Welcome. Mic on? Mic on? Yeah, it should be on. Should be on? No? Can you hear me on the mic? Awesome. Yeah. Excellent. Jay and Rohan, welcome to day one of Communication and Technology Conference 2025. It's your fourth year in a row in September, I think. Delighted to have you guys here and delighted to have our clients join us from literally all parts of the world. We've had clients from Europe, Asia, the U.S., and I'm missing many other countries. My apologies for that. It is going to be an exciting, action-packed, information-packed, insight-packed conference. AI is going to not kill software, but it's actually going to make software flourish. It's the longest introduction to a technology conference. With that out of the way, Jay, I've had the pleasure of hosting you many, many, many years ago when the company was small. Congrats on how long the company has come. If you look at what you've been able to accomplish the last several years, frame it in the context of where you're going. What is your vision for the company and how that has changed? What are the things that have changed? What are the things that have not changed as to what you want Confluent to be in the next four to five years? Yeah, that's an exciting way to start. When we started the company, the goal was really to be able to connect all the parts of an organization and have data flow to all of them and make it so organizations could act in real time on whatever was happening in any part of the business. I think in large part that vision hasn't changed, but we've come a long way in actually realizing it. The adoption of the core open source Kafka has continued. Our cloud offering has grown to be more than half the business and now very prevalent in the set of users of Kafka. We've started expanding from just having the streams of data to having a rich set of functionality around that. How can we capture these streams of data from all the systems that hold data in an organization? How can customers process that data in real time? We use an open source technology called Apache Flink, but we built a managed service around that that's on a very fast growth trajectory. How can they govern the flow of this data? How can they ensure that it's correct, that they understand where it's going, understand the quality of this data, be able to really build and depend around it? Probably one of the biggest changes in what drives this is the rise of AI. When we talk about AI, we often talk about this challenge of building models, which is a fascinating problem. It's a problem that roughly five companies need to go solve. For every other company in the world, the AI problem is in large part about how they can combine their data with these off-the-shelf models, how they can put those two things together to actually be able to make decisions that are unique to them, that are timely, that take into account all the information that they have. That's where I think Confluent can have a ton of value in gathering that information from across the enterprise. I think that's a significant catalyst in moving out of kind of slower batch data collection mechanisms. If you think about agents or software acting as part of the business, I think it ultimately has to be acting with knowledge of the current state of the world. How are things right now? I think we're seeing that in our customers as they start to think about the architecture that enables them to take advantage of AI. We're seeing it in the use cases around that. I think that's a significant driver in this space. Not surprisingly, some of the first use cases for open source Kafka, when we had built it and rolled it out at LinkedIn, were around these machine learning use cases, which were going to bring together data and make smarter decisions. I think this has become something that has a much broader scope in the types of problems that you can attack. It is actually just much easier for companies to use than traditional machine learning. I think those are some of the things that have changed as the company has grown and got to scale. As you look at the core value proposition of Confluent, being able to collect the real-time data, does AI represent just a continuum in that you are able to better enhance the value proposition of Confluent? Is it like a break point where customers are like, okay, so this is what Confluent does, and I'm trying to still connect it with what AI can do for me, or is it more seamless and more fluid in the way? Yeah, you know, maybe a little bit of both. I'll start by just saying, like, hey, what's our role in AI use cases? And then, you know, has that changed things or not changed things? So, you know, there's two use cases around streaming and AI. The first is, you know, how can you actually get the right bits of context data, transform it in the right way, and apply it at the right time? And how can you iterate on that, evaluate it as this changes, be able to reason about whether, you know, your agent or AI application is getting better or worse as you're adding bits of data, as things are changing, as you're upgrading the model. This is something that's particularly well suited to streaming. If you think about it, there's just not that many other ways of collecting and working with real-time data that are common. The second is actually applying, you know, agents or AI to events in the stream. You can think of it as a kind of streaming agent. When people say agent, there are lots of things they mean. They may mean fancier chatbots, they may mean something that kind of clicks in your computer. What I mean is, you know, kind of the particular set of use cases around business automation, where there's some background task, something happens in the business, you're trying to kick off some activity, that's something where we've added functionality. Those are the two things we're doing, right? Both of these have gotten good adoption throughout our customer base. Now, okay, is that a continuation or a break point? Yeah, I think in a way it's a little bit of both, right? If you think about how the technology is used, this is similar, you know, getting real-time data is similar to how we were used all along. In terms of what we're for, it's a continuum. I do think that the way in which it can be, you know, something quite different is if you feel that these AI use cases are going to be a significant portion of what companies are doing over the next five years, maybe half of the new use cases will be that. I think you can say, okay, that's a bit of a phased change in, you know, when you think about your investment in data or in software, that represents a sizable change. There I think you could see it, you know, kind of this way, like, you know, if you go back five or ten years, you know, what did it mean for a company to be data-driven? Maybe five, ten years ago it meant, okay, we could get all our data into our data warehouse and that data was correct and we could build good reports and our executives knew what was going on and they could make informed calls, right? I think if you think about what does it mean now, what is it going to mean going forward to be data-driven, it's like, yes, you still need that, right? That hasn't gone away. I think the promise of this technology is to be able to bring that decision-making into the flow of the business, right? Transaction by transaction, customer by customer, be able to do something that's very personalized, that would not be scalable with humans or executives, be able to do that in a way that's efficient and takes over a lot of the manual effort in the business. That need is very different from kind of batch collection of data at the end of the day. It is much more an operational proposition. I think if that becomes something that's quite widespread in companies, then I think that does represent a phased shift in how they think about data. Just listen to, promise me to ask the question. How do you see this evolution of Confluent with AI in the years ahead? Does the tech roadmap kind of evolve with it? You hired a new Chief Technology Officer. How do you see the roadmap and how the product strategy kind of changes to evolve the AI infusion into Confluent? Yeah, yeah, yeah. This is obviously something we've watched for quite some time. We wanted to make sure that first we had just really good integration into the kind of analytics and lakehouse ecosystem. It shouldn't be the case that you're kind of choosing between analytics and operational applications in many ways. AI is kind of blending between those two. We've done work to take all the streaming and fully integrate that into Iceberg and Delta and these open table formats. You can kind of take the stream and have it also be the long-term store of data. That matters for AI because AI use cases are both acting on the new stuff, but also kind of combining some of the historical data. The next thing is really opening up the access to the best models in the streaming and stream processing, building integration into a lot of the different data stores and frameworks that are part of the modern AI stack. Those have all been things that we've been quite active in pacing, even in a landscape that's shifting pretty rapidly. A big part of really realizing the opportunity is actually bringing our stream processing story to scale, right? A lot of being able to act on data starts with these processing capabilities. For data systems, this is something almost every data system has as some processing layer. It's a big part of the value they add. In the real-time streaming area, it's been seen as harder. It's been seen as more expensive, more elaborate, more difficult to do. What's changed in that space is I think we've really understood the computer science of how to build real-time processing in a way that it need not be more expensive or more complicated just because it's done continuously, and with low latency. A lot of this has come out of this open source project Flink. This is something where we brought in a lot of the core people for that. We've been really building out a first-class serverless stream processing layer. That went GA last year and just got out to the different clouds with all the private networking technology. Over the first six months of this year, it grew 3x in revenue. It's still early part of the business, but very rapid growth. I think one of the drivers there is this kind of modern analytics and AI opportunities that blend the world of data from batch all the way out into real-time. We think that there's a ton of promise in that area as that further unlocks. That's one of the most important things I think to go capture that opportunity. Got it. A couple for you, Rohan. We had Q2 results, had about 21% subscription revenue growth rate. We also talked a little bit about consumption headwinds. When you look at things today, when you look at the different customer segments, what customer activity are you seeing with respect to consumption, uptake, downtake, et cetera? Yeah, I mean, we're in Q3, month three of the quarter, and a few weeks still to go. At this point, I can tell you that the team is very heads down and just focused on executing and having a strong close to this quarter. We generally do not provide mid-quarter updates on consumption trends, but what I'll tell you is, in our Q2 earnings. Except at Goldman Sachs. Except. Safe cash. What I'll tell you is some of the green shoots that we called out in our Q2 earnings call, you know, be it just a progression of pipeline and generation of late-stage pipeline or the Flink momentum. We're actually pleased with the execution that we are seeing. Obviously, it's still early days, but we are pleased with the progress and execution that we are seeing. Specific to consumption trends, we'll provide a more detailed update in our Q3 earnings call. Got it. Also, along with that, NRR had moderated a bit. We can understand the reasons why. What's the best way to think about this metric on a go-forward basis? What drives this metric? Yeah, I mean, when you look at the trajectory of NRR, there tends to be puts and takes. I'll reiterate a couple of comments I made at our Q2 earnings call. First is, you know, when you look at our cloud business, both NRR and GRR, the way it is calculated, it is essentially you take the last quarters of consumption and then you annualize that metric. Typically, consumption trends in more recent quarters have a more amplified impact on this metric. Just given that dynamic over the short term, you know, that's going to be a headwind to this metric, right? However, having said that, some of the green shoots that we spoke about, some of the growth drivers that we spoke about, and our new CRO, Ryan McMahon, is driving, be it the Flink momentum, be it late-stage pipeline progression, warp stream, these are all going to be tailwinds. There are some puts and takes, but as we look ahead at balance, you know, we have a bunch of growth drivers and that can sell into our existing customer base. As I think about medium to long term, there are a bunch of growth drivers for this NRR metric to get more tailwinds. Jay, I want to jump back. Anybody knows Matt Martino on my team? Everybody, watch out for this space. Thanks, Josh. Jay, I want to jump back to the AI piece, right? I think, and maybe I may be misquoting here, but I think you said 10x production use case improvement within a subset of your largest customers, right? I think when you look at the broader kind of consumption landscape, you tend to hear kind of mixed signals around where we are in terms of the shift from experimentation to production. Why do you think that Confluent is kind of seeing it first, if you will? Yeah, you know, I don't think we're alone in this, right? There's definitely a set of companies that are kind of part of the stack for AI workloads. You know, and there's probably some which are earlier in that adoption cycle where they're probably part of the prototype. We're probably more part of the production usage where it's like, hey, where do you need the real-time data flow at scale where it drives consumption? Probably as that thing goes to production. We've seen a pretty steady progress of customers. It kind of starts with the tech companies that tend to move the fastest, but it's really getting out to, I think, a broader set of organizations. If you look at what they're doing, it's true that there's a lot of experimentation and probably not every project is going to make it, but there are very significant things that are working. We've seen that when we count something as this kind of consumption pipeline, that means that we think it's a needle mover in that account in a way that's significant enough for the field team to track. That's definitely gone up 10x from last year. I think that trend will continue. Right now there's a set of things that can be done with these models. I think a lot of organizations are learning how and where they can actually put that into practice and get results. I think as that learning curve continues and as the models continue to improve, we're going to see continued adoption of this stuff. I think probably even better focus of where it's being directed. Ultimately, kind of as I said, I do think these AI problems, when it comes to enterprises putting it into practice, it actually does play out as a data problem for them. That is the part that they have to put effort into. Both the bringing together of that data, but also the iteration on this as they try and get something that works well enough that they can actually put it out into production, be able to iterate on that, run evals against it. That kind of iteration loop, I think, is very well suited to the streaming world, which can go from the kind of real-time data to the historical data and back and forth, which is very much what's needed in that space. What are some of the more powerful use cases you're observing in the business today? Yeah, I think there's a set of things that I think are really interesting, right? Early wave of use cases around support, which is kind of like, hey, how can we bring together the information about the customer to be able to get them the right thing at the right time? We talked about a large online travel company that had done this across all their different lines of business. We've seen some of the tech companies that have embedded this in their products, like Notion as a customer of ours, and it's talked about how we're writing the flow of all the data for AI for them. I think it's a particularly sophisticated user of AI that was very early on the adoption. Now you would start to see this getting out beyond the Notions and the cursors, which of course are those customers that are right at the center of things, out to insurance companies, companies in healthcare that are really starting to apply this to some of their core business flows. I think we're seeing success in some of those projects. Now, anything to move the needle overall for the company, it's not about one use case here, one use case there. You have to really take it to scale. I do think the most interesting use cases are these ones that are unique to these businesses where it's right at the flow of what they do. It's an insurance company working on claims processing. It's a bank working on some of their core flows. I think those are the most exciting. Sometimes they take a little bit longer, but the ROI on those things I think is very high compared to what they can get out of other packaged solutions. I think there's a huge opportunity to be around that. Yeah, very interesting. Rohan, for you or Jay, the field sales reorganization in 2Q, can you just tell us a bit about, you know, why now? How do you see this kind of benefiting the business long term? Talk about some of those changes that you're making. Yeah, do you want? I can start and maybe at least jump in. Yeah, you know, we've been very focused on the go to market over the last, you know, really year and a half. The goal has been to orient around consumption, drive more efficiency. We've made a whole set of changes around that. One of the things, as we were bringing Ryan McMahon into the CRO role that we wanted to really get right was some of the assignment of technical resources to accounts. Really as part of our efficiency work, we tried to spread that as thinly as possible. One of the things we found was, hey, if you want to go drive consumption workloads and you want to do that at scale, one of the most important things is technical resources that have continuity in the account and getting that right. The focus on how we land customers, getting that right. We're doing that both with takeouts of existing open source as well as some of the offerings from CSPs. Really orienting around those projects. Those are some of the key things that we felt like, oh, there's an opportunity to make this better. We were able to do that in a way that was cost neutral, which was important for where we're at. We've seen, I think, very good results off of that so far. We look at a couple metrics. We talked in our last earnings, we measure the rate of progress for these projects that will be consumption drivers. It's kind of the consumption equivalent of pipeline, but it's not like a PO, it's like an app, right? That was a substantial step up Q1 to Q2, going through these changes in process. That's been reflected in the kind of growth in RPO and CRPO, which would reflect step ups in commitments, right? As customers feel like, hey, we have these projects coming, we want to take up our commitment with Confluent. Those are, I think, both positive early indicators of what we've seen. We've also seen very strong results from what we would call DSP, a Data Streaming Platform, going from just the stream of data Kafka to the full set of offerings around. This is a smaller part of the business. At cloud, I think a few quarters back, we said it was about 13%, but that's growing very fast. The question for a company like Confluent is, how can you go from having one thing to having multiple things and executing on all of them at once? Getting that specialization model right, I think, is very important. I talked about the growth of our Flink offering. I think one of the key aspects of that is the maturity of the technology. I think there's a ton of latent customer demand, but one of the key aspects is actually having the right go-to-market model where you're not asking too much of the line sales rep. There's somebody who's a deep expert in real-time processing who can kind of help them out as something becomes more significant. I think that that's worked. Even though that Flink offering is very new, we're seeing hundreds of companies with early use cases and then a set of customers that have really gone big and gone big to success where we have customers that are over $1 million in just their Flink spend and have really taken whole complicated batch workloads with hundreds of individual batch jobs and moved them over. It may not be easy to appreciate, but this area of kind of serverless data processing is very technically hard. The hardest thing to do is have a bunch of workloads that were written for some other system that are going to just get force fed over to the new thing. We're actually lucky that we can do that now in part with AI. These AI coding tools are very good at these migrations, but in part, the underlying system then has to have a certain level of reliability and maturity for that to operate. I think that's a very promising thing. Obviously now our goal is to continue that growth for that offering and scale it. This work on kind of getting that specialization and field model right as we have more parts of our platform is a key aspect of making the go-to-market successful. We want it to be the case that new offerings speed up the progress of the sales team rather than slow them down. Rohan, maybe for you, Confluent Cloud is now 56% of subscription revenue. You know, but the guidance for the back half of the year sort of implies some stability in that mix between platform versus cloud, given some of the consumption headwinds. Maybe talk to us about your long-term aspirations for platform versus cloud, especially with even some of the emerging products like Flink seeing a pretty equal split there. Yeah, you know, let me take a step back and just talk about some of the growth drivers in the business as we think about the business over the medium to long term. I'll put it in four categories. The first one is just the large Kafka streaming opportunity we have. For context here, we have over 150,000 organizations using open source Kafka. We've built a billion-dollar plus run rate business by monetizing approximately 5% of that ecosystem. The first opportunity falls for both Confluent platform and cloud. Candidly, probably cloud has a bigger opportunity there. That's number one. Number two is around our Data Streaming Platform. I mean, Jay touched on it. We're making this shift from selling just streaming to selling a platform. All the individual products of our Data Streaming Platform, they're operating in fairly large markets, large market opportunities. From a growth perspective, they're in their early stages of their growth curve, S curve. On average, they're growing faster than the company. That's a growth driver. That's primarily going to be coming on the cloud, but also, there is a big platform opportunity there as well. The third aspect that, again, Jay touched on was AI. I mean, we know every AI problem today is a data problem. We are going to play a very important role in this ecosystem. Just the modernization of data architecture, data infrastructure that's going to go on to power these agentic workloads. You know, Confluent is going to play a very important role. The last one is around our ecosystem. The partner ecosystem is pretty important because it kind of provides us with scale and amplification of our story. When you think about these four growth drivers, they apply to both platform and cloud. However, in general, we expect the cloud business to grow over the long term much faster than the platform business. That's how I'd say it. We will see this mix shift continue over the long term to be more cloud. Just to add on to that, I think that there's obviously many things you could worry about about any company. The idea that companies will continue self-managing these big complicated data platforms, I think it's very unrealistic. In our product offering, we've expanded the set of cluster types we offer for our cost offering, and they've become very cost-effective. Now, compared to doing it yourself, we would be cost-competitive just against your cloud infrastructure spend. How can we do that? Is that sustainable? The way that we do that is both by really good engineering, which of course we put a lot of effort into, but also by multi-tenancy, by actually collapsing a lot of customer workloads on the same pool of infrastructure. That's not an opportunity that an individual company can do. The resulting product is not just a better deal. It's actually a much better experience because having that pool of infrastructure kind of pre-allocated and shared across customers means it's elastic. It always has the right capacity for their need to scale up and down instantly. That's just the Kafka component. Increasingly, if you look at what's working in the cloud, customers aren't just looking for point solutions. They're looking for end-to-end problems and ecosystems being put together. I think one of the things that Databricks has done such a good example job of is going from a better Spark to a complete solution in the data science space. There are a few years ahead of us in kind of executing that, but I think it's exactly the right way. It's exactly what we're doing with our Data Streaming Platform offerings. It's kind of taking the end-to-end problem around real-time data and bringing it together. I think that's a really big deal for customers because trying to tie together a bunch of little pieces yourself is actually quite labor-intensive and hard to really make progress on and hard to use against the end use cases. Both in the kind of TCO of the core thing, as well as solving the larger problem for customers, I think there's no question that customers are moving towards managed solutions. There could be competition in doing that. There could be puts and takes in a rate at which customers are adopting versus optimizing. All those things can happen. The idea that they will kind of continue doing it themselves forever, I think is highly unlikely as the end outcome. I wanted to ask you, Databricks has been clearly, there used to be the hypothesis before Databricks became a success story, that it's hard to build an open source software company. Red Hat was successful, you guys. What is it that you think Databricks is doing better versus what the industry expected that has made them successful? What are the things that applied Confluent's evolution in converting the Kafka install base, you said, Rohan, into a commercial install base? What are the things that you've been able to absorb from their example? Yeah, first of all, you know, it's true, it's very hard to build an open source software company that's, call it whatever, a billion dollars or more. That said, it's very hard to build a software company that's a billion dollars or more. If you say, is it much easier to build a proprietary infrastructure company that's a billion dollars? I don't know, that's also quite difficult, right? Particularly when you think about these new things, and I think what they're doing is a little new, what we're doing is a little new. If you want to get something new out in the world, you have to have some tailwinds that help you. Open source can definitely be one of those things that kind of draws you into conversations. It gives you access to a lot of these companies. The alternative universe where we're kind of going door to door trying to tell people about a new way to think about data, I think would not be positive. That's the first answer I would give. In terms of what they've done well, I think it's a set of things, right? They started by building something that had traction, open source Spark. Then they made a faster, better TCO Spark, right? They did a number of iterations on that where it was like, hey, if you swap this in, it'll be easier for you. You won't have to manage it. It'll be more cost-effective. You'll get your results faster. They expanded from something that was mostly Spark to really a larger ecosystem around data science, around analytics, as they brought in more tools for machine learning, as they brought in more of the governance of the lakehouse, the open formats, the tabular acquisition, et cetera. If you think about what part of that applies to Confluent, I think a lot of it applies to Confluent. I think we've done a similar thing in making a better way of getting and consuming Kafka. I think that we're now well into the execution of broadening that platform to be something like a data lake, but for real-time data. We're the operational side of the business where how businesses are going to actually have apps that execute. The story is not done, but if the first part is kind of building some of that functionality and then getting it onto that scaling curve, I think now we have a number of these product offerings that are very much on that scaling curve. I think that combination is a much broader value proposition for customers than a point solution and ultimately a much more strategic position for them, rather than something serving here and there. This is something that will really act as the central nervous system across data for them. I think those are some of the points of similarity. I think that Databricks is also a company that has done an excellent job of staying with what is current and executing well against what's moving now in the industry, whether that's big data or AI, et cetera. I think similar opportunities for Confluent to do that well. Got it. Any questions from our clients in the two minutes that we have? Rohan, go ahead. Hey, good to see you. Hey, yeah. How's it going? I think that we're about to go. You mentioned Databricks right now. When customers go to the data shared with you, kind of feeling that they're updated or getting ready for data, why don't they end up picking a third-party vendor or trying Confluent? They expect Databricks, for example, already has a stupid solution and doesn't have to go there. What are the cons of switching to a nested cloud? Yeah, yeah, yeah. Just to repeat the question, it was, you know, broadly, what does Confluent add, you know, if you're kind of getting ready for AI, if you have Databricks? I think the big thing that we add is all the streaming data, the ability to actually do this across the application estate. Spark has some kind of real-time processing capabilities, but this larger ability to actually have streaming data across the organization, that's the thing that we would bring to the equation. I think that's the basis for the partnership with Databricks. We're not going out and competing in the market. We're actually cooperating. That's not just a surface-level thing. The teams are very actively working together. It's not the kind of thing where like in Salesforce, we would have some kind of competitor field with the Databricks in the dropdown. To customers, these are very complementary technologies that would work together. Obviously, between any two data platforms, there's some bits you could do with us or you could do with them. There's some kind of border dispute you could point out. Ultimately, if you think about what is it that they're trying to do, what's the territory that they're trying to occupy, and what's the territory we're trying to occupy, those are largely just joint parts of the organization. On that note, we will wrap it up. Before I do that, I won't call out by name, but there are two people here in the audience that I've actually known for 30+ years. You are one, and the guy standing in the back. I won't call you by name. You know who I am. Anyway, that's quite a milestone. We've been doing this all together. On that note, thank you so much, Jay and Rohan, for joining us. As I said, this is just the beginning. It's four days of action-packed, insights-packed conference. Hang around because we're going to learn a lot together. Thank you so much.