough to where there are so many customers that have so much technical debt that they have had backlogs that they want to modernize and migrate, and they are taking those projects on because now the AI tooling and capability is there for them to do so, and being able to migrate in literally days and weeks instead of years. It is all of those things together, I will tell you.
Patrick Colville here from Scotiabank. Lots of juicy, fantastic metrics. To me, the most interesting was the 3%-5% from AI natives. I guess to me that metric speaks to the success but also the opportunity because I compare it to another infrastructure company headquartered in the city, and they are at 3x that level, right? I guess maybe for you, Mike, and also for the broader team, what would it take for that metric to get to the double- digits, the mid-teens point contribution from AI natives?
Yeah, thanks for the question, and I will let you ask them all that piece of it. For us, look, we have said it, Patrick, the whole time, which is it is still a relatively small piece. It is workload by workload in there. If you look at AI natives and Frontier Labs, they are going to be different. We love the AI natives. They are all still pretty small. You heard of one today, Emergent, which is doing a nice job, and they are growing. So for us to get that in double digits, it is really two things. It is more workloads in the larger labs, clearly, and then that AI native cohort starting to scale and starting to grow as well. Because those are the workloads that we should scale with them as they grow their business and continued customer acquisition. We have talked about this.
This is where the new customer metric comes in. You saw that double the number of customers that are AI native. That is going to expand as well. So it is still small, growing very quickly. For us, that is a relatively good thing because we also do not have dependency on any one, and we do expect that to continue to drive growth.
Great. Thanks. Kirk Materne, Evercore. I think this is probably for Ryan, but you all are obviously pursuing a dual-pronged strategy, obviously getting bigger with your biggest customers, at the same time, going after builders. How do you think about just monetization and pricing along those two different tracks? Obviously, your bigger customers have to think about security, statefulness. Smaller customers just might want to go. I was just curious how you think about making sure that you are thinking about monetization the right way across those two vectors and anybody else that wants to add to that?
That is a great question. Look, it comes up all the time, and it is something that Erica and I continually look at, as well as with the engineering team. First, it starts by listening, right? Listening to our customers, whether it be one that comes through self-serve and is somewhat small and from our largest customers. The flexibility that we have, though, with both Atlas and EA today is pretty helpful as we see more customers focusing on a hybrid approach, I will tell you. Then it is really just listening. So we are going through that, and we will continue to evaluate that, and we will make sure that we adjust accordingly, right? With keeping an eye on margins.
Yeah. I think, Ben, maybe you can just talk a little bit about how our pricing approach is much simpler than some of the other infrastructure providers.
Yeah, for sure. So a couple things is that a lot of the other infrastructure providers, it does not matter which one, they bill customers, especially in cloud, by read units and write units. We do not do that, right? Customer comes in, and I will just talk about Atlas Dedicated because it is the GA product. You pick your capacity, you pick your storage. If you do not exceed that capacity, your bill is the same every month, and it is predictable. Customers love predictability. Enterprise customers love predictability. But what this also means is as enterprises start to prototype different types of workloads or they add a little bit more data, unless they exceed those bounds of the box, so to speak, we are not going to charge them more. So on the weekends, they want to try an agentic workload and throws 100,000 queries at it. They did not exceed their allocate. Great.
I think it is really important to understand that one of the reasons why enterprises love us so much, besides the security and the governance and compliance and all these things, is that we provide them a very predictable surefire way that they can actually understand their spend, understand their usage, and they are not worried about the runaway bills that we have heard from the past.
Hey. Thank you, Raimo Lenschow from Barclays. The big debate in the market is people start working on AI, and the database you start with is the starting point. Then if someone comes in later and says they have the better database, that becomes a database migration, which is kind of a headache that we all know. So in that respect, where are we on that winning back the base? What are the proof points that you see? Pablo, that is probably you.
Yeah.
Sorry. But where are we on that journey? Thank you.
Sure. Raimo, it is good to see you again. Look, as I said, one of the key things for us that is super important is making sure that we are where the builders are. I think as you know this, we have always been a builder company. So I went through what we are doing to show up from an awareness and from also an accuracy perspective. We are seeing good early wins in terms of driving both of those. That is sort of the start of the funnel. We also need to be in all the right places for customers as they are building those applications. Frankly, we are just getting into a lot of those marketplaces.
It is creating a lot of that opportunity so the coding agents can actually pick the best platform for the job, where we are as it relates to JSON and the document model and also our integrated vector search and capabilities like Voyage AI. That is a starting point. The other thing I will also say is when we have seen this with some of the at scale customers, we have had Frontier Marketplace that migrated to us in 4 weeks off of a PostgreSQL database for conversation history. The other thing also to think about is, as Dev Ittycheria said earlier, it is a new workload, and there are a lot of those that are coming up and spinning up. I think we are participating in that market, and it is in the enterprise and in production still pretty early.
The other part of this is we are also working very closely with a lot of the builder communities, like the Emergent Labs, to make sure that we drive that awareness early on.
Raimo, you have heard us talk, and they were at our MongoDB.local in London, ElevenLabs, where they started on Firestore, and they just got to a point of scale as they saw their market opportunity increase, and they said, "We need to get off," and got on MongoDB. There is a whole host of other those that as a startup, you are just trying to make your next paycheck, trying to find your product market fit. Then when they hit a scale point, they tip over, really. We are there to help them.
Let me just add one other thing. The way I like to think about this is the developers of today are different from the developers of three years ago when the tech stack was determined for them, and then they started writing the code. Now, Grok Build, Claude Code, whatever, makes it so they do not care about the tech stack until it is maybe too late. If you think about the selection that happens in that tech stack decision, the model makes, or the agent or whatever, pick your poison, makes two decisions, one of them really cool and one of them shitty. One of them that is cool is it is always modeled on JSON. That makes our migration easier later on. The mistake that the models make is that it picks PostgreSQL on JSONB most of the time, which to Pablo's point, that is what we are trying to change.
When they scale and reach some sort of exit velocity, just like Emergent did, easy migration. We will get them up here. I am happy with the enterprise. I am happy at the mid-tier. We will migrate them every single time, no problem. Pablo and the team, we are focused on the long tail of the funnel as well.
Let us go to Matt.
Hey, Matt Martino, Goldman Sachs. Dev, great to see you again. This question is for you. I wanted to talk about the modernization opportunity at AMP specifically because it started under your tenure. I guess what I am curious about is because it seems like we went through a bit of a quiet period on AMP over the last nine to 12 months. I am curious what has changed from a technical and commercial standpoint where you can actually see some follow-through in the numbers around the modernization opportunity broadly.
Yeah. Just to level set with everyone, what I started talking about maybe three years ago was the fact that when we had lots of customers who felt trapped on legacy platforms and it was easy to move the data, it was very hard to rewrite the application code. That was the big tax, and because these were a lot of old brittle systems, a lot of stored procedures, it was a very hairy, complex job. With AI, we thought that that problem would get solved. What we underestimated, can be, was that the maturity of the AI models took longer than we had expected. Obviously, we have seen rapid innovation of model capability and performance. Mike, being the great CFO he is, he is good at managing expectations, says, "Investors are a little tired about hearing AMP.
Next time we talk about AMP, let's have something really hard to deliver. I would tell you that we're really excited about what we're seeing. We made an announcement with Cognition this morning, who's now our partner, who's helping us essentially work with us to help migrate, and they've become very successful as a coding tool to help migrate customers and deal with that application logic problem. And obviously, we provide the tooling to migrate the data as well as put in all the guardrails to make that a production system. I think we didn't want to overplay our hand today, but I think that's a really important growth vector. As Mike talked about, it's one of the upside cases in our model.
Yeah.
Yeah. Sanjit Singh, Morgan Stanley. I wanted to start with you, Dev, on, when we go back to the IPO, and you were for years consistently saying, "We're about the developer, and we're focused on the developer." Obviously, the developer role is changing. From writing the code to orchestrating a harness. Do you see risk in that changing roles of the developer and how MongoDB's value proposition ultimately changes, and what are you doing to mitigate that risk? And then a follow-up question for Ben on, what was the question here? On memory. Is that like a net? Have you sort of productized that capability? Because I kind of thought that was part of the core value proposition of having a document model. So with the new agent enterprise offering, have you sort of productized a memory capability? Just want to get some clarity on that.
Yeah. Sanjit, to your first question, there's no question in my mind that over the next, say, X number of years, whether that's three, five, or 10, humans will no longer write code. You see that coming, and it's become very clear. I think there may be human in the loop. I think there'll be checks and verifications and all that, but most code will be written by agents. Yes, the original value prop of MongoDB, one of them, was that it makes the developer's life so much easier because it's so consistent with the way they code and they think, especially we get all the drivers, all the popular languages. I would say what also was one of our value propositions was the ability to essentially scale up and scale down. A lot of in the AI workloads we talked about is the bursty traffic.
A lot of that traffic is ephemeral. We've always been a very scalable platform, so that plays to our strength. Ben mentioned JSON, the fact that we speak the language of AI. You can even prompt in JSON. I don't know if people have actually done that, but you can prompt an LLM with JSON, and it will respond because JSONs are, I mean, LLMs are designed and built around JSON. We think those attributes are just as important, and that's why we're building all this capability around price, performance, Infinite, and Agent Engine to provide all the wrapping to enable agents. We recognize that we may have been a little slow, like with other companies like Supabase, who have done some work around making Big much more agent-friendly. Those are easy gaps for us to fill, and you'll see us address that pretty quickly.
Just to then answer that. There's memory of today, which I'll talk to you about, and then we'll talk about memory tomorrow or now today, because we announced it today. Before today, when we talked about memory, you're 100% right, it was just another database workload for us. Agents were speaking JSON, great, stored in MongoDB, and we saw that organically start happening at our customers, as well as other context like chat conversations and other things. But the memory of today, what we announced today, is a little different, and it gets into more of the business logic in this.
Yeah. What we announced today, memory actually is a composable component. As Ben said, we've got customers who've been using us for memory because they can, and they've been building on top of it. But what we actually do, and it's both a memory and a compute problem, we actually bring in long-term memory, so all the stuff that comes in from previous conversations, stuff that gets inferred and brought back from the LLM. All that stuff comes in, and we actually break it down from a long-term perspective into four different types of memory. We do that under the covers for the customer. We break it down into procedural memory, so these are the steps that the agent will learn and memorize. Episodic memory, so the episodes and events that it's remembering.
Taxonomic memory, which is the taxonomy that may be for that agent, for that team, or for that organization, the terms that are very specific to that team, like Magenta, the code word that we had for H10 before we launched it. And then semantic memory, which is sort of the knowledge of the organization. That all happens in the system, and we aggregate it, and then because we have Atlas Vector Search Voyage AI under the covers, we're able to bring back the right context for the agent at the right time, which both drives accuracy, meaning fewer turns, and lower cost as we're trying to drive some of the token usage. That's all available to customers. We provide it as an SDK, and we're externalizing also through MCP.
Alex Zukin.
Thanks, guys. Alex Zukin with Wolfe Research. A lot of great numbers, especially like the AI native ARR. I think we are all a little surprised about the magnitude of the higher Atlas or the confidence in that mid-20s over the next three years. Maybe just dive in a little bit between those two, AI natives versus core, what is driving that confidence? Do you expect to re-accelerate the non-AI native part of the business, given, to your point, now you have a CRO, you did not have one in the first six months. Maybe just help us understand that, and I have got a quick technology question.
Sure. As we look at Atlas, and we have said it multiple times in the last several quarters, we continue to feel very good about the Atlas business. Technically, going from three to five to three was a big thing, Alex. It allowed us to say, "Hey, what do we really see calling in the near- term?" We have seen all the growth that has been driven by the move-up market, which has been significant, that as we look at the geo expansion, the new products, as well as the platform. That is going to, over time, hopefully start to monetize. Again, good usage, but it needs to monetize more as we look at that. Then, of course, you have the AI natives.
Underneath the covers, when you look at the AI natives are 3%- 5%, it means the core, by definition, as a percentage, continues to call it slow down, but the dollars continue to grow. We look at that over the next three years, and we feel very good about growth in both of those, and we really have not baked in a lot of upside for enterprise AI. It is the thing we talked to you folks the most about in the question earlier on, when is that going to pick up? At some point, now keep in mind, AMP is great. Dev did a great job of going through that. We also want sales to be able to sell it. That is going to help both self-manage and Atlas, but we do expect growth there.
The confidence is in the go-to-market engine, the growth we see in our larger customers, the new product, and continued growth in AI natives.
If I could just add one other thing that I think, from working, it also gives me confidence that I had the benefit of stepping back and now stepping back in, is that when we launched 8.0, there was some concern both in the investor community, is that going to cannibalize some of your business, right? Because if you are providing a 30%, 40%, 50% faster machine, this is going to need to use less gas. Well, it turned out that they wanted to drive that car more because they put more workloads on it because they saw so much benefit. And the fact that 80% of our fleet is already at 8.0, faster, 2x faster than any other prior release, gives us confidence that same phenomenon will happen with 9.0 because 9.0 is that much more performant, and that's a pretty material data point.
Whether Atlas Infinite, where you feel like is there cannibalization risk concern? Is that completely net going to grow?
No, let me address that. Ever since I've been at MongoDB has been viewed as a great technology. No one has ever questioned the caliber of the technology we've built. Some people have questioned, some people use the term Ferrari or turbo sports car. It's great for certain workloads for the mainstream meat and potatoes. Sometimes those internal workloads that don't need all the sizzle, maybe MongoDB is not the right profile for. By increasing the price performance, we make it just that much easier for people to put more and more workloads on MongoDB. So it helps our expansion to larger accounts because there's more workloads we can track. It helps our upsells and cross-sells in our existing customers. And obviously, we can go after these new AI workloads as well.
Super dry with Oppenheimer. Building on the enterprise AI opportunity, which you've talked about unlocking there, can you give us some perspective on what you think the system integrators can do for you when you start expanding there?
Yeah. For me, the systems integrators are about credentializing us, giving us access to buyers that we would not have otherwise faced off against, and helping us drive true transformation at the business level. That is what the SIs are going to be able to help us do. What is important for us, I know you heard it from Vikas, and this is where it is actually coming true, so it is an example, is a lot of our enterprise customers are reliant on the system integrators to help them make the decision around what technology and what innovations they want to use for their AI transformations. Because there is so much innovation out there, it is complicated to make those decisions. Getting with the SIs and getting embedded into their frameworks, they are giving us a natural entry point into those buyers, into those enterprises.
We are credentialized from the start because they are walking in and saying, "As part of our stack, we are embedding MongoDB." So for Infosys, we are part of Topaz's framework now. When anyone walks into their innovation lab around the world, on the wall, it is going to say MongoDB, and when they show Topaz, it is going to say we are the data platform for it. So that gives us access that we might not already have, and that is going to be true expansion for us in terms of new customers, new stakeholders, and new revenue.
Yeah. I also want to add, just one of the things that surprised me, it kind of has not surprised me, but in some ways surprised me, is that the gap between where the AI capabilities are and where the enterprises are is wide. There is a massive, what people call the diffusion gap, and if you go back in history and look at when PCs were deployed and networking was deployed, it took time for people to reengineer their work around this new technology, and the productivity gains did not really happen until three, four, five, six years post the deployment of the technology. I definitely see, having been at Sequoia and I was working with them over the summer, clearly the AI natives are all AI pilled, and they are being very aggressive. But when you talk to the enterprises, they are still moving very slow.
The example I gave you of some very sophisticated customers, they are just pleased to have an AI chatbot. So it is going to take some time for them to really deploy and leverage the sophistication. In the last month, we have seen more AI models deployed with performance gains that are unbelievable, and each one of them would have been like a noteworthy front page newspaper announcement 12 months ago. Now you just see them coming out almost every week, and these enterprises just cannot keep up. So the SIs play an important role in bridging that gap between where the capabilities are and where the customers are, and I think that is where they are going to add a lot of value.
Steve in the back.
Great. Thanks. It is Steve Koenig from Macquarie. I want to build on the good question that you answered earlier, Dev, about the nature of the developer changing in the future. I want to bring together a couple of points that I have heard here. One is that the agents are currently going with PostgreSQL with their JSON extensions instead of being aware of MongoDB. I want to also bring in your point about you are working harder to move up the decision-making ladder in the enterprise. Lastly, I just want to bring in the fact that, Mike, when you were citing three of the top use cases for agentic, the coding use case, which is huge in the enterprise and in AI natives, was not one of those three use cases.
It smells like there is an opportunity here to enable the coding agents in creating applications and enabling them to decide to use MongoDB and putting that in at an enterprise level. I do not know if I am stretching this too far, but how will MongoDB. Is there an opportunity here? If there is, how would you look to enable this?
Yeah. Pablo, you want to take that?
Yeah, I will start. Look, I think, as you know, in the enterprise, you have your enterprise-approved solutions, and we fare very well in that space. One of the things that we had not been doing well is making sure that we show up in the coding agents, making it super easy for developers or for the executives to choose a framework. As you know, enterprise, a lot of those decisions are getting made by enterprise architects in terms of for their production at scale systems, and we have made that a lot easier for them. We made it a lot easier for those systems. And one thing I will highlight in our partnership with Cognition, where they are getting a lot of reach into some of these enterprises, we are actually seeing that happening and those choices, both the coding agent and the platform, is being made together.
When you think about it, in an enterprise, they are not going to let developers build production scale applications using any single solution. Those typically go through the architecture team. And that I think, one, we show up well there, but really, a lot of the stuff that we have been doing for both builder awareness, accuracy, and being in those platforms is really helping us make it super easy for those teams to build on us.
All right.
Great. Thanks. Yanni Samoilis with Baird. Just going back to the enterprise AI transformations for a moment. Those can be super lucrative, but also they require a lot of handholding. The SIs can help with that, of course, but I was wondering if you think you need to send more of your technical folks in as well, embed them in those customers' operations to be able to help with some of that handholding, or if the SI partnerships are sufficient enough?
It's going to be both is the answer. We are building FDE capability right now, starting in our product teams. It'll migrate over to professional services. We'll have that capability, and we'll be able to deploy that alongside our partners and our customers to help them accelerate their revenue. We absolutely believe that's necessary. We know our customers, they have to set up their schemas, their indexes, everything correctly in order to make sure they get maximum value of the platform and to make sure that then they want to consume more. That's going to be critical. It's a motion we're driving and frankly, it's essential for delivering the kind of customer experience we want and getting the level of consumption we need.
The thing I'll add to that, and this is true for both AMP and for Atlas Agent Engine, where we do have FDE capabilities. It's not just about helping the customers. It also creates a super tight feedback loop for us. As we've been in design partnerships and in our early previews with a lot of these customers, those FDEs are embedded. They're there, they're either physically there or in with those teams, and we're getting that feedback back super quickly, which is helping us reduce the cycle time for new innovation.
Hi, this is Palak. Hello. Yeah. Hi, this is Palak for Miller Jump from Truist. Thank you for taking questions. I just have one quick one on Enterprise Advanced. Enterprise Advanced has been doing really well in the past few quarters and is benefiting a lot from Atlas Vector Search. Given the double-digit growth opportunity in the next three years, could you just talk about the opportunity and where do you expect Enterprise Advanced to continue benefiting from?
Yeah. As I articulated about the Run Anywhere strategy and vision, that is 100% something we're committed to. We're going to continue doing that, which basically means as we do more and more inside of cloud and other things inside of Atlas, where it makes sense, we'll continue making more parity decisions to bring those into the self-managed, I would say, ecosystem. But just from a market perspective, there's a lot of things going on. We talked about AI adopted in enterprises. Some of it is people, some of it is laws that haven't even been passed yet. They don't know what they're doing in France. They don't know what they're doing in England. How is GDPR going to apply to this? Does FDA have to approve this model for this particular use case?
Do we have to give an agent a Series 7 so they can look at financial data and give advice to your customers? We don't know what this actually looks like. When they figure out all that's when we're going to see enterprise AI go really crazy. Until then, we're going to see what we're seeing as well as where EA plays into a part is, one, regulated industries where they can run it inside of their fiefdom and then make it for more internal use cases. Two, public sector and air-gapped environments, because that's really important. They're not going to slow down. Then three, there is plenty of geolocations in the world that doesn't have a cloud provider that either needs to be served by a Neocloud or an Equinix or pick your poison, or just doesn't even have anything, so it has to go there.
The banks or the insurance companies still want to serve that area, so EA's going to play a part there. EA really has a massive play in this thing, and it's not an or, it's 100% an and.
One thing I'll add to that. EA is a deployment option for us. Search and Atlas Vector Search was introduced recently in the last couple of months. So when I talked about the Atlas Vector Search growth over the last couple of years, it really was around our cloud. The thing there, as Mike touched on, those multi-feature use cases really do resonate with customers. So one thing I think about is we've seen the match and we've seen the market fit as we've deployed this in Atlas. So as we go into EA, we're earlier in the journey, but we're very optimistic.
We're up on time. Thank you everyone for the great questions.
Thank you. Thanks everybody.