MongoDB, Inc. (MDB) Earnings Call Transcript & Summary

All right. Welcome to another great session on day 4 of the Morgan Stanley TMT Conference. I'm Sanjit Singh. I cover the infrastructure coverage within the software team at Morgan Stanley. We're super excited to have the management team here. From at MongoDB, we have CEO, Dev Ittycheria. I think Serge is going to be joining us on stage in about a minute. He's getting miked up. But given that we're a few minutes late, we're going to get started. Before that, let me just get through the disclosures. For important disclosures, please see the Morgan Stanley research disclosure website at www.morganstanley.com/researchdisclosures. If you have any questions, please reach out to your Morgan Stanley sales representative.

So maybe just to level set, Dave, when we think about this year, another solid year, overall revenue growth 19%. Your Cloud business on a reported basis was 27% growth. Reported earnings results last night, again, solid quarter, 20% overall growth. Cloud business growing 24%. But the outlook did disappoint the market expectations for fiscal year '26. Before we dive into the details and the debates around Q4, I guess the high-level question that I have for you is from a big picture perspective, do you believe that MongoDB can get back to that 20% -- 20% revenue growth profile that you've historically achieved since you guys have been a public company?

The short answer is yes. And I think one of the key points, we want to emphasize from the call yesterday, and we tried to do that in some of the one-on-one meetings earlier today is that Atlas growth has stabilized. Consumption growth has stabilized, and that's a question a lot of investors were asking us. And if you do the math based on the feedback we gave on the non-Atlas business. I think you can reverse engineer into a pretty solid growth rate for Atlas. And that's a function of a couple of things. One, the base itself has stabilized, two, the workloads that we acquired last year seem to be bigger and growing slightly faster than the workloads we required in the year before period. And that's also giving us confidence about the workloads we will acquire this year as we move upmarket and address more and more sophisticated use cases.

Yes. It's a great way to frame it out. So let's talk a little bit about, Serge, maybe you can walk us through the guidance philosophy, the assumptions that underpin that guidance, the puts and takes and why -- why that sort of outputting and netting out to sort of a 12% to 14% guide?

Yes, yes. So the total guidance is 12% to 14%. What we called out is that the non-Atlas revenue will decline in the high single-digit range. And that is entirely due to the fact that we're facing a very difficult compare when it comes to multiyear non-Atlas deal. So due to ASC 606 when we sign a multiyear deal, we recognize the term license component upfront. And we had 2 very strong years, exceptionally strong years in a row of multiyear -- non-Atlas deals fiscal year '24 and then fiscal year '25. So what that means as we look into fiscal year '26, the opportunity set of deals in our renewal base to do multiyear deals is simply lower than it was in the prior 2 years. And even though we assume that we will be equally as successful, just a smaller opportunity set yields $50 million less in multiyear revenue. And that's the reason why the non-Atlas revenue is declining in the high single digits. They've already talked about Atlas. What's implied in the guide is stable consumption growth in Atlas fiscal year '26 versus fiscal year '25. We find that very encouraging in the context of a growing business, and that's no small feat. And again, we feel like we can accomplish it because of the strength we're seeing in fiscal year '25 workloads as well as our increased investment, up market and our strategic accounts where we see better productivity. And so those are the puts and takes when it comes to the guide. I think that perhaps a better way to think about the underlying growth of the business is to normalize for the $50 million. And so the way that I would do it is I would remove it out of the fiscal year '25 revenue base. So that's how that 12% to 14% becomes roughly 300 basis points better. And that's more reflective of sort of the underlying growth of the business, minus sort of the accounting lumpiness.

One of the questions I'm getting this morning from investors is if you just look at big picture and you look at the trend line of growth. You guys, I think, a couple of years ago were high 40s year after low 30s, this year 19% now guiding if we sort of normalize mid-15s. So the question I'm getting is, is there a competitive issue here? Is there any issue with retention in terms of that line of concern, Dave or Serge, how would you address that?

Yes. So our win rates are still very high. What we see, there's really 3 layers to the cake in terms of our growth. One is the base business, workloads that required post 2 years ago. And obviously, that's the largest part of our Atlas business. There's the workloads required in the prior year and then there's new workloads. We did have issues last year this time where the base was not growing as fast as we thought. And then the workloads we had acquired in the previous year were not tracking as fast as we thought. And then we had a late start to the year with some organizational -- just changes at the start of the fiscal year. We fixed all that. So we feel like that's the big reason why we're calling out stable consumption growth in Atlas. I would say that we do think this year is a year of transition. We are really excited about the opportunity in AI, but we also do recognize that customers especially large enterprises are moving quite slowly in the deployment of custom AI apps. Most of the AI use cases are fairly simplistic, chatbots, document management, use cases, et cetera. But we are seeing people get very interested. And we think architecturally, we are well designed for the world of AI. One, we support different types of structured, semi structured, unstructured data. Two, we're highly elastic and scalable. Three, we natively embed text or keyword search and semantic search through our Vector Search capability. And then we just announced last week the acquisition of Voyage AI, which is the best-in-class embedding and reranking models and that's all designed to really essentially reduce the risk of hallucinations. And one of the things that we find with our customers is there tends to be 2 issues holding them back. One is the skills issue -- skills gap issue and the second is a trust issue. And so the way we're addressing the trust issue is to obviously do things like Voyage where they can get better predictability and quality of the outputs from these AI applications. And on the skills issue, we're not just coming to technology, where we're coming with a very solution-oriented approach. One way of that is modernizing existing applications using AI and the second one is really helping them build custom AI applications that really transform their business.

Yes, a lot there, and I definitely want to dive deeper into a lot of those points that you just made. Kind of tying the bow on last night's earnings results. Serge, going into this fiscal year '25 that we just completed, you guys had historically been kind of prudently conservative on the non-Atlas business and the multiyear deals. What is it about fiscal year '26 that's going to be different, resulting in that sort of high single-digit headwind? Because there was, I think, a $40 million anticipated headwind for this year that didn't seem to materialize. Why won't that resolve itself to the upside in fiscal year '26?

Yes. So if we rewind sort of the story of multiyear deals, just to make sure everybody is on the same page, we had an exceptionally strong fiscal year '24. The biggest deal was Alibaba, but there was strength across the board, well above sort of the average. So going into fiscal year '25, we assumed the $40 million headwind, which would result in fiscal year '25 being roughly an average year when it comes to multiyear performance. In Q3, we call a significant outperformance. In Q4, part of our outperformance was also due to multiyears. So most of that $40 million headwind actually did not materialize, right? So we outperformed our expectations that we had in fiscal year '25. And you have to understand that like forecasting multiyear deals is inherently particularly difficult. Sometimes those happened at the very last minute. And so it's not conservatism. It's just difficulty of forecasting is how I would put it. And so -- but what that means now is we look into fiscal year '26 and we think about sort of what's available for us to go get using historical sort of occurrence, if you will, of multiyear deals, whether they are renewing multiyears or new multiyears, when we apply to a lower opportunity set in the renewal base, that's what results in the $50 million. So in order to do better than that, we would need to see better -- a greater percentage of customers signing up for those that has been the case for the last 2 years. So that's not what we're assuming. It's just that the opportunity is lower.

Got it. And then the last one on -- coming off of last night's results, on margins. So op margins in terms of the guidance, you look at the midpoint, sort of targeting about 10% after delivering about 15% operating margins in fiscal year '25. Understanding the $50 million headwind from the non-Atlas business, which is a super high margin. Why not philosophically take the decision to protect margins after a weaker-than-expected outlook?

Yes. So first of all, just to echo what you said, that $50 million is roughly half of the margin decline from 15% to 10%. And then the second is related -- and I'm sure Dave is going to want to chime in here, about our confidence in terms of the investments that we're making and the opportunity going forward. So where we're disproportionately investing in the business is 2 areas. One is R&D, because we see an opportunity to continue pushing the envelope in terms of performance in the core database plus the investment in Voyage AI and creating this out-of-the-box GenAI solution that we think would be unique in the marketplace. And the second area of investment is awareness. We hear over and over again, even from some of our largest customers that a small percentage of their developers really knows the full capabilities of our platform. So that's an obvious opportunity to increase our ability to acquire new workloads. And that's an area where we're going to invest this year and invest more consistently going forward. So that's the rest of the bridge down to the -- from 15% to 10% in addition to the $50 million. The only thing that I would add is -- so then if you kind of turn around to like what's the philosophy underlying? Those are the puts and takes of what's the philosophy. So we observe our margin performance over the years, and we have high confidence that the business scales. So we don't have to go back to the IPO, which you remember when the margin was negative 37% and kind of celebrate the progress that we made from there. Even over the last couple of years, whenever we slow down investments, we just see margins shoot up. That's the underlying unit economics of the business showing up. So this is a proactive decision at a moment in time that is unique in our opinion, to invest in certain areas of the business to maximize the opportunity going forward. And we have the confidence that the business will continue scaling. This is not a forever state. This is a moment in time state. That's why Dave refers to it as a transition year. And we're making that knowing that the opportunity set is coming to us, and we want to make sure that we maximize it with the purposes of being the biggest possible company we can be in 5 years.

Yes. I would just double-click on a couple of points. One, we just printed a 21% op margin quarter, right? It's the best quarter of op margin we've ever had. So the unit economics of our business are very strong. I would also tell you that, obviously, we were private at the time, but we saw similar moments happening with the cloud, and we invested very aggressively in building Atlas, obviously, that wasn't available to all of you, but our existing investors saw the investment, and obviously, that's paid massive dividends over the last 7, 8 years of our growth. We see a similar opportunity with AI. Again, I cannot reinforce how -- if you believe in a world where the world is going to change only more quickly with AI, you need a data foundation that is designed to be flexible and adaptable. There's no more flexible and adaptable platform than MongoDB. And we're well optimized for the world of AI. And so I think we are investing as a vote of confidence. It would be -- candidly, it would be a much simpler conversation with all of you to say, "Hey, we're going to keep margins where they are, maybe even increase margins just because of what's happening in the marketplace." But we're actually investing as a vote of confidence because we think we see that opportunity. Customers are telling us that they want to use us to do both build new GenAI apps and help modernize their existing legacy infrastructure. And I think those things will pay dividends in the years to come.

Yes. I want to spend the bulk of -- we have roughly 20 minutes left. I want to spend the bulk of the time diving to why MongoDB is well positioned to power the next wave of modern AI applications. And maybe we sort of level set the conversation. Dave, give us your thoughts about roughly the last 12 to 18 months as 2024 progress, what's your latest thoughts on where we are in the cycle? Are we moving out of the test eval, proof-of-concept phase of the market and actually getting applications into production?

Yes. You're talking about AI applications, yes. So I think this is a gradual journey. When I look at the enterprises today, so I'll give you a couple of anecdotes. I was meeting with 2 large financial services company in New York, and I asked them how many AI apps do you have in production? One person told me about 25, 30, and other person told me about 20 to 15. And I asked them, how many of those AI apps are customer-facing? Both of them said 0, right? Because they're very, very nervous about the risk of hallucinations, especially in the regulated business like financial services, right? So while they're very interested the use cases they're looking at is more around document management, customer service, just streamlining efficiencies to sort of play with agents. But the challenge you have is that unless you really feel confident about the outputs you're getting, you're going to be very measured in terms of the deployment model. That being said, when we think about -- and I just kind of -- I'm old enough to remember the Internet area where like people are kind of building simplistic static web pages, and no one thought the Internet would transform not only everyone's business, but everyone's life, right? I think AI will have the same impact. And I think you're seeing the innovation happen. We started with basic infrastructure, basic apps. Now you get into more sophisticated infrastructure and you're starting to see more sophisticated apps. I think that's happening. Candidly, I think the first generation of GenAI apps will show up as ISVs. You have companies like Harvey and like the code generation tools and all that, I think that's where it's going to start showing up first. But I think the only way an organization can really differentiate themselves is by building custom purpose GenAI apps that are meaningful to their business. Then why is MongoDB well positioned, right? So let's talk about that. One, fundamentally, we are a distributor architecture where we use a document-based approach to manage structured, semi-structured and unstructured data. A lot of people say, well, Postgres supports JSON, can Postgres do what you do. If you look at any performance results, as soon as you introduce a 2 kilobyte JSON document or object, Postgres starts having performance issues because Postgres or relational databases need to do something called offload storage. And offload storage is a different technique for managing data that can fit in rows and columns. In fact, Postgres has this approach called the oversized-attribute storage technique or TOAST, which is a way for them to deal with unstructured data, but it comes at a performance hit. Postgres is also very rigid in schema. So it's not very easy to change. Postgres is not designed to be a single node system, not designed to be a scalable system. Some people have tried to make it a little bit more scalable, but it still has inherent scalable limitations. So when we think about a world that's going to change and adapt and deal with different types of data and different types of modalities, we think we are well positioned. And just to be clear, a lot -- I get the question, what is the role of the database in the world of AI. I think of the LLM as the brain. I think of the database as the state machine and the memory machine. And then like I get a question like what's Voyage AI. Think of Voyage AI as a very fancy library. So let me explain what I mean. Imagine you hired Albert Einstein to be your personal assistant, the smartest person in the world. You asked him a question about chemistry or biology, even no matter how smart that person is, they still need to go do research to essentially find -- give you the right answer to a hard problem. And the challenge is like they could go read every book in the library or -- and they could essentially get all the information, they come back and answer but that will take a long time and cost a lot of money. Or you can say for this particular question, go to this section of the library, go to this aisle, this shelf, this book, this page, and this section of this page to get the answer you're looking for to formulate a response. And that enables you to be much more efficient about how you do very sophisticated search and retrieval. Remember, this is data sitting in an enterprise. Obviously, the LLMs have been trained on Internet data, but they don't have data in the businesses that you're in or large banks or financial services, telcos, media companies, tech companies. So you need embedding models to really help the LLMs become much more efficient in order to produce the results and produce high-quality results. And so we feel like as the world produces more software, and you're going to see the software use cases expand dramatically, that traditional software not dealing with open-ended questions, reasoning, natural language kind of interaction, different kinds of user interface, et cetera. So if the envelope of software is going to increase, then by definition, you need more data infrastructure to support that software. And you need real-time data. That's the other question I get is, what about you versus or Snowflake or Databricks. Imagine you're -- again, back to an agent doing investment decisions. For you to make an investment decision, you might have been trained, but to act on that decision, you need to know what does that stock trading at what is the volumes, any other kind of real-time data to make a buy or sell decision. If you're a customer chatbot, you need to know exactly what's happening with that customer to be able to respond appropriately in terms of how to answer that customer's question. So that's where real-time data becomes incredibly important for these kind of mission-critical applications. And we think when you look at all the requirements, we are well positioned for that.

Yes. To pick up on some of those things that you just laid out and sort of incorporating the Voyage AI acquisition, kind of rewind back 2 years ago, vector database companies was kind of the hot thing, increasingly database companies across the market have embedding Vector Search capabilities and now you guys seem to pushing the puck forward with having world-class embedding models, reranking capabilities. So it sounds like the unlock here is about bringing a solution to the customers. So...

Not just a solution, but a solution and an elegant user experience, right? It's back to the point like the most successful companies have been able to take friction out of the user's workflow and be able to do things. Customers still have to go get their data embedded to use -- you can't use a vector database without using -- without embedding your data, but they'd have to go to OpenAI, they have to go to Cohere or they have to go to some other third party. And we said that's a very painful process. Most enterprises don't know which embedding models to use, don't know which operation store to use, not sure which LLM to use, and then they got to figure out what vector store to use and stitch it all together. That's why we want to make it much more simple and easy for customers because we can bring everything to bear in a very elegant user experience.

The classic kind of MongoDB value prop.

Exactly.

And so could you talk about what have been -- what's been the storyline in terms of Vector Search adoption, RAG adoption, you guys released it generally available earlier last year. How has that been building momentum? And do you think the acquisition with Voyage AI is going to unlock more of those RAG and the Agentic use cases.

Yes. So the uptick has been good. We've talked on the last call on this call, like we have had thousands of small customers building AI apps. We have a couple of large well-known AI companies using us as their memory and state machine for the use cases they're doing. Obviously, with the competitive dynamics of AI, they don't really want us to talk about who they are right now. But we're seeing -- starting to see some of those apps starting to take off, and these are like 7-figure workloads. And what I would say is -- in terms of your question around vector, we're seeing adoption, but I think the voice thing really makes it so much easier and so much more compelling to use MongoDB. So that is truly a one-stop shop.

What's going to be the time line to integrate the embedding models and the re-ranking capabilities into to the core...

Yes. So the way we're doing it today, you can go to Voyage AI and you can get their models either from them or you can get them from AWS, Bedrock and a few other places. What we're going to do later this year is do something called auto embedding. So you can choose as data comes into MongoDB, you can choose that you want that data embedded. So out of the box, all that is taken care for you. Then we will focus on building domain-specific models. Right now, there are models for financial services, their models for coding, but we see a lot of customers like health care customers saying, "I want models," but a lot of health care data is not publicly available. So we can go to health care customers and say, we can build models for you that are optimized for your particular use cases, your particular data and then enable you to leverage the power of AI to really do profound things in your business. And that's something that customers are quite excited by. I was talking to early-stage company that's growing very, very quickly in the -- in the life sciences and biotech space, and they're super excited about being able to leverage some of these models to just improve the performance and the accuracy of the outputs that they're getting. So there's a big opportunity. And then there's other sophisticated things we can do like instruction tuning, where you can do very sophisticated things around the models that we will just have a roadmap for going into next year.

You talked a little bit about the advantages or maybe the limitations of Postgres. If you look at big picture at the operational database market, Relational is still 2/3 to 70% of the operational database market. So my question is that it seems like Relational has a sort of, I would call it, a supply-side advantages. Kids go to school, undergrads in computer science, they still learn SQL. So you're pumping out thousands of students a year that now SQL and Relational. And so if we take a step back, what is the MongoDB strategy to penetrate that supply side and get developers to learn MongoDB whereas every year, you have new developers being pumped out learning relationship.

Yes. There's an old Chinese proverb, we love to use inside MongoDB saying, if you tell me, I will hear, if you show me, I will see, if I experience it, I will learn. And what we find is MongoDB is one of those technologies that's well known but not known well. I'll give you a simple example. We have a large financial services customer 2 years ago who was spending a little bit over $1 million with us. Fast forward 24 months later, now they're spending $22 million with us. So 8-figure account, first digit starts with -- it's not a 1, okay, pretty good growth. How much upside is there in this account. During the account review, I almost fell off my chair when I found out that only 5% of the developers know how to use MongoDB. Now they have tens of thousands of developers, but only 5% know how to use MongoDB. So that 8-figure account could easily be a 9-figure account. And our -- what we have to work on is being able to get those developers to more easily understand how to use MongoDB, not just know about MongoDB. How easy is to work with an organized data, how easy is to chart or scale out data, how easy is when you natively embed text and vector into your platform and then now with embedding models that user experience is so much more simple and less complex. And when we do that, we see the growth, which is why we're also investing up market because we see a disproportionate -- the sales productivity is disproportionately higher in the high end the market than the mid-market. And because of that exact reason that there's so much opportunity these accounts even when we have a large presence.

A key part of your strategy for growth. One of the pillars I should say, for your growth is your AI Relational Migrator service. Now there's been several pilots going on at some very large customers. What have been the results thus far from the customer's perspective? And what is the go-to-market and investment plan to scale these early successes to the rest of the customer base?

Yes. So just to level set, the database market is one of the largest markets in software. But a big part of that market is these trapped legacy applications running on legacy databases. That's frankly very, very hard to move or change, and they're hard because, one, there's lots of lines of code and, two, they're crown jewels of an enterprise, and so people get very nervous about messing around with them. However, those companies are facing technical debt. So 90% to 95% of the budgets are just spent on maintaining those applications. They're running into end-of-life issues where a lot of those technologies are getting end of life. There's compliance and regulatory issues where the regulators are saying, those applications are becoming a systemic risk, especially in like financial services or health care and other places. And then when you say, okay, how are you going to leverage AI to modernize your business, they can't do that on those existing applications. So there's a confluence of events happening with saying, we got to do something different. So when we approach customers about modernizing the applications, the receptivity is very high. And then when we show them that there's typically 2 objections at surface. One, are you selling me snake oil because customers tend to be a little cynical. So we run typically like a 6-week proof of concept where we say pick an app, and we'll show you how you can modernize and the results end up being very positive. And in some cases, the customers have stopped those pilots or POCs because they've seen enough to say, can you convince me. The second question is, okay, why MongoDB and then we walk them through what I just talked about in terms of our architecture and everything that comes with it. And we get buy on that front. And then we start monetizing. We already have a couple of proof points. We already have issued some press releases. And what we have realized is that there's so much demand, we want to move -- be able to move fast. So we've actually decided to focus on Java apps running on Oracle. Just to give you an example, we're working with 1 large insurance company to basically modernize the key underwriting application. There's tens of thousands of lines of store procedures sitting inside the Oracle database, right? Typically, the way people build applications previously was to write application logic but also application logic -- the app tier as well as application logic at the database tier to get better performance. The problem is that over time, it becomes the spaghetti code that becomes very difficult to unravel and you kind of get locked into the platform. And what we can now show is that we can reason through all that code, we can chunk up that code, peel off pieces of the functionality, and we're doing engagements right now with this global insurance company in Asia, and they're going country by country, and as soon as we knock off the country, they bring us to another country and the pipeline of opportunity just in that account is growing dramatically. That's just one small example of what we see across other financial services customers, telecom customers, older ISVs, who want to modernize. We had an ISV in Germany that asked us to modernize financial application, which you think financial applications all structured data. But they said they can get far better performance and be able to add features more quickly if they built it on MongoDB and we've done that.

And this sounds like super exciting opportunities. And so if I look at like I'm going to call it your new workload opportunity, including the AI Relational Migrator service. Let's put Vector Search, RAG in there as well, let's put Stream Processing there. If we think about these as a bucket or a class of opportunities, when do you think that starts to move -- benefit growth in your cloud business? Is that something that happens this year? Or is it...

I mean I think part of the reason for stable consumption growth is just starting to show up in the numbers, at least the search is starting to show up in our numbers. I guess it's all part of Atlas consumption, it's hard to disaggregate, but that's part of driving our confidence on the stable consumption growth that we talked about in Atlas. And I think, as I said, this is a year of transition, we clearly have an appetite to grow faster and deliver better margins. I recognize there's a lot of people in the audience who might be wondering, is this what we expect the business to do? And I can tell you absolutely not. We are very, very motivated to grow much more quickly and do it much more efficiently.

Yes. And so maybe with the last minute or so, maybe they just take the opportunity to sort of speak to what excites you about MongoDB today. The business is a $2 billion scale. You've got a world-class cloud business. Looking forward, what excites you? And why do you think MongoDB is a good investment opportunity at these levels?

Well, I'm happy that we're a $2 billion company, but in some ways, I'm a little pissed that we're not a $20 billion company, right? And I think that's the long-term opportunity sitting in front of MongoDB. And as excited I was about building Atlas. And again, we were a private company and a lot of people are skeptical, wait a minute, you're going to partner and compete with the hyperscalers, who's done that, how can you prove that why won't they strip mine your product, you know all the bearish cases against MongoDB, and we were able to prove that. I'm equally excited, if not more excited about the opportunity that AI presents. And I think it will be -- I think you're starting to see the market shake out. We saw DataStax get acquired by IBM. I think a lot of these kind of single-function databases, a point solutions just can't scale and grow very, very quickly, and I think that gives us more opportunity.

Maybe just to leave the last one, on of the debates that we've been getting this morning since we started a little bit late. A common question I get, hyperscaler competition. Is there any truth to that in terms of that being the ability to bundle, maybe Cosmos or DocumentDB? Are you seeing any signs of that as a potential headwind on the business?

Yes, that's been a question we're getting ever since the AWS launched DocumentDB in January of 2019, and our Atlas business only grew faster since then. What I would say is our relationship with the hyperscalers has never been better. It's actually really good. They're actually working with us on some of these app monetization efforts and programs. And they do that by funding some credits and so on and so forth to get customers to move more quickly. And then we're partnering in the field from a sales point of view. Clearly, there's an era of coopetition. They do have their first-party services, but we've been in this business long enough. We know how to partner and compete. And what we find is like if there's a hyperscaler who doesn't want to work with us, there's 2 others who are happy to go after that opportunity together. So we know how to kind of use that motion to help drive business.

Awesome. With that, we're out of time. Thank you, Dave. Thank you, Serge for updating us on the MongoDB opportunity.

Thank you.