CFLT - NASDAQ

Welcome to the Confluent Third Quarter 2025 Earnings Conference Call. I'm Shane Xie from Investor Relations, and I'm joined by Jay Kreps, Co-Founder and CEO; and Rohan Sivaram, CFO. During today's call, management will make forward-looking statements regarding our business, operations, market and product positioning, growth strategies, financial performance and future prospects, including statements regarding our financial guidance for the fiscal fourth quarter of 2025 and fiscal year 2025. These forward-looking statements are subject to risks and uncertainties, which could cause actual results to differ materially from those anticipated by these statements. Further information on risk factors that could cause actual results to differ is included in our most recent Form 10-Q filed with the SEC. We assume no obligation to update these statements after today's call, except as required by law. Unless stated otherwise, certain financial measures used on today's call are expressed on a non-GAAP basis and all comparisons are made on a year-over-year basis. We use these non-GAAP financial measures internally to facilitate analysis of our financial and business trends and for internal planning and forecasting purposes. These non-GAAP financial measures have limitations and should not be considered in isolation from or as a substitute for financial information prepared in accordance with GAAP. A reconciliation between these GAAP and non-GAAP financial measures is included in our earnings press release and supplemental financials, which can be found on our IR website at investor.confluent.io. References to profitability on today's call refer to non-GAAP operating margin, unless stated otherwise. And with that, I'll hand the call over to Jay.

Thanks, Shane. Good afternoon, everyone, and welcome to our third quarter earnings call. We're joining from New Orleans, where in 2 days, we'll host Current, the data streaming event where real-time data and AI come together. Turning to the quarterly results. We delivered a strong Q3, exceeding the high end of all guided metrics. Q3 subscription revenue grew 19% to $286 million. Confluent Cloud revenue grew 24% to $161 million, and non-GAAP operating margin expanded 3 percentage points to approximately 10%. This performance underscores strong consumption growth in our cloud business, the deepening commitment of our customers and our disciplined focus on driving efficient, sustainable growth. Last quarter, we outlined 2 areas of focus in our go-to-market and several areas where we were drilling down on early success, all aimed at accelerating use case expansions and supporting the long-term growth trajectory of our cloud business. I'll give a brief update on each of these. The first area of focus was tightening field alignment to drive more use cases into production. As we shared last quarter, we saw strong momentum in late-stage pipeline progression, a metric that tracks the dollar value of new use cases moving into production. That momentum continued in Q3 with more than 40% sequential growth and progressing late-stage pipeline and an accelerating pace of new use cases. This positions us for durable consumption growth and was a key driver of our cloud performance this quarter. In parallel, we continue to build momentum in expanding our large customer base, delivering the largest sequential net add in $100,000-plus ARR customer count in the past 2 years, along with continued acceleration in million dollar plus ARR customer growth. Together, these results underscore the depth of opportunity within new workloads and the continued strength of expansion among our large customers who are increasingly standardizing on our data streaming platform and relying on Confluent to meet their business needs. Our second focus area is centered on accelerating the build-out of our DSP specialist team to drive multiproduct selling. We previously highlighted Flink momentum in the first half of the year, and we're pleased to report another strong quarter with Q3 Flink ARR for Confluent Cloud growing more than 70% sequentially. Flink usage has continued to expand across our customer base. More than 1,000 customers used Flink during the quarter. Stream processing is key as it enables companies to act on data the moment it's created, turning information into real-time decisions and results. A great example of the power of our Flink offering is Siemens Healthineers, a global leader in medical technology with operations in more than 70 countries. The company develops imaging systems, lab diagnostics and connected medical devices used by hospitals and clinics around the world. Behind these life-saving technologies is a constant stream of data that determines equipment reliability, accuracy and ultimately, patient outcomes. But Siemens Healthineers was hindered by disconnected systems that isolated critical data in silos, lengthy file transfers, manual handling and periodic batch processing often delayed insights by weeks. These delays prevented timely action to improve equipment performance and product quality. So they turn to Confluent Cloud with fully managed Flink. With Confluent, Siemens Healthineers built a unified real-time data backbone that streams and processes millions of events from imaging, lab and devices daily. Flink continuously filters, joins and enriches these streams to deliver timely, trustworthy operational insights that help improve device reliability, manufacturing, quality and consistency of diagnostic data across its installed base. This foundation now gives Siemens Healthineers real-time visibility and the agility to move faster as it advances digital and AI initiatives that enhance care delivery and improve patient outcomes worldwide. Next, our partner ecosystem continues to deliver strong results. As of Q3, partners sourced well over 25% of our new business over the last 12 months. This is a clear sign of the consistency and scale we're building through our established partner relationships, which are instrumental in broadening our footprint and driving customer expansion. Confluent was named a MongoDB Partner of the Year and served as an AWS launch partner for the new AI agents and tools category in the AWS marketplace, further strengthening our position at the center of real-time data and AI. Lastly, we remain as competitive as ever replacing CSP streaming offerings. We have maintained win rates well above 90%, with average deal size, more than doubling over the past 2 quarters, all while continuing to increase our add bats. This is made possible with multi-tenant Freight Clusters, Enterprise Clusters and WarpStream, which together have delivered a 4x increase in consumption over the past 3 quarters. Because of their multi-tenant architecture, we believe adoption of these new clusters is a tailwind to subscription gross margin over time. These differentiated offerings provide superior performance and lower TCO to our customers which also helps us soak up more of the world's Kafka workloads. This includes one of the world's largest fintech companies who signed a 7-figure deal in Q3 to move their large-scale logging and telemetry workloads from open source Kafka to Confluent. Another great example of this is EVO Banco, a digital native bank in Spain serving hundreds of thousands of customers through its mobile-first platform. As transaction volume grew, its open-source Kafka clusters became increasingly difficult to scale and secure with rising operational costs and downtime during peak loads. To address this, EVO Banco migrated to Confluent Cloud as its central data backbone. The platform now streams and processes hundreds of thousands of financial events per day across payments, fraud detection and customer channels and with stream processing and fully managed connectors, EVO Banco integrated core banking systems and analytics tools in real time without managing infrastructure. Since moving to Confluent, the bank has improved reliability, lowered costs and accelerated delivery of new banking features. Q3 also marked the 1-year anniversary of our WarpStream acquisition. Over the past year, WarpStream has seen 8x growth in consumption, and we've closed multiple 6-figure deals with marquee customers across different industries, including a Fortune 5 customer. We're encouraged by WarpStream's strong first year performance and remain incredibly excited about the significant opportunity ahead. Next, I want to spend a few minutes on a key aspect of Confluent's opportunity in the AI space, providing context data for AI agents and applications. We're seeing a clear pattern across the industry. Many companies have shown they can successfully prototype AI, but fewer can get those systems into production. AI models are clearly capable, but a recent MIT study found that though enterprises are investing tens of billions of dollars in generative AI. Most of these initiatives haven't delivered the desired results. The challenge isn't building a prototype. It's being able to build reliable business systems powered by AI that makes trustworthy decisions and takes appropriate actions. There are 2 factors that fundamentally drive the quality in AI systems, the models capabilities and the data it has access to. Both of these are significant challenges, but they fall on different people to solve. Improving the quality of large-scale AI models is a challenge largely driven by a small number of LLM producing research labs. Enterprises can easily harness the results of this work by simply pointing their apps at a new model. But getting data into shape to act as context for AI is a problem every enterprise must solve with their own data. This is where Confluent can help. One of the reasons AI demos are often so successful is because they can be powered by a onetime manually curated data set. But to take an agent to production, it must have an up-to-date comprehensive view of all the inputs needed to do its work. This isn't just a matter of trying to hook the model into every source system directly. The source data is generally too messy and application specific to lead to good results. And AI Ops can't be splunking around in production databases, reading through everything and potentially leaking the wrong data to the wrong user, that would be wildly expensive, create unsustainable production workloads and be fundamentally insecure. Rather, the problem is about curating the right data for a given problem and creating a data set an agent can be tested with and evaluated against. Maintaining that live context is what determines how well an AI system performs. That's where accuracy, relevance and trust are won or lost. What businesses need is a system that can keep data in motion so it can be processed, reprocessed and served continuously as it changes. Our data streaming platform was built for exactly this problem. It works to connect data from every system application and cloud and support just these kinds of complex pipelines. With Kafka, Flink and Tableflow, teams can process in real time, combining history and live events with one unified engine. With logic changes, you can go back and reprocess data to create the new data set. Tableflow and Flink work to combine the best aspects of real-time capabilities with the long-term historical store of data in the lake. As this goes out to production, the stream of feedback data can also be captured to measure the effectiveness of each change. And in 2 days, we will host Current and unveil new capabilities that are designed to make this even easier for customers and strengthen how our platform delivers real-time government context. Confluent data streaming platform is becoming the context layer for enterprise AI as businesses move from AI experimentation to production, from static data to living context, and from analysis to intelligent action. One customer that really illustrates this is a multibillion-dollar health and fitness chain with nearly 200 clubs in a rapidly growing digital platform. As the company expanded into AI-powered wellness, its data from wearables, class bookings and mobile apps was siloed and processed in slow batches. This made it impossible to provide real-time personalized guidance through its Gen AI companion. With Confluent Cloud as a streaming backbone, this customer now continuously ingests and enriches this data in motion. Wearable metrics, work out history, purchase activity and engagement events are streamed and combined with contextual data, like recovery status or performance trends before being routed into AI systems to fuel personalized recommendations. Confluent enables them to deliver AI insights in seconds instead of ours, scaling to millions of real-time interactions while enabling security and compliance. Fully managed infrastructure frees engineers to focus on innovation, helping the company turn decades of wellness expertise into intelligent, context-to-ware experiences that deepen member engagement and fuel digital growth. As AI evolves from innovation to utilization, context will define who wins, and we are committed to making Confluent the company enabling the shift by turning data and to continuously refresh trustworthy context for AI systems everywhere. In closing, we're encouraged by the strong out consumption growth and the traction we're seeing for our complete data streaming platform, particularly with Flink. As AI becomes operational across every industry and geography we believe that the demand for real-time context powered by data streaming will only grow. It's an exciting time for Confluent and we're just getting started. With that, I'll turn it over to Rohan.

Thanks, Jay. Good afternoon, everyone, and thank you for joining our earnings call. Our strong third quarter performance highlights the momentum of our data streaming platform and our diversified growth strategy. We delivered strong top line growth, stabilized our net retention rate, increased the adoption of new products and drove continued margin expansion. These results demonstrate our ability to drive durable, profitable growth at scale over the long term. Turning to the results. Q3 subscription revenue grew 19% to $286.3 million and represented 96% of total revenue. Confluent platform revenue grew 14% to $125.4 million, driven by healthy demand in financial services. Cloud revenue grew 24% to $161 million, representing 56% of subscription revenue compared to 54% in the year ago quarter. We are pleased with our cloud performance this quarter, which was driven by stronger consumption across core streaming and DSP, including acceleration of new use cases moving into production. Turning to the geographical mix of total revenue. Revenue from the U.S. grew 13% to $172.1 million. Revenue from outside the U.S. grew 29% to $126.4 million. Moving on to rest of the income statement. I'll be referring to non-GAAP results unless otherwise stated. While driving top line grow at scale, we continue to show significant operating leverage in our model. In Q3, subscription gross margin was 81.8%, above our long-term target threshold of 80%. Operating margin increased 340 basis points to a record of 9.7%, exceeding our guidance by 270 basis points. This was driven by revenue outperformance and improved sales and marketing leverage from continuing to streamline coverage to drive growth. Adjusted free cash flow margin increased 450 basis points to 8.2%. Net income per share was $0.13, using $370.6 million diluted weighted average shares outstanding. Fully diluted share count under treasury stock method was approximately 382.4 million. We ended the third quarter with $1.99 billion in cash, cash equivalents and marketable securities, reflecting the strength of our balance sheet. Turning now to customer metrics, $20,000 plus ARR customer count increased to 2,533, up 36 customers sequentially. $100,000-plus ARR customer count was 1,487, up 48 customers quarter-over-quarter, representing the largest sequential increase in 2 years. New $100,000-plus ARR customers include many leading AI companies such as Forbes 50 AI analytics provider, an AI-powered SIEM cyber security vendor, a next-gen AI automation platform company. Our $100,000-plus ARR customers continue to account for more than 90% of our ARR, $1 million-plus ARR customer count increased to 234, representing growth acceleration of 27%, driven by new use case expansion across cloud and platform. Additionally, more than 10 of the 15 net new $1 million-plus ARR customers increased their spend on DSP products over the previous quarter. NRR for the quarter stabilized at 114%, while GRR remained close to 90%, driven by stronger consumption growth in our cloud business. Turning to our outlook. For the fiscal fourth quarter of 2025, we expect subscription revenue to be in the range of $295.5 million to $296.5 million, representing growth of approximately 18%. Non-GAAP operating margin to be approximately 7% and non-GAAP net income per diluted share to be in the range of $0.09 to $0.10. For fiscal year 2025, we expect subscription revenue to be in the range of $1.1135 billion to $1.1145 billion, representing growth of approximately 21%. Non-GAAP operating margin to be approximately 7%, non-GAAP net income per diluted share to be in the range of $0.39 to $0.40 and adjusted free cash flow margin to be approximately 6%. For modeling purposes, we expect Q4 cloud revenue to be approximately $165 million, representing growth of approximately 20% and accounting for approximately 56% of subscription revenue based on the midpoint of our guide. Turning to the key drivers of our business. We saw strong demand in our core streaming business and good momentum across DSP, AI and our partner ecosystem. First, our continued focus on field alignment is delivering strong results. In Q3, we accelerated the pace of moving new use cases into production and sustained strong momentum in building our late-stage pipeline, which once again grew more than 40% sequentially. We're also seeing customers commit to larger and longer-term deals, reflected in RPO growth of 43%, another quarter of acceleration. Together, these trends give us greater visibility into near-term consumption revenue and increase longer-term visibility with improved RPO to revenue coverage. Second, we saw good DSP momentum across cloud and on-prem in Q3. Building on the momentum from the first half of the year, we delivered another quarter of strong performance for Flink with particular strength in cloud. Q3 Flink ARR for Confluent Cloud grew more than 70% sequentially, and we now have more than 1,000 link customers, including more than a dozen customers with greater than $100,000 in Flink ARR and 4 customers with greater than $1 million in Flink ARR. This comprehensive breadth and depth represents the foundation for scaling into a very significant Flink market opportunity ahead. Here are 2 customer examples to illustrate how Flink begins to drive ARR expansion in our customer base. These customers are spending currently north of $100,000-plus and $1 million-plus Flink ARR, respectively. Notably, in the last year alone, adoption of Flink has supported both customers to more than 6x total spend. Third, we are strongly positioned to deliver contextualized, well-governed and AI-ready data to companies. We now have more than 100 AI native customers, including 21 with $100,000-plus in ARR demonstrating Confluent's highly strategic role in the age of AI. Fourth, we are pleased with seeing continued traction in our partner ecosystem. On a trailing 12-month basis, Q3 partners sourced deals increased to more than 25% of our new business, up from more than 20% last quarter. As we grow beyond the $1 billion-plus revenue scale, we expect partners to play an even bigger role in driving growth and leverage in our business in the years ahead. Lastly, we've continued to demonstrate the effectiveness of our disciplined ROI-driven capital allocation strategy, especially in M&A. Q3 marked the 1-year anniversary of our WarpStream acquisition. And in just 1 year, WarpStream's consumption has grown nearly eightfold. Following the Immerock acquisition, we shipped our Flink product in spring of last year. And since then, we've scaled Flink into a low 8-figure ARR business. The strong financial performance underscores the successful path both products around and reinforces the strength of our overall capital allocation strategy. In closing, we delivered strong third quarter results demonstrating durable top line growth and margin expansion at scale. We are encouraged by the strong consumption growth in our cloud business and remain focused on continuing to execute on our key growth drivers across core streaming, DSP, AI and the partner ecosystem. Looking forward, we believe we are well positioned to take advantage of the large market opportunity ahead. Now Jay and I will take your questions.

All right. Thanks, Rohan. [Operator Instructions] We ask that you kindly keep it to one question and one follow-up. And today, our first question will come from Brad

Great. And good to see the good results, especially the accelerated bookings, really impressive. Jay, I want to follow back on some of the go-to-market changes that you made last quarter. The field alignment, changes in coverage ratios. And it's great to see the momentum in late-stage pipeline continue. What are the learnings now that we're another quarter into these changes? And what conversion trends can you share on all this new pipe? And how should we think about the capacity to effectively work that much incremental pipeline?

Yes, those are great questions. So yes, we put a number of things in motion heading into this year. And particularly over the last few quarters, I called out some of those, the specialization model for DSP. That's really important just to be able to take these new products to scale, and it's working really well. A number of aspects of just kind of field execution around consumption. I think that's one of the biggest drivers of that kind of progression in consumption pipeline. And on that pipeline, I think we have very high confidence in it. These are ultimately customer workloads that they have people building that are reaching production that then go drive consumption in the quarters ahead. And so it's a little bit more than just an entry in sales force. And that's why we feel that it's a very promising stat and why we track it very religiously quarter-to-quarter. So I think these are really solid improvements. I've been very impressed by the execution in the go-to-market team over the last few quarters to get this in place and do it quickly. And I think that gives us a lot more ability to help drive these consumption workloads ourselves, right, really land in the right use cases, make sure that they're using our complete product, the full DSP in the best way possible and make sure that, that gets out to production without snags and reaches its full potential. So yes, I think very promising in what we're seeing.

Great. And maybe just a quick follow-on for Rohan. RPO and CRPO, both accelerating very nicely. Why or why shouldn't we look to that as a reliable leading indicator for Confluent specifically?

Yes. Great question, Brad. Thank you. You're right. RPO, in general, what I've shared before is when you think about our business for Confluent platform, absolutely. RPO is the single most important leading indicator with respect to the forward-looking organic growth of the business. For Confluent Cloud, it's a tad bit nuanced where over the short term, I think what we've internally focused on is the momentum of new use cases moving into production and which was a check in Q3. So overall, we feel with the short-term drivers. But over the long term, I think, coverage of RPO to revenue to cloud revenue, that has continued to increase through the year. I mean this particular quarter was the fourth consecutive quarter of accelerated RPO that we've delivered. So yes, like from the cloud business perspective, short term is new use cases moving into production and our ability to drive growth in the new business, newer products, and long term is around the RPO. So that's going well. And for Confluent platform, absolutely, it's a leading indicator. So that's how I think about it.

All right. Thanks, Brad. We'll take our next question from Sanjit Singh with Morgan Stanley, followed by JPMorgan.

I guess it's a very simple one, Jay, and it's with multiple sort of vectors that you guys have in play to drive growth, including with all of the sort of rejuvenation activity within the go-to-market organization. When do you think we can see growth start to bottom is the first question.

Yes. Yes. I mean, look, first of all, I think we're very pleased with the results that we brought, the strength in cloud, pleased to be in a position where we're raising guidance for Q4. I think ultimately, the cloud business has been quite strong. When you look at the growth rate for Q4, there is some impact from a particular customer. We kind of talked about that dynamic last quarter. if you normalize for that, you are seeing kind of stability in the overall cloud growth rates. So overall, we feel pretty good about that. And then when we talk about kind of some of these tailwinds some of the GSP offerings, including Flink getting to scale and starting to contribute more sizably. The overall execution within the field team around consumption and the ability to drive use cases, I think, those are positive trends.

When it comes to the growth that you're seeing in the core streaming business, given the big ramp in like things like WarpStream and enterprise, that sort of kind of the cannibalization question. Are you seeing that kind of net accretive impact from the rise of those offerings? Or do you feel like there's any cannibalistic effect on some of the core streaming business?

Yes. Yes. It's a very fair question as we added new offerings that were particularly cost effective. Is this going to be a tailwind or a headwind. I think it's proven to be a substantial tailwind. So we called out in the call that we've seen substantial improvement in overall deal size, which is maybe counterintuitive, but in fact, it's not because customers are leaning in with bigger workloads, bigger migrations that might have been harder or taken longer in the past. And because of the architecture of these offerings, the multi-tenant clusters with enterprise and freight, WarpStream with BYOC, they're very cost effective to run. So they are a tailwind to gross margin. So it's really good on both sides. It's good deal for customers, they're leaning in and going bigger. And it's a good deal for us. It's ultimately more profitable.

All right. Thanks, Sanjit. We'll take our next question from Mark Murphy with JPMorgan, followed by Barclays.

Yes. Great. So Jay, you had mentioned, I think you said more than 40% sequential growth in progressing late-stage pipeline. And it sounds very promising. I'm not sure we have historical context on that metric. Can you speak to what is driving such great traction there. And then what is the normal level of sequential growth you'd see in that late-stage pipeline?

Yes, yes. So yes, it's a great question. We're obviously not trying to turn that into some kind of external metric, but one of the things we set for ourselves as a benchmark of improvements in the field motion around consumption was, hey, get the new use cases, get into the new use cases, get them to production. And so we measure the dollar amount of those use cases. And we've seen that as these use cases hit production, they ramp up, they take traffic, they drive consumption in the quarters ahead. So it's a reasonable indicator to pay attention to in a forward-looking way. So yes you're asking, hey, what's the normal growth quarter over quarter? Well, over time, if you're bringing more dollars of use cases out to production, those are dollars that you're realizing in future quarters. It takes some quarters for different projects to ramp up. So it's not one is to one, but that's roughly how I would think about it. We haven't given kind of the full history of the metric, and that isn't the intention. It really is, I think, being used by us as a benchmark of execution of the field. And we felt that kind of internal metric was one of the best representations of that. We have made a number of adjustments in how folks are working these consumption projects. And I think it really has worked quite effectively.

Okay. And then as a quick follow-up, Jay, how is the early response to the launch of streaming agents on Confluent Cloud? Because I think we would all agree, for sure, agents need access to real-time data. Frankly, they're going to look pretty unintelligent, right, and out of date if they don't have it. But then companies are -- they're so risk-averse, and they're struggling to give -- to get comfort giving agents free rein to all their data, right? It sort of scares them. And you laid out a nice -- very nice architectural vision for that, right, in the webinar, but I'm just wondering how is the customer readiness for that product? And just could you speak to -- I mean, if this takes off, can agents become pretty big in the mix a few years down the road?

Yes, I think that they absolutely can. So there's a few opportunities around AI for Confluent. One is around making the agents real time. One is about the provisioning of real-time data sets. Both of those are actually substantial, and you can do them both together or you can do them separately. And for those who follow us closely, we -- I mentioned in the prepared remarks that we're here in New Orleans for our conference Current, and that's in a few days. So we'll have some announcements in this space that I think we'll fill out the picture a bit more. But already, the streaming agents have caught on, we talked about one of the customer use cases in the call earlier. And it makes a lot of sense. This is a really easy way that you can run the agent on the kind of historical data, kind of benchmark it, be able to play with it almost in a batch model, but then have it translate into production and run in real time against the data that's there. It makes that kind of development much easier. And I think this is going to be a critical part of the stack. One of the things, I think, software teams are realizing is that this kind of agent development is actually a bit different from traditional software. You have to do it with the data. Traditional software, you can kind of write some program, run some unit tests against it with fake data. If that all passes, it works, you're good to go, your program is good. But these AI systems are not that way. You can build some support agent and say, oh, this answers support questions really effectively. But if you haven't tried it with the actual customer data on actual customer questions, if you're not really developing that way, you're not doing anything. And so the need is to be able to work iteratively with data, but then also launch something that will run in real time in production and be able to keep those 2 in sync as the team moves. And so I think we have really foundational capabilities. Like in many ways, that is about what streaming is, which is this ability to take some of the ideas that we had off-line with batch data processing, be able to translate them into continuous processing. And so I think it's a huge opportunity for us. In many ways, it's an acceleration of what we were doing for customers anyway. Even if the intelligence was just smart rules in a production application that was driving personalization or customization or relevance, we're already doing lots of that. And I think the AI opportunity is, in many ways, a huge generalization of that of allowing not just hard rules, but broad capabilities to access the same kind of data to make data-driven decisions, take smart actions. So hopefully, that's helpful. And stay tuned for the next couple of days, we'll have a few more announcements. It's hard to always figure out the timing of these things. But since that's 2 days later, we don't get to talk about all the new products until then.

All right. Thanks, Mark. We'll take our next question from Raimo Lenschow with Barclays, followed by Wells Fargo.

Perfect. Can't wait for conference then. The -- 2 quick questions, one for Jay, one for Rohan. Jay, Flink, you gave us some extra data points. At Flink, we've been waiting for -- while I don't want to call it an inflection point, but like the uptick here. What do you see there, how customers are using it and what you're seeing in the pipeline? Does that kind of increase your optimism like talk a little bit about how that kind of translates into the business going forward? And then, Rohan, one for you. You've raised the subscription by more than the beat in Q3. Obviously, that's a good sign for Q4. What drove that? Was that kind of the one AI customer maybe doing a little bit more with you? Is that overall business doing a little bit better? Can you speak to that, sort of what gave you the confidence there?

Yes. I'll start with Flink that we're hugely excited. So I do think externally, this was a little bit of an unusual product development cycle because we changed our stream processing strategy and bought a Flink company, but it wasn't a Flink product. It was just the team that had built the open source. So then we were effectively starting product development with an announcement about Flink. So then we had to build the product. And I think the team has done an amazing job of that to really build a modern data -- serverless data processing layer, but do it in a way that supports high availability, real-time processing. It's a big undertaking. I think the growth of that since it's kind of reached GA and kind of got into the critical enterprise features over the last year has been spectacular. And that's absolutely as much as we could ask out of a kind of first year of selling for the product, and that trajectory remains very strong as we look ahead. So yes, we are -- I think as we've communicated as we started this effort, we think that potential for that offering over time is huge. The market for data processing is really big. There's all this stuff in these old batch jobs that needs to move into real time. And now I think we're starting to realize that opportunity. And it's an interesting intersection with the AI question as well because one of the things that actually aids these conversions is AI. So if you're converting these batch quarries to streaming queries, we have a set of capabilities to just help customers do this, just goes through and makes the little minor adjustments. I mean, largely, it's very similar. These streaming queries or SQL or similar language to the batch stuff, but of course, getting all the nuances, right? And so that's been one of the accelerants that's helped customers that are trying to go big with a lot of real-time jobs all at once, help them move faster. So yes, long story short, we're very excited about it.

And Raimo, before you answer the question, I'll just add a quick point to what Jay said. From my lens, when some of these new products are ramping, I think, there are 2 things that I'd like to focus on, the breadth of adoption and the depth of adoption. For Flink, specifically, when you look at the breadth of option, we have over 1,000 paying customers for Flink. And on the depth side, we have about 12 customers spending over $100,000 in ARR and 4 customers spending $1 million in ARR. So that's actually a good position to be in and on the heels of 3 quarters or 9 months of very solid growth that we've seen. So just to add to what Jay said, we're excited about what lies ahead on that side of the business. So coming back to your question on subscription guidance for Q4. Yes, we are pleased to raise our Q4 subscription guide, and that's mostly coming from the Confluent Cloud side of the world. So if I take a step back and then analyze the Q3 performance, I'll call out 3 things. The first one is something that Jay called out in his prepared remarks. That's just the momentum of new use cases moving into production. And we saw 2 consecutive quarters of acceleration over there, so which is good. The second area is around -- we are seeing more normalized levels of optimization. I would actually put it in the category of healthy levels of optimization. So that's number two. And the third is continued strength in Flink and the cloud side of Flink. So these are some of the drivers and the momentum builders in Q3, and that's giving us confidence with respect to our Q4 cloud guidance. And I'll leave you with one more big picture thought that I touched on my first response that is these are short-term visibility drivers for the cloud business. When I take a step back and look at the long term, the RPO to cloud revenue coverage through the year has continued to increase and improve. And that's less of a Q4 visibility, but more of a slight long-term visibility, we feel good with that increasing coverage as well.

All right. Thanks, Raimo. We'll take our next question from Ryan MacWilliams with Wells Fargo, followed by Piper.

Jay, as enterprises continue to move from testing to production with AI use cases, are there any AI use cases that come to mind that involve Confluent that could be more likely in production in the near term, like a customer service use case or an IoT use case.

Yes. Yes. We're seeing -- these tend to be quite broad, right? So there's similar patterns around customer support. There's patterns around anomalies and investigations. Many businesses have some operational side kind of looking for the bad thing and then diving into the bad thing that cuts across businesses that might be doing IoT, manufacturing, different production processes, but also things like retail. But even businesses, financial services, insurance, companies you might think of this being more risk averse. I think have very active projects in this area. And so I think for all of these, it's about whether they can really complete that connectivity and make it into production with these systems. We think that a big part of that is about data flow, data quality, the ability to actually iterate and test and get from something that kind of 99% works to something that 99.99% works. So it sounds like a small difference, but we operate already in a business where, operationally, the difference between 99% and 99.99% is actually a really big deal for our customers. And so you can totally see why on the quality side for any of these things. It's hard to get that last bit done. And I think why we think we're well positioned for it.

I understand it as well. I got 99 things right and 1 thing wrong, you remember which one. And then for Rohan, you mentioned last quarter that a large AI native company was moving to self-hosted after signing a self-managed deal in the third quarter. Any commentary on how much that large customer contributed in the third quarter? And as that large customer spend drops off from the cloud next quarter, could the self-managed portion step-up, contribute further? Just any commentary on the mechanics of that large customer deal could help?

Yes, yes. Ryan, a few data points that I'll share. First, and reiterate what I said in the Q3 call, and what we said in the Q3 call was this large customer basically made this move from Confluent cloud to on-prem. And as a result of this dynamic, their spend towards Confluent would be significantly reduced. So that's the data point. And what that would do is it would have a low single-digit impact to our Q4 cloud revenue. And Jay called out earlier, when you normalize that impact of the low single digit and you compare our Q4 guidance versus Q3 actual cloud performance, you'll see somewhat flattish year-over-year growth rate. So that kind of signs of stabilization. And specifically, that large customer, obviously contributed in Q3 from a revenue perspective and the real impact, the low single-digit impact we are going to see from our cloud business is in Q4 and that's incorporated in our guidance for Q4.

All right. Thanks, Ryan. We'll take our next question from Rob Owens with Piper Sandler, followed by William Blair.

Jay, maybe you could elaborate a little bit more on the CSP replacement opportunity? Just how big you think it is? And why do you think this is inflecting over the last couple of quarters?

Yes. Yes, it's quite sizable. We also, of course, are continuing to do very large open source takeouts, and there's quite a lot of the open source. But both for the open source and the CSP offerings I think one of the -- there's really 2 things that I think are making this something customers really want to take action on right away. The first is the TCO of making the change. And that comes out of the fundamentally, the improvements we've made in Kora that enable things like Enterprise Clusters, Freight Clusters. It's just something that's kind of better, faster, cheaper. And I think that's very compelling. Secondly, I think these DSP capabilities have become just a bigger and bigger part of what customers think about when they think about streaming and what they need to do to be set up to use this technology in their organization. And I think that's really quite appealing to customers making the move. So I think those 2 things are the 2 biggest needle movers. The biggest enabler, I would say, on our side, is really working on tools around migration, making it easy. I think once you have a bunch of customers that want to do it, well, this is a big live data system migration, we want to make it as easy as pushing a button, that's ongoing work to really make that easier and easier. And as I think we continue that, I think we'll see an even faster transition of these systems, which is great.

Great. And then as a follow-up, Rohan, in your contemplating guidance for the fourth quarter, you mentioned healthy levels of optimization. And I know this has been an issue in the first half of the year. When you -- I'd actually just to parse the question a little bit more on the comments a little bit more, is this healthy levels from prior optimizers or these net new optimizers that aren't to the same extent that you saw before. And so I guess within that question, maybe an update on optimization, is it still relevant as a headwind from the first half. And is this more a balancing active net new or kind of the whole thing in aggregate?

Yes, Rob, when I think about the cloud business or rather how we manage and run the cloud business, they are typically like 3 things that are important to focus on, right? The first is, as you're entering a quarter, you're entering a quarter with a book of business and like for the existing customers, what is the growth that they are showing. And that's where optimizations generally come up. And as we've said, optimization is kind of part and parcel of every cloud business. And we want our customers to fine-tune and kind of use Confluent in a more efficient manner. That's part and parcel and that's something -- that's why I called it healthy levels of optimization. Which compares to prior historical optimizations that we've seen and which is not an outlier. So that was my comment. The second data point around how we kind of look at the business momentum is net new use cases moving into production. And the third is around adoption of new products. So when I talk about our guidance or just the momentum in Cloud business, these 3 kind of all go hand in hand. And the optimization levels to specifically answer your question, are in the ranges that we've seen historically, that is kind of more normalized and again, healthy and good optimization.

Thank you. We'll take our next question from Jason Ader with William Blair.

Yes, I know we've seen better cloud consumption trends across the vendor landscape, really over the last quarter or so. How much of the better performance that you guys saw in Q3, do you think is due to better sales execution versus overall macro tailwinds, including AI?

Yes, it's a great question. It's -- obviously, it's always hard for us to pull ourselves out of the environment in which we operate in because we only get to run each quarter once, there's no counterfactual where it was a different environment. That said, I do think some of these improvements are kind of very mechanically obviously helping things. And so I do think we've made a set of structural improvements that are paying off. The new products are obviously new products, which are kind of bringing in Flink revenue or Connect revenue or governance revenue that we would not otherwise have had in those customers. So yes, I can't ascribe it between the two. I am aware that there was kind of good results in some other providers. But we do feel like we've made some pretty important structural improvements in what we're doing.

Okay. And then Rohan, for you. You didn't talk about U.S. Federal at all, but the shutdown here is going into week 4 or something. Did you bake that in? Did you bake in some conservatism to your Q4 outlook, especially on the Confluent platform side from potential weakness in U.S. Federal?

Yes. For -- Jason, that's a great question. I mean, before I go into Q4, our Q3 federal performance, which is generally a big federal quarter, was in line with our expectations. So pretty much in line, no surprises there. And when you look at federal as a percentage of total revenue. I've shared this before. It is in the low single digits for us, which is good and bad, good as it's a big opportunity for us as we look ahead. And so that's great. And from a Q4 perspective, we have a couple of deals that are appropriately baked into our guidance.

All right. Thanks, Jason. We'll take our next question from Mike Cikos with Needham, followed by Wolfe Research. Hey, Mike, we can't hear you. You may be still on mute. All right. Why don't we go to Alex

Can you hear me okay?

Loud and clear.

Perfect. Maybe just first one for Jay. Of the 21 AI native customers that you guys signed over $100,000 or that are using the product for over $100,000, is there a common pattern in how they're using Confluent? Are the AI products built around Kafka or Flink? Or are there use cases similar to what you're seeing with other companies? Because that's a really, really powerful stat. I wanted to see if you could unpack it a little bit.

Yes. So first of all, AI companies or tech companies, so they have a set of usage patterns, they're exactly like every other tech company, which is they use it for a bunch of different stuff, right? But there is a set of use cases that are common in these companies, which are very specific to AI. And that's about the flow of data about the suggestions, recommendations, actions are being taken. So I kind of touched on this briefly in the script, but the big difference in these AI systems is it is not just upfront testing. You need to do this kind of ongoing evaluation, which is really looking at what are the actions that's taken, are they good? How are we going to evaluate that? You have a bunch of different ways of doing that, including just asking humans to judge it, asking the model to judge it. But the flow of that data is really kind of right at the heart of a lot of these systems, and it's a very natural kind of streaming problem. You're going to collect that in real time, it's going to flow out maybe through table flow or other mechanisms into kind of long-term storage. You're going to be able to iterate on that. It's also a very important kind of real-time analytic in terms of how well you're doing for your customers minute-to-minute, as you're out there, if you release if you take in a new model or you make changes to your system, ultimately, are you doing better or worse with your customers? That's kind of the fundamental question. So in many of these systems, that's one of the use cases. And this is not surprising. This is a similar use case we had with more traditional machine learning use cases. I think it's just now translated into the AI era.

Perfect. And then maybe just a quick one for you, Rohan. You gave us a lot of stats that are really encouraging. RPO accelerating and the coverage ratio is improving. You talked about I think, being past kind of a peak negative optimization headwind where it's kind of stabilizing and you're talking about more visibility longer term. And you gave guidance for cloud platform revenue -- sorry, for cloud revenue for Q4, but not guidance, sorry. You gave a modeling point of being around 20%. And as I look at a year ago when -- at least for my model versus the out year, there was about a 2-point delta in that. And so I guess I know we're not guiding to or maybe even in modeling points yet for next year. But as we look at our models and that 20% exit rate, do we -- what kind of step down given some of those dynamics that are maybe headwinds for Q4 that reversed? Should we think about as we look at next year cloud revenue?

Yes, Alex. As we speak, we're kind of dotting the I's and crossing the T's on our fiscal year '26 plan. So I'm not going to be providing guidance either for total revenue or cloud revenue in this call. Having said that, I think, it's important to reiterate some of the data points that I shared around like, I think, you said it late-stage pipeline moving into production, the optimization levels being stabilized, normalized and which I like to call healthy, right? And the Flink driver of business, Flink has been really good. So we expect and coupled with the long-term visibility. So when you think about these and then you couple that with the low single-digit impact that we saw in Q4, which will obviously have an impact over the first half of next year, right? So those are some of the puts and takes. If I were you, I would look at as I think about fiscal year '26. But in our Q4 call, I'll be sharing a lot more color and details around our cloud revenue guidance.

Thank you. We will try Mike Cikos with Needham again, followed by Guggenheim.

Can you guys hear me okay?

Loud and clear.

Sorry about that. And thanks for the second shot here, Shane. I just wanted to come back to Rohan first. On the consumption trends, can you just give us maybe a little bit more granularity on how those month-over-month trends played out in Q3? You obviously outperformed the guide here. But I don't know that we necessarily broke it down to the month-over-month trends the way that we were getting that detail in Q1 and Q2 of this year.

Yes. For our month-over-month trends, obviously, we spoke around the performance drivers for Q3, which were 3, I just laid out. And given these drivers, our month-over-month consumption growth rates improved sequentially. And in general, going forward, I will try to avoid providing that level of detail. But specifically, we brought it up last quarter. So our month-on-month growth improved sequentially, and we were pleased with it.

That's great to hear. And if I could just tack on one more, I know that you guys had the double down initiatives and some of the near-term focuses that we went through last quarter. Jay, maybe for you. But on the DSP specialization team, again, encouraging to hear some of these data points. Has the team been built out at this point? I know last quarter, we were talking about accelerating that buildout. Are the bodies in the seats? And where are we in maturing the playbooks and that team at this point?

Yes. Yes. Yes, that team is built out and in full execution mode.

All right. Thanks, Mike. We will take our next question from Howard Ma with Guggenheim, followed by KeyBanc.

I appreciate all the commentary on the optimization trends. And I get that the Q3 outperformance sets the bar higher heading into Q4. But for -- I guess, one for Rohan. Does the Q4 cloud guide specifically still assume optimizations or consumption trends well below historical trends? And then when you take into consideration the large AI native customer, does it imply that NRR will be -- will decelerate versus the 114%?

Yes. So I'd say a couple of questions, so I'll break it down, Howard, to start off like we always have optimization. And that's all quarters, there is optimization. And that's why I kind of made sure that I commented around like normalized level of optimization that we saw in Q3. So that is hopefully answering the first part of your question. And when you kind of normalize the impact of the one large customer that we called out last quarter, our Q4 guidance is when you look -- compare the year-over-year growth rates, it's roughly flattish to what we saw in Q3. And from a net retention rate perspective, obviously, we are pleased with stabilization of our net retention rates. And when you think about what are the drivers, it's primarily around our stronger consumption growth that we saw, both in core streaming and DSP, both are drivers of stabilization. And from a net retention, again, I'm not going to guide for Q4 or fiscal year '26 for net retention, but I'll leave you with 2 data points. In the short term, net retention can generally fluctuate. But over the long term, some of the opportunities that we are focused on, be it core streaming, DSP, AI partner ecosystem, these are going to be the drivers of net retention rate. And all of these drivers have had positive results in Q3.

Got it. And Rohan, given how important Flink is as a driver now, you gave the disclosure Flink low 8-figure ARR, Flink on cloud up 70% sequentially. I think if you triangulate it, you can get to maybe low single digit, call it, $2 million to $3 million of sequential increase in the cloud side. So is that fair? And should we expect that sort of sequential -- assuming that number is right, increase on the cloud side going forward, maybe as a baseline.

Yes. Again, I'm going to stay away from providing guidance, but we are very pleased with our Flink performance. And from a Flink performance, again, I'll say because it's important to note both the breadth and the depth of our Flink performance is something that we should note. We have a lot of customers, over 1,000 customers using Flink and then we have 12 customers spending over $100,000, 4 customers spending over $1 million. And in Q3, we just reported greater than 70% quarter-over-quarter growth for that business. So we're very pleased with how the Flink business is progressing, and it will be a material contributor to Confluent Cloud in fiscal year '26.

All right. Thanks, Howard. Our last question today will come from Eric Heath with KeyBanc. Eric?

Great. Maybe a lot of good questions have been asked. Maybe if I could just come back to Flink for a minute here, Jay. I am just curious to hear more maybe about some of the easy wins you're seeing with Flink customers. Some of the learnings you are applying to scale that Flink adoption across the customer base. I know we talked a lot about go-to-market and the DSP team, but any color there? And Jay, maybe just lastly, any thoughts or the feedback on how we should think about competition with Databricks structured streaming product that was announced this quarter?

Yes. Yes, happy to talk about both. So yes, there's actually a very broad set of use cases for Flink. If you were trying to bucket them, there's a bucket that's kind of these real-time data pipelines, getting data to some AI application or agent getting data into the analytics ecosystem, Databricks, Snowflake, cloud provider things. And then there's a set of use cases, which are acting on the data, right? Trying to predict fraud, personalize things for customers, do something smart in reaction to an event in the business. Those are the 2 kind of buckets that we see customers using. Both are actually doing quite well. And both are represented in kind of the numbers that we would overall describe. I would say some of the larger customers are customers that are kind of taking existing batch processes and converting them over. So I talked about a number of customers just doing these kind of migrations. That's obviously the most challenging to orchestrate for a new product is to really take something that's been built up over many years and kind of move it over. But we're finding that we're now at a point of maturity where we can start to do that and do it successfully with customers. So that's I think that's an exciting thing. Relative to Databricks, we remain actually very close partners working together on applications for hundreds of joint customers. We're providing a set of real-time data that often flows as their environment. There are some overlaps and capabilities in both what we're doing and what they're doing. I think in practice, we tend to serve different constituencies. We tend to have more kind of real-time operational application systems, software engineers, they would tend to have more data engineers, analytics, data scientists, type user base. But for sure, there's some things that you could do in either product. On the whole though, I think we've been pretty complementary in going to market together. And even though that kind of overlapping feature set may increase, I think that will remain the case. Ultimately, customers have chosen us for that real-time hub of integration for data and many customers have chosen Databricks as the kind of lake destination where all the data goes for historical analysis. And so ultimately, customers want those things to work together, we're happy to serve them together.

Great. This concludes our earnings call today. Thanks again for joining us. Have a nice evening, everyone. Take care.