The primary incremental takeaway from cloud monitoring and analytics software company Datadog (DDOG) Q3 2026 earnings call is that growth momentum is spreading from a handful of large AI customers to a broader base of enterprise clients.
Revenue growth from non-AI customers accelerated for the fifth consecutive quarter, indicating that AI-driven demand stems not only from model companies expanding themselves, but also from traditional enterprises using AI to write more software, speed up application modernization, and consequently increase their need for continuous integration, testing, deployment, and observability tools.
At the same time, the integration of Datadog's software ecosystem platform is raising the initial contract size of new customers: clients such as a U.S. federal defense agency and major Japanese banks adopted multiple products from the very start of the relationship, while management explicitly noted that these newly signed customers have not yet fully ramped up their usage, and their revenue contribution is still waiting to be unlocked.
The analyst Q&A session further clarified in detail the three factors behind the Q4 guidance: large customers receiving more favorable unit pricing after early renewals, the largest customer being counted in guidance only at its minimum consumption commitment, and seasonal effects from year-end holidays and code freezes.
Early renewals come with higher committed consumption amounts and broader product adoption, with short-term unit price adjustments and long-term customer expansion occurring simultaneously.
Management also made clear that these early renewals are mainly related to customers' own business growth and are not directly connected to enterprises cutting spending to control token budgets.
The largest customer's actual consumption in Q3 was significantly above its minimum commitment, but the company did not extrapolate its future consumption from that; other customers were conservatively discounted based on recent growth trends rather than uniformly counted at minimum commitments. Formal 2027 guidance will be announced at the year-end call.
On the product and business model front, Bits AI has become the fastest-adopted product in the company's history, with billing shifting from charging separately per investigation task to AI credits covering multiple usage scenarios.
Management emphasized that pricing will remain close to inference costs, passing savings from model efficiency gains to customers to support more investigations and broader platform usage.
The self-developed site reliability engineering model Scout, along with reinforcement learning and post-training capabilities, is pushing the company from providing monitoring tools toward autonomous operations. Its commercial value includes both direct revenue from Bits itself and increased adoption of products such as logs, metrics, traces, security, and cost management.
Another key incremental point is that GPU monitoring and training observability remain in the early stages of commercialization. Management explicitly stated that current growth in the infrastructure monitoring business is not directly driven by GPU monetization, and related products have not yet generated widespread revenue; as more enterprises train, customize, and post-train their own models, this could create new demand space.
At the same time, real demand for agent observability is spreading from AI-native companies to traditional enterprises, while AI cost management benefits from enterprises' need to track rapidly growing token spending, usage effectiveness, and return on investment. The company is thus participating in both growth paths: "using AI to manage software" and "managing AI itself."
The following is the full transcript of the Datadog earnings call (AI-assisted translation):
Conference Host
Greetings everyone, and thank you for patiently waiting. Welcome to the Datadog Q3 2026 earnings conference call. Currently, all participants are in listen-only mode. After the speakers' introductions, we will open the floor for questions. To ask a question during the Q&A session, please press star and 11 on your telephone. You will then hear an automated voice prompt informing you that your hand has been raised. To withdraw your question, please press star and 11 again. Please note that today's call is being recorded. Now, I will turn the call over to the first speaker, AJ Lubitsch, Senior Vice President of Strategic Finance and Investor Relations. Please go ahead.
AJ Lubitsch (Senior Vice President of Strategic Finance and Investor Relations)
Thank you, Victor. Good morning everyone, and thank you for joining us to review Datadog's Q3 2026 financial results. We have published these results in the press release issued this morning. Joining me on the call today are Datadog co-founder and CEO Olivier Pomel, and Datadog CFO David Obstler. First, I want to thank Yuka Broderick for her outstanding work leading investor relations over the past five and a half years; she will be moving on to a new role within Datadog. I am pleased to be returning to the investor relations role. I worked with many of you during the company's 2019 IPO. During this call, we will make forward-looking statements, including statements regarding our future financial performance, our Q4 and full-year 2026 outlook and related commentary and assumptions, product capabilities, and our ability to capture market opportunities. Words such as "expect," "believe," "continue," "estimate," "anticipate," "intend," "will," and similar expressions are intended to identify forward-looking statements or other similar future expectations. These statements reflect our views as of today and are subject to various risks and uncertainties that could cause actual results to differ materially. For a discussion of significant risks and other important factors that could affect actual results, please refer to our 10-Q for the quarter ended June 30, 2026. Additional information will be included in our upcoming 10-Q for the fiscal quarter ended September 30, 2026, and other filings with the SEC. This information, as well as a replay of this call, is also available in the Investor Relations section of our website. We will discuss non-GAAP financial measures. A reconciliation of these measures to the most comparable GAAP financial measures is included in the earnings press release, available at investors.datadoghq.com. Now, I will turn the call over to Olivier.
Olivier Pomel (Co-Founder and CEO)
Thank you, AJ, and thank you all for joining us today to review our third quarter results. Let me start with the business drivers for the quarter. I am pleased to report that we delivered strong business performance in Q3. All aspects of the business performed well, including across product lines and a diversified customer base. Revenue growth from non-AI customers accelerated for the fifth consecutive quarter, growing in the high end of the 20% range year over year, up about 1 percentage point from last quarter and above the roughly 20% level from a year ago. Additionally, our AI customer cohort remains a strong growth driver. This cohort continues to grow and become increasingly diversified, both in terms of the number of customers we serve and their scale. Overall, we continue to see healthy demand. Customers across the spectrum, from the most technically forward-looking startups to the world's largest enterprises, are using AI and modern technologies to build. We believe this is driving further adoption of cloud and the Datadog platform. Regarding Q3 financial performance and key metrics, our revenue was $1.2 billion, up 35% year over year, above the high end of our guidance range. We ended Q3 with approximately 33,500 customers, up from about 32,000 a year ago. Customers with annual recurring revenue of $100,000 or more totaled approximately 4,860, up from about 4,060 a year ago. These customers contributed approximately 91% of our annual recurring revenue. We generated $251 million in free cash flow, representing a 21% free cash flow margin. Turning to product adoption, our platform strategy continues to gain market recognition. Currently, 61% of customers use four or more products, up from 54% a year ago; 41% use six or more, up from 31%; and 16% use ten or more, up from 8%. So we are increasing our customer count and creating value across more products through a unified platform. One milestone we are very proud of: annual recurring revenue for our infrastructure monitoring product suite has now surpassed $2 billion, growing in the mid-30% range year over year. Infrastructure monitoring was our first product, brought to market over a decade ago. Today, customers rely on it for full visibility into their environments, including cloud infrastructure, on-premises servers, containers, serverless applications, and GPU clusters. While reaching this milestone is very exciting, we still feel we are just getting started. According to industry analysts, we rank number one in global observability market share, yet our share remains in the teens. Moving to R&D. As AI changes how customers build and run software, we have been investing and advancing in two directions: AI-powering Datadog, and using Datadog to serve AI. First, AI-powering Datadog. These AI products and features make the Datadog platform more useful to customers and increasingly automated. First, our flagship agent Bits AI has had a very strong start. Since launch, Bits has become the fastest-adopted product in our history. In October, Bits Testing became generally available. With a single click, it can discover key user flows in an application and generate tests covering them. Meanwhile, our AI research team is pushing the boundaries of Bits and the platform. Last month, we launched Scout, our self-developed reasoning model specifically optimized for site reliability engineering tasks. Scout was built using our internal distributed reinforcement learning infrastructure and currently powers Bits investigation capabilities. This is exciting for two reasons. First, for alert investigation, it performs on par with or better than foundation models at a fraction of the inference cost. This is valuable because it allows Bits to investigate more alerts, intervene earlier, conduct deeper investigations into incidents, and become customers' always-on first responder. Second, it demonstrates what can be achieved by training models on our own production data, key user workflows, and internal tools. Scout is just the beginning. With the Adaptive ML team that joined us in June, we are accelerating our work in reinforcement learning and model post-training across the board. Over time, we expect this research to drive step-function improvements in platform capabilities, bringing us closer to autonomous operations. Next, using Datadog to serve AI. These products provide end-to-end observability and security for the new AI technology stack. As AI moves from experimentation to production, customers now need to answer three key questions: How is it performing? Is it secure? And what is the return on investment, or is the spend worth it? And because AI does not run in isolation, they want these answers on the same platform where they manage the rest of their stack. On the performance side, we cover the entire AI stack, from compute resources all the way up to agents. GPU monitoring now goes beyond hardware reliability, with new capabilities helping customers maximize cluster utilization, improve training throughput, and reduce per-request inference serving costs. Year to date, the number of GPU devices we monitor has grown eightfold since the start of the year, and some frontier AI lab customers are co-developing these capabilities with us as design partners. Further up the stack, our agent observability product helps teams monitor and evaluate agents in production, including response quality, latency, and token usage. Year to date, the number of trace segments sent to agent observability has grown ninefold since the start of the year. We have also made agent observability available on the Anthropic app marketplace and are a launch partner for OpenAI's new enterprise app marketplace. So customers can now buy our products directly through the frontier AI labs they already work with. On the security side, our AI Guard product discovers custom agents running within an organization and protects them at runtime from prompt injection and data exfiltration threats, leveraging the same real-time data collected to monitor these agents. On the cost side, enterprises everywhere are grappling with rapidly growing token and AI spending. Beyond the three major cloud providers, AI providers are now the largest spending category monitored by customers in our Datadog cost management product, or CCM. Now, through AI cost features in CCM, customers can understand which teams, models, and agents are generating AI spend, and use budgets, forecasts, and anomaly alerts to manage it. Currently, more than one-third of CCM customers are tracking AI costs. The number of customers doing so has grown more than 11 times year over year. Further up the stack, our agent console builds on CCM data to show exactly which agents are actually being used by whom, what tasks they perform, how much they cost, and most importantly, the associated outcomes. As we continue to innovate around AI-powering Datadog and using Datadog to serve AI, independent research firms are taking notice. Forrester named Datadog a leader in its 2026 AIOps platform evaluation, and Gartner in its August 2026 AI vendor race report called Datadog the company others in the observability platform space need to catch up to. We believe this reflects our leadership in observability and our ability to help customers manage the complexity of the AI era. Now turning to sales and marketing, let me walk through a few deals our go-to-market team closed in what was a very strong quarter. First, we signed a seven-figure annualized expansion contract with a leading software company. This customer more than doubled its annual commitment, is putting AI agents into production, and will replace a standalone commercial tool with Datadog's agent observability product to monitor cost and quality. By using Flex Logs, they can also ingest and correlate larger volumes of logs at lower cost. In total, they will consolidate three tools onto Datadog and increase their Datadog product usage to 20 products, including On-Call and Bits AI. Next, we signed a five-year, seven-figure annualized contract with a U.S. federal defense agency. This customer runs thousands of cloud accounts across hundreds of mission owners. As it modernizes, manual tools have left it facing fragmented visibility and security blind spots. They chose Datadog as their enterprise observability and security platform and will start with nine products, including infrastructure monitoring and cloud security. I want to congratulate the public sector team on this milestone customer win. This is a direct result of our investment in the government space and the FedRAMP High certification we obtained earlier this year. We are also actively expanding our government-facing platform to support workloads up to Impact Level 5, or IL5. This is the highest certification level for unclassified data and part of our long-term investment in the U.S. public sector. Next, we signed a seven-figure annualized contract with a leading home healthcare software company. As this customer grew, fragmented tools slowed incident response and hurt customer experience. They will consolidate four tools onto Datadog, replacing their existing logging and security vendors with Flex Logs and our cloud security information and event management product, and rely on our sensitive data scanner to help maintain HIPAA compliance. This customer will start with nine Datadog products, including incident management and Bits AI, to resolve issues faster. Next, we signed a seven-figure annualized expansion contract with a leading online entertainment company. This customer experienced a traffic surge during peak season and scaled effectively with Datadog. Their own AI agents already use a Model Context Protocol server, and they will now add Bits AI. This customer will increase its Datadog product usage to 23 products, including agent observability and cloud cost management. Next, we signed seven-figure annualized contracts with two of Japan's most important banks, which are modernizing their core financial systems. Both banks chose Datadog as their unified standard to gain a consolidated view across cloud and on-premises environments. Both customers are adopting the full-stack platform, covering all three pillars, as well as digital experience, security, software delivery, service management, and Bits AI, moving toward AI-driven operations. Next, we signed a six-figure annualized contract with a top European AI lab. This customer's developers were spending valuable time on slow builds and flaky tests, delaying releases in a highly competitive market. They chose Datadog's continuous integration visibility and test optimization products, replacing a commercial tool, so their engineers could focus on shipping product. Their engineers also use Bits AI to reliably build monitors and custom analytics. Next, we signed a seven-figure annualized incremental expansion contract with a leading AI coding company, bringing the total annualized contract value to eight figures. This customer is growing rapidly and renewed early to keep up with its expanding scale. In this expansion, they will add Flex Logs and our new unlimited cardinality metrics capability to effectively control costs as data volumes grow. They are also one of the largest users of our Model Context Protocol server. This customer will increase its Datadog product usage to 13 products. Finally, we signed an eight-figure annualized expansion contract with the AI division of one of the world's largest technology companies. Just two quarters after initially signing this customer, their annual commitment has already grown to nearly three times the original amount. They are expanding Datadog usage across large-scale AI training workloads. They are also adopting application performance monitoring and Bits AI to further reduce friction in engineering work. Alongside these customer wins, at the other end of the customer size spectrum, we are also increasing our investment in the many AI-native companies emerging every day. Through our investment in the Datadog startup program, we are expanding its reach to a large number of earlier-stage startups and deepening engineering collaboration with those experiencing explosive growth. It is truly an exciting time to work with startups. Now, before I turn the call over to David to review the financials, I want to say a few words about the long-term outlook. Our overall view is unchanged: digital transformation and cloud migration are the long-term growth drivers of our business. Today, AI is also another major growth driver for us, as AI workloads moving into production drive more cloud resource consumption and greater use of data platforms. Today, we are already seeing this across our business: all of the top 15 AI leaders are our customers. More broadly, we now serve over 850 AI customers, up from about 750 last quarter. Across our broader customer base, we are also seeing a significant increase in AI activity. Currently, about one-third of customers use one or more of our AI products, including three-quarters of customers with annual recurring revenue of $100,000 or more. Many of these customers are just beginning to use these products, so there is still significant expansion opportunity ahead. We continue to see rapid growth in related activity, with Model Context Protocol tool calls year to date up 50 times since the start of the year. We could not be more excited about this direction. As AI writes and runs more customer software, customers are shipping more code faster across an ever-expanding portfolio of models, agents, and infrastructure. Of course, the resulting complexity is rapidly outpacing even the best engineering teams' ability to track and manage on their own. This is exactly the problem we are investing to solve. AI-powering Datadog takes more of this work off their hands; using Datadog to serve AI helps them run AI with confidence; and our AI research turns data into models for greater autonomy over time. That is why we believe more than ever that we are ideally positioned to help customers of all sizes and industries, and all types of users, whether human or AI agents, transform, innovate, and create value through AI and cloud adoption. Now, I will turn the call over to our CFO, David.
David Obstler (CFO)
Thank you, Olivier, and good morning everyone. Our Q3 revenue was $1.197 billion, up 35% year over year. We continued to see strong usage growth from existing customers, as well as rapid ramp-up from new customers. Across our broad non-AI customer base, revenue growth accelerated to the high end of the 20% range year over year, up about 1 percentage point from last quarter and above the roughly 20% level from a year ago. We saw strong growth across customer cohorts, spending tiers, and industries. Excluding our largest customer, revenue growth accelerated again. As we communicated in August, our largest customer reduced usage in Q3 but remained above its minimum commitment level. Meanwhile, AI customer growth continued to significantly outpace the rest of the business. This cohort of over 850 customers continues to expand and diversify, including 39 customers paying us more than $1 million per year. We continue to land larger new customers and grow faster, with new customers contributing an increasing share of revenue growth. In Q3, new customers accounted for approximately 30% of our year-over-year revenue increment, up from 25% a year ago. Now on retention metrics. Our trailing twelve-month net revenue retention rate was slightly above 120%, similar to last quarter; churn remained low, with gross retention stable in the mid-to-high 90% range. We believe this metric highlights the platform's importance to customers' critical operations. Now to financial results. Billings for the quarter were $1.16 billion, up 30% year over year. Remaining performance obligations were $3.85 billion, up 38% year over year; current remaining performance obligations grew in the mid-30% range year over year. The duration of remaining performance obligations extended slightly compared to the same period last year. As a reminder, we continue to believe revenue better reflects business trends than billings or remaining performance obligations. Let me review some key P&L results. Unless otherwise noted, all metrics are non-GAAP. We have provided a reconciliation of GAAP to non-GAAP financial data in the earnings press release. Q3 gross profit was $972 million, with a gross margin of 81.2%; compared to 79.6% last quarter and 81.2% a year ago. Q3 operating expenses grew 25% year over year, compared to 26% last quarter and 32% a year ago. Q3 operating income was $334 million, with a 28% operating margin; compared to 23% in both last quarter and the same period last year. Now to the balance sheet and cash flow statement. We ended the quarter with $5.2 billion in cash, cash equivalents, and marketable securities. Operating cash flow for the quarter was $290 million. After capital expenditures and capitalized software, free cash flow was $251 million, representing a 21% quarterly free cash flow margin. The timing of collections and payments can vary; on a trailing twelve-month basis, free cash flow margin was 26%. Now to our outlook for Q4 and fiscal year 2026. First, our overall guidance philosophy remains unchanged. As a reminder, we base our guidance on trends observed over recent months and take a prudent approach to these growth trends. Regarding our largest customer, our Q4 guidance counts revenue at their minimum commitment amount. For Q4, we expect revenue between $1.19 billion and $1.2 billion, representing 25% to 26% year-over-year growth. Excluding our largest customer, the revenue growth implied by our guidance is in the low-to-mid 30% range for Q4. We expect non-GAAP operating income between $280 million and $290 million, corresponding to a 24% operating margin. Q4 includes significant marketing spend, including AWS re:Invent. Based on approximately 379 million weighted average diluted shares, we expect non-GAAP net income per share between $0.69 and $0.71. For fiscal year 2026, we expect revenue between $4.515 billion and $4.525 billion, representing approximately 32% year-over-year growth; we expect non-GAAP operating income between $1.095 billion and $1.105 billion, corresponding to a 24% operating margin. Finally, based on approximately 375 million weighted average diluted shares, we expect non-GAAP net income per share between $2.73 and $2.75. A few additional notes on guidance: we expect net interest and other income of approximately $200 million for fiscal year 2026. We expect cash taxes of approximately $30 million to $40 million, and we will continue to apply a 21% non-GAAP tax rate for 2026 and beyond. Finally, we expect capital expenditures and capitalized software to total 3% to 4% of revenue for fiscal year 2026. Overall, we are pleased with our Q3 execution, delivering a combination of strong growth and strong profitability. We continue to invest to capture the long-term opportunities in cloud and AI. I want to thank Datadog colleagues around the world for their efforts. Now, let us open the floor for questions. Operator, please begin the Q&A session.
The following is the Q&A session of the earnings call (AI-assisted summary and translation):
Conference Host
Thank you. To ask a question, please press star and 11 on your telephone and wait for us to announce your name. To withdraw your question, please press star and 11 again. Please hold while we compile the Q&A roster. Please stand by while we connect the first question. The first question comes from Sanjeet Singh with Morgan Stanley. Your line is open.
Sanjit Singh (Morgan Stanley Analyst)
Alright, thank you for taking the question, and congratulations on the strong Q3 results. David, I wanted to start with the Q4 guidance. If we go back 90 days, compared to then, have there been any changes to your underlying assumptions across different customer cohorts, including the core business and the largest customer? Or does this outlook simply reflect the minimum commitment method you are applying in Q4? I am just trying to understand whether your view of the outlook has changed.
David Obstler (CFO)
Yes, a few things. Thank you, Sanjit. There are a few things to discuss. First, I think we said the largest customer has renewed, with spending above the minimum commitment amount. That was indeed the case in Q3. We said that although their Q3 spending was above the minimum commitment, we would only count the minimum commitment in guidance. As we discussed, excluding the largest customer, our growth accelerated in Q3. So I would say the only change is that in Q3, we completed renewals and contract extensions with some large customers and large AI customers. As is our practice, we basically extended terms, increased commitment amounts, and offered more favorable pricing. This happened with several large customers. This is our normal practice. That is really the only thing that has changed since our last communication. Ali, anything to add?
Olivier Pomel (Co-Founder and CEO)
Yes, we discussed several of them on the call. We mentioned some large customers whose renewals came much earlier than we had originally expected. There was another renewal in October that we did not discuss here. So these are all good news. But they do create a headwind to near-term revenue, and that is a change compared to last quarter. Beyond that, everything else has largely played out as we expected, along with some pleasant surprises, particularly the continued acceleration in non-AI customer growth, which overall has been very good.
Sanjit Singh (Morgan Stanley Analyst)
Understood. If we consider growth excluding AI-native customers, that growth rate has been accelerating for several consecutive quarters. For this core non-AI-native business, your AI product portfolio clearly has many favorable factors. It appears that the cohort of customers with annual recurring revenue of $100,000 is adopting more AI products. Do you think this can continue to drive sustained growth? Or after multiple quarters of acceleration, is the growth rate of the non-AI-native customer base leveling off? Thank you.
Olivier Pomel (Co-Founder and CEO)
I mean, look, these customers are adopting AI, consuming more cloud resources, building more software, and changing the way they work. So we expect these customers to experience a long growth phase. As for how the overall growth rate will evolve — will it continue to increase significantly, or remain stable? The current growth rate is already high, and we feel quite good about it. It is hard to say where it will go from here, but we are very confident in the underlying trends supporting this growth.
Conference Host
Alright, thank you. Please hold while we connect the next question. The next question comes from Gabriela Borges with Goldman Sachs. Your line is open.
Gabriela Borges (Goldman Sachs Analyst)
Hi, good morning. Thank you for taking the question. David, I wanted to follow up on the AI-native customer cohort, because I think you have been very transparent in discussing the largest customer versus the rest of that cohort. What I want to ask is, your sequential performance has been somewhat irregular, because over the past twelve months, this cohort delivered very impressive growth, and now the largest customer has renewal-related changes. You also talked about early renewal dynamics. What I want to ask is, what thoughts can you provide on how to model the sequential performance of this business? Is it that in about two more quarters, we will return to a steady state? Or because of early renewals, might we see another step-down in sequential performance every year? Of course, that itself is a great situation, because you are signing these very large contracts, and customers get more favorable pricing precisely because contract amounts are increasing. So maybe you can talk about what positive and negative factors we should consider when assessing the sequential performance of the AI-native customer cohort, and when it will return to some kind of "normal" state. I also know that "normal" itself is hard to define.
David Obstler (CFO)
Yes, that is a good question. Regarding the largest customer, as we said, this customer has renewed. We decided to count only the commitment amount, which is the minimum commitment. So this customer's Q3 spending was significantly above the minimum commitment. Therefore, a very important factor affecting sequential performance is what happens there. We cannot control it — customers cannot spend below the minimum commitment, but they can spend above it by varying amounts. So that is one factor. Basically, if that amount stabilizes and no longer exceeds the minimum commitment, it would reduce the sequential increase in guidance or actual results. So that is the first point.
Olivier Pomel (Co-Founder and CEO)
I want to broaden the perspective a bit. What I mean is, part of this situation is very common: we keep acquiring these customers, they keep growing and renewing; when they renew, they get more favorable terms and also adjust their traffic over time, all while growth continues as the foundation. This is what we have seen since the company was founded. What is new is that some of these companies are expanding at an astonishing rate, to a degree we have never seen before in this industry — actually in any industry — which creates new dynamics. That is why we are somewhat surprised when we see customers renewing early just a few quarters after starting to adopt products. Overall, the number of these customers is increasing. The AI-native customer cohort is expanding. There are more customers in this group, and many more large customers. We named several customer wins — for example, we signed another AI lab this quarter, as mentioned in the prepared remarks. So in the long run, this cohort will continue to expand and represent a much larger proportion of our customer base, so stability will also increase over time. That said, I still do not think this is a very unstable business. I mean, we are only talking about potential differences of a few percentage points between quarter-to-quarter outcomes and our expectations. But you know, ultimately, we believe this is a usage-based model, and our visibility into the medium and long term is higher than into the very short term.
David Obstler (CFO)
I want to remind everyone that, all else being equal, sequential growth from Q3 to Q4 is typically lower, because the late-December holidays and some code freezes affect our usage-based products. Depending on other factors, this effect may show up or may be masked. But Q4 does have this effect, while other quarters do not.
Gabriela Borges (Goldman Sachs Analyst)
That is very helpful. Thank you.
David Obstler (CFO)
Alright, thank you.
Conference Host
Thank you. The next question comes from Raimo Lenschow with Barclays. Your line is open.
Raimo Lenschow (Barclays Analyst)
Thank you, thank you. Can I ask one last question around this? My understanding is that customers renew early to get more favorable pricing, but their usage is clearly growing significantly, which creates a near-term headwind because the same usage now corresponds to a lower price; but then usage continues to grow, and after a few more quarters, we should see normal growth trends. Is that understanding correct? I remember this has happened in the past, I just want to confirm. Then I have a follow-up.
Olivier Pomel (Co-Founder and CEO)
Yes, that is indeed part of it. Typically in these situations, there are other things happening as well, namely that customers try to figure out what they will need. So they look at the services they are using and think, okay, maybe we do not need as much of this part, and maybe we need more of that part. Or in some cases, we tell them that given the way you use this type of data, perhaps you should adopt a different data retention tier. So when we work with customers, we also make some optimizations to their account usage configuration. But all of this is happening against a backdrop of growth. These customers are growing, deploying much larger workloads, and want to expand their usage on our platform by two times, three times, or even ten times over time. That is basically what we are working on with them.
Raimo Lenschow (Barclays Analyst)
Okay, very clear. Then I mainly want to ask about the business excluding AI customers. In the past, we would think of this growth as coming from cloud migration, customers modernizing and doing more business. You mentioned another aspect, which is that because of AI, we can write more code, application modernization increases, and all sorts of things are happening. So, is this changing the way your business develops? That is, in the past, customers typically needed Datadog when they were modernizing, but now the way enterprises work is actually changing, and this is starting to drive non-AI business growth as well? Can you talk about that?
Olivier Pomel (Co-Founder and CEO)
Yes, I would say there is one thing I did not mention on the call, but it has always been there. That is, we are also doing more for customers. We have consolidated many tools that customers used to use. We have more products and can cover a broader scope. As customers use AI to build more things, some of our other products also become more important. For example, customers are developing much more software, so the pressure on their continuous integration, deployment, and testing has increased significantly, so they are also adopting our continuous integration products more. I think we also mentioned several customer wins in this area. So this opens the door for us to promote more products beyond core observability.
Raimo Lenschow (Barclays Analyst)
Yes. Okay, very clear. Thank you. Well done.
Conference Host
Thank you. The next question comes from Samik Chatterjee with JPMorgan. Your line is open.
Samik Chatterjee (JPMorgan Analyst)
Hi. Thank you for taking my question, I have two questions. First, I know you mentioned that the largest customer did spend above its commitment amount in Q3. Can you tell us, relative to Q3 guidance, how much of the overage came from this large customer spending above its commitment, and how much came from the rest of the business? Of course, that depends on how you forecasted this customer in Q3. I also have a brief follow-up.
David Obstler (CFO)
Yes, we do not provide specific numbers, but it is true that if your guidance is basically below their — we said we would take a conservative forecast in Q3, and we said our forecast was conservative relative to their spending level at the time. For Q4, we have made clear that we will base it on the minimum commitment amount. Indeed, if their spending is above that level, then naturally that means — mathematically, yes.
Olivier Pomel (Co-Founder and CEO)
We have not changed our assumptions over time in this regard. We have always kept our original assumptions unchanged. I think one difference we mentioned earlier, a difference compared to last quarter, is that there were some positive trends in overall customer growth, along with some large customers renewing early, including some other AI customers. That is the change you should keep in mind.
Samik Chatterjee (JPMorgan Analyst)
Okay, understood. And a related follow-up. The revenue contribution from new customers has continued to rise steadily. Can you talk about what the average initial contract size is for new customers now, how it has changed, and what behaviors you have observed? Looking ahead to next year, how do you see new customers contributing to overall growth? Thank you.
David Obstler (CFO)
We have not provided guidance on this yet. We have said that over time, we have seen average initial contract sizes increase, because the platform continues to expand and, as our disclosed product adoption metrics show, customers are adopting more parts of the platform. Ali, do you want to —
Olivier Pomel (Co-Founder and CEO)
Yes, I think this is what we are observing. One thing we are seeing is that among newly formed companies and AI-native companies, as discussed earlier, some are starting at a larger scale earlier. However, we are also seeing enterprises scale faster than before, so they can generate more revenue more quickly. But we are also opening new segments and signing large customers from day one. We listed several examples in today's prepared remarks. I remember we mentioned a U.S. government defense agency. We also mentioned Japanese banks, such as signing two of Japan's top three banks in the same quarter. These initial contract sizes are quite substantial. If you compare the business now to five years ago, this is clearly a sizable new business that we may not have had before. By the way, the wins I just mentioned have not even shown up in revenue yet, because they have only just become customers and usage has not yet ramped up. Thank you.
Conference Host
Thank you. The next question comes from Matt Hedberg with RBC Capital Markets. Your line is open.
Matt Hedberg (RBC Capital Markets Analyst)
Alright. Thank you all for taking my question, much appreciated. I wanted to start with training. You mentioned in your prepared remarks an expansion deal with the AI division of a large software company, which I believe had an eight-figure annualized value. It sounds like they are using you for AI training workloads. To us, this seems like a very notable growth opportunity. Can you talk more broadly about what the opportunity is in training and what it could mean for your growth?
Olivier Pomel (Co-Founder and CEO)
I would say this area has been developing in a positive direction. If you look back one or two years, basically no one was doing training except a few frontier AI labs running their own training, and no one else was doing it. Now, we see an order of magnitude more large AI labs or new labs doing training at scale. And beyond these labs, there is a whole category of relatively smaller enterprises starting to do their own training, including us. We have started doing fairly substantial training ourselves and showcased some of those results on this call. So I think there may be some notable developments here: training becomes part of every company's business, and every company customizes models and builds its own models. So we are investing. I think it is still too early to call this a mass market. Currently, we see some notable signs. We also see that we can play a very large role for some very large customers. But I think we still need to invest, still need to develop, and still need to see how this market evolves.
Matt Hedberg (RBC Capital Markets Analyst)
Understood. David, if I may, I wanted to follow up on one more question. Regarding the largest customer, you used the phrase "significantly above its minimum commitment," which I found interesting. I know you are only using the minimum commitment in your Q4 assumptions, but from an overall judgment perspective, considering this customer and its recent usage patterns, could this change your view of how it will use Datadog over the long term? For example, beyond current usage, could new use cases emerge that make this customer's growth opportunity very promising?
David Obstler (CFO)
I think this applies to any customer — Datadog has many different use cases. Specifically, as we have emphasized over many quarters, we cannot control it — we have contracts and renewals — but we do not forecast these situations. However, for our entire customer base, it is indeed the case that there are many use cases within a single enterprise, whether across different departments, end users, or other aspects. Ali, please add.
Conference Host
Alright, thank you. The next question comes from Carl Kirstead with UBS. Your line is open.
Carl Kirstead (UBS Analyst)
Alright, thank you. Olivier, this question may be for you. In your remarks, you mentioned that many customers are dealing with higher token costs and AI spending and are trying to control these costs. I wanted to know, based on your various conversations with customers, given their efforts to control token costs, how is enterprise spending behavior changing? Also, if some of the customers renewing early to get price discounts are not AI-native companies but traditional enterprises, do you think this might be because they are dealing with budget pressure? Thank you very much.
Olivier Pomel (Co-Founder and CEO)
Yes, I think this is a very common issue. We are seeing explosive growth in AI adoption, especially in coding agents. And this was not in the plans. Basically, a year ago, no one had budgeted for the kind of AI spending we see today, so they really had to figure out how it works, which parts they actually need, and what they get from it. So there is strong demand for tools in this area, and that is what we provide. This is a great opportunity for us, and our cloud cost management product has performed well in both business growth and adoption. As for the early renewals we have seen, they are not directly related to this. I think these renewals came from large customers that are either AI labs or, broadly speaking, technology companies. And this actually has little to do with their own token consumption, and more to do with the growth of their own businesses.
Carl Kirstead (UBS Analyst)
Alright, thank you.
Conference Host
The next question comes from Fatima Bulani with Citi. Your line is open.
Joel (Citi Analyst)
Good morning everyone. This is Joel on for Fatima. I wanted to ask two quick questions. First, on margins and operating leverage on operating income, perhaps for David. This quarter's performance was strong, at least relative to our expectations, but we also know you are investing quite heavily in certain areas. So I want to understand the specific drivers of this quarter's performance, and what your updated, or latest, investment approach is. The second question is also brief, so I will ask both together: regarding Bits AI penetration in the customer base, is there any data you can share? Thank you.
David Obstler (CFO)
Do you want to — okay, let me start with margins. Okay, okay. This is very similar to what we have said before. We are actively investing in go-to-market. Whether by geography or end market, we still have uncovered white space, such as the federal government market. We also continue to successfully scale our quota-carrying sales team, and the results are showing. We continue to invest in R&D. In this area, we are investing in both AI and talent. We see many opportunities. I think Ali will talk about products next. Do you want to talk about Bits AI?
Olivier Pomel (Co-Founder and CEO)
Alright. On this point, we mentioned on the call that its adoption has been excellent. If you look at the number of independent customers or users using it, this is the strongest product we have launched in terms of post-launch performance. So we are very pleased with it. Over the past quarter, we also redesigned how it is packaged. We used to charge specifically for investigation tasks. Now, we use an AI credits model covering a range of different use cases, with Bits AI as our agent, and we have seen very good results. So this is very exciting. I would say it is still early, because we are still developing many features for Bits AI: there is much more customers want us to do for them, and there are many opportunities to automate workflows. Still, we are very, very pleased with where things stand.
David Obstler (CFO)
Let me go back and add one point. I think I missed your mention of gross margin. I believe this is very similar to what we have said before, driven by a combination of platform penetration and engineering work. Our gross margin has been around 80%. This quarter it ended at about 81%. So this is normal practice for us: we invest, optimize, and do engineering work. The gross margin we disclose reflects the stability of this metric, which has also been evident in prior quarters.
Conference Host
Thank you. Please hold while we take the next question. The next question comes from Kirk Materne with Evercore ISI. Your line is open.
Kirk Materne (Evercore ISI Analyst)
Alright, thank you very much for taking the question. I wanted to ask David or Ali about these AI-native customers renewing slightly early. Clearly, they are coming back to renew because their spending with you is high and they want better pricing. I am curious, in these recent renewals, what is the attach rate for other products? When they come back to renew early, do you still have time to cross-sell some of your other products to them? What are you seeing? I know this creates some near-term revenue headwind, but I imagine it also creates an opportunity to discuss expanding their use of multiple products. Thank you.
Olivier Pomel (Co-Founder and CEO)
Yes, I mean, if you look at these customer cohorts, especially compared to the customer base we might have had five or six years ago, they adopt far more products. So we see that these customers adopt products and the overall platform much more broadly than five or six years ago. This is indeed happening. Some of it happens at renewal; sometimes it happens naturally as new needs emerge and usage expands to more teams within these companies.
David Obstler (CFO)
But I think the important clarification is that these are indeed renewals, but renewals with higher committed consumption amounts than before, and that is precisely why they receive volume discounts. When we renew with these customers, their committed consumption amounts all increase.
Kirk Materne (Evercore ISI Analyst)
Okay. David, I wanted to ask you separately, these customers entering the renewal phase earlier is, I think, good long term, but it does make quarter-to-quarter predictability a bit weaker. Is that what Ali meant just now? I think it is good long term, but it creates a little more volatility when setting next quarter's guidance. Is that accurate?
David Obstler (CFO)
I would say it creates some variation in the timing of these customers' revenue growth paths. The sequential situation here actually involves three factors. First, this factor. Second, as we mentioned, the largest customer in guidance, and you know we do not forecast what they will do next. Third, we know there are some seasonal factors as well. So that is one factor, and we try to get it right. Of course, as our remaining performance obligations and contract durations extend, sequential performance can indeed become somewhat more variable at the margin.
Olivier Pomel (Co-Founder and CEO)
Again, these are all good news. We operate a usage-based model, and when working with fast-growing customers, these customers themselves are not entirely sure how they will grow in the coming months, which parts of their business will grow, and how usage will change. So we are happy to work with them to ensure we support their growth. That is how it is. The only time we find this tricky is when preparing for these calls. We have to figure out which parts can be communicated externally, which parts we know, and which parts we do not know. But overall, these are all good news: great customers, strong growth, excellent adoption, and powerful trends supporting the entire industry. So we feel very good about it.
David Obstler (CFO)
And this is similar to what you just mentioned, something we have been doing since the company was founded. Our approach is to increase customers' committed consumption, expand platform usage, and extend contract durations. That is what we have always done. It creates some volatility, but it is very similar to what the company has been doing throughout its history. And it is a good thing.
Kirk Materne (Evercore ISI Analyst)
Great, thank you very much.
Conference Host
Thank you. The next question comes from Peter Weed with Bernstein Research. Your line is open.
Peter Weed (Bernstein Research Analyst)
Thank you very much. We touched on this a bit, but it is exciting to hear about the Bits AI Scout launch and strong early adoption. Can you provide some more specific data and remind us how this product is priced? Does it generate revenue directly, or mainly drive adoption or usage of Bits AI? Overall, it would be very exciting to understand how broadly Bits AI is currently being adopted and how it might drive overall growth, perhaps even becoming another major pillar of the business.
Olivier Pomel (Co-Founder and CEO)
Yes, actually it is both. It is built to a large extent on all the data we collect, the way we process that data, and the various products we have. That is the core. But on top of that, we also have a model that specifically charges for AI credits. The model we chose is to keep pricing fairly close to inference cost, so that as model efficiency improves, we can pass savings to customers and also conduct more investigations. We did this because this space is changing too quickly, and we are improving the efficiency of these models very quickly. So when you expect the number of tasks performed to increase by orders of magnitude while prices also fall by orders of magnitude, setting a fixed price for a strategic activity you perform is not very effective. So we have adopted this model, and it has worked very well. So in the long run, we want both to generate revenue directly through Bits as we do now and to drive broader platform adoption, delivering more value to customers. As for which will become the primary monetization method, it is hard to say right now, and we do not particularly care. Again, we charge entirely based on usage, so as long as usage increases and value delivered increases, we are fine.
Conference Host
Thank you. Thank you. The next question comes from Koji Ikeda with Bank of America. Your line is open.
Koji Ikeda (Bank of America Analyst)
Alright, thank you everyone. I wanted to ask a question about infrastructure monitoring. You disclosed that this business has now surpassed $2 billion in annual recurring revenue, growing in the mid-30% range. Compared to earlier this year when you last disclosed this business was $1.5 billion, can you quantify how much the growth rate has changed? How much of that growth comes from AI infrastructure? With improving GPU efficiency, how sustainable is this growth rate? Thank you.
Olivier Pomel (Co-Founder and CEO)
We are seeing acceleration across the overall business, driven not only by AI-native customers but also by other customers. We mentioned this earlier, and the same applies to infrastructure monitoring. Currently, this growth is actually not directly related to GPUs. Based on what we see from customers, GPUs are still mainly used for training. We have products in this area, but we have not yet widely monetized them. So for us, this is a future opportunity. This opportunity is not yet reflected in current growth and current performance figures.
Conference Host
Thank you. The next question comes from Patrick Covo with Scotiabank. Your line is open.
Patrick Covo (Scotiabank Analyst)
Thank you very much for including me. For me, one of the most interesting points in the prepared remarks is that Datadog is training its own frontier models. Olivier, you then said this has not yet reached the mass market, but you are watching how the market develops. I wanted to ask you, since you are at the forefront of these changes, how does this affect Datadog's choice between proprietary models and open-weight models, internally or customer-facing? Then I wanted to ask David, if the shift Olivier described accelerates, what would that mean for Datadog's profitability? Your profitability this quarter already improved considerably. Could this change also potentially help improve net income? Thank you.
Olivier Pomel (Co-Founder and CEO)
What we see internally and at customers is a mix of different models. So I think here we can use "we" broadly to include both customers and ourselves, because the approaches are quite similar. We make heavy use of top models from top labs, and we also use open-weight models, with open-weight model usage growing. We also train our own models and post-train them. I think this may represent the direction of the entire industry. We do this because some capabilities can only be obtained from top models trained by other organizations. At the same time, there are many things for which readily available open-weight models are already good enough and cost-efficient, and you can just run them. Finally, there are things where efficiency can be greatly improved through your own data, your own workflows, and your own post-training. Some of the models we announced use exactly this approach. So I expect this mix to remain roughly the same. As for the impact on margins, ultimately, you can control to some extent where the money is spent. If a capability is very proprietary and you can achieve better cost efficiency, then you absolutely should do it. If not, for example if it is low-differentiation and applies to some scenarios in your business as well as millions of other enterprises, then using an off-the-shelf model may be fine, and you can continue to use it.
Conference Host
Thank you. The next question comes from Alex Zukin with Wolfe Research. Your line is open.
Alex Zukin (Wolfe Research Analyst)
Hello everyone, thank you for the opportunity to ask a question. Ali, let me ask you a brief question first, then follow up with Dave. To what extent are you seeing demand for agent observability from AI-native customers and core customers driving demand for products like your tracing products? Given increased adoption, what has changed in this area over the longer term? Also, can you talk about cybersecurity and your plans in this area?
Olivier Pomel (Co-Founder and CEO)
Yes, we are seeing growing demand for agent observability, and there has also been notable demand from areas like AI. But the most notable thing here is that non-AI-native customers are also increasingly moving forward, and as they go deeper into AI, we see real demand, actual usage, and very exciting sales opportunities for these products. So we believe this is arguably crossing the chasm, which is exciting for this product, although we are not disclosing specific numbers today. On cybersecurity, AI brings many new opportunities. We are very pleased to see that we can get large customers to adopt the entire platform. We mentioned a defense agency and several other win cases, and I believe in those cases we had customers adopt not only the observability platform but also our security platform. So we are confident in product-market fit and pleased that the AI-driven wave of tool refresh and rebuilding is creating favorable conditions for us.
Alex Zukin (Wolfe Research Analyst)
Great. David, maybe I can ask you to clarify next. First, it is great to hear that you completed a renewal with the largest customer. I think we fully understand the approach of guiding Q4 at the minimum commitment amount. What guidance can you give on how to model 2027? Because this is probably a one-year contract. Should we think of it as staying at some level? Or should you exclude it from the model? Also, if you look at how you guide for some other AI-native customers that renewed early, are those also counted at minimum committed consumption? Or slightly above that? I think any help on these two points would be useful.
David Obstler (CFO)
Yes, as a reminder — yes, yes, sorry. We will provide 2027 guidance on the year-end call. I think over the past several quarters we have said that we guide based on contractually specified minimum commitments, and we reiterated that this quarter. We will provide more detail when we give 2027 guidance. As for the rest of the business, our approach, as we have said for many years, is to base it on these customers' growth rates and then apply a discount. That is the method we use across the entire customer base: start with recent organic growth rates and then discount. For customers that renewed early and for the broader customer base, we of course apply this approach.
Olivier Pomel (Co-Founder and CEO)
Thank you everyone. Alright, I think that concludes today's Q&A. Thank you all for listening. I also want to thank every Datadog colleague for helping create an outstanding third quarter. I know we are now in an even better fourth quarter. Everyone is very busy. We will meet — in a few weeks, I will meet many customers at the largest AWS conference, and I hope to share more exciting product announcements with you soon. Thank you very much, and see you next quarter.
Conference Host
Thank you for attending today's conference call. This concludes the call. You may now disconnect. Have a great day.
Comments