Thanks, everybody, for joining us. I've been popping around between these rooms, and this is my favorite room because it's nice and crisp. These guys in the back know that I like the temperature in here, so we're keeping everybody on their toes, and it's very serene. The mural, the water.
Very serene.
Yeah. The only thing we're missing for the fireside is a little fireplace.
We are. That would be helpful.
Next couple, we will see if we can arrange that. All right. I have been busy all day today, and there isâ
Is there anything going on in the world?
I do not know. I have been in this conference room all day, so I do not know. Who knows? Maybe there is another breach.
Maybe an IPO pricing today.
Yeah, I don't know about that. It is certainly a wild world, and you guys are at the center of a lot of this digital transformation, AI exhaust. David Obstler, CFO for Datadog. AJ, there he is, up in front. A familiar face from a lot of you guys from Datadog's first iteration, thanks for joining us. We were here this time last year, and you had just reported what was a transformation quarter in my mind, when you started to see inflectionâ
Yeah.
âin growth. It was 3Q of last year. That's continued through 3Q of this year. From your perspective, there's probably the obvious answer of AI and whatnot, but what's fundamentally changed in your business that's driven such acceleration, broad baseâ
Yeah.
âAI natives, non-AI natives across the board, and I guess confidenceâ
Yeah.
âthat these trends are durable and they can continue for years to come?
Yeah. Good question. I think, as those of you who have known us, when there is evolution on the tech stack and an impetus for replatforming, modernization of applications, Datadog has tended to benefit.
Yep.
I think we are having that right now for a lot of reasons. AI is one of the factors that is causing enterprises to put an emphasis on modernization of their tech stack, modernization of applications. I think everybody knows that they are not going to be able to take advantage of the AI world unless they modernize their applications, put them in the cloud, et cetera. That has caused a very good buying environment. When things get more complex, when things change, when there is a need to observe more things and how they fit together, that has benefited Datadog. That happened with containers, that happened with serverless. It is happening again, and I think that is the backdrop to what, on the infrastructure side, companies like us are seeing as a strong demand cycle in the market.
Yeah. If we double-click on something that was a real theme for me over the last 12 months, and really, you were the one that really kicked it off, was that non-AI native cohortâ
Yeah.
âaccelerated. This last quarter, they accelerated again, high- 20s growth. There's other companies who feel like they're hitting a little bit of AI roadblock.
Right.
FDEs are being deployed to help AI deals cross the finish line. What's unique about your opportunity, and what does that tell you about where enterprises areâ
Yeah.
âin their AI adoption? Again, the non-AI native cohort.
Yeah, good question. I think, as I mentioned, we have a good end market that is leaning in towards replatforming. The other side of it is what Datadog has done and can do. I think Datadog, all credit to our founders, Oli and Alexis, has over the years broadened its platform significantly, modernized our own platform so it can observe these mission-critical workloads, and expanded our go-to-market at the same time. What that's enabled us to do against this good backdrop of demand is to increase our market share and cross-sell and get good platform adoption.
Yeah.
What we've been able to do with investing over the last few years is to be in a position to take advantage of this change in the market and monetize.
Yeah.
I think we're having consolidation opportunities, cross-sell opportunities, and platform adoption opportunities that are consistent with this need to observe more modern workloads.
Yeah. We're a regulated bank.
Yeah.
You're dealing with a lot of non-AI natives out there. When we and the broader non-AI native cohort starts to move, these are like battleships that take time to turn.
Yeah.
Is there any sense from just kind of what the field is saying around what inning are we in? It feels like more the cloud modernization, but AI modernization. I imagine it is still pretty early for these non-AI natives.
Definitely. Yeah.
Is it first inning, second inning of AI adoption within some of these?
I think we are very early. We are starting to see some metrics reflect the adoption, including the use of our Agent Observability and our Bits products. But I think we are very early. I think at the beginning, there was a very large investment in infrastructure and the foundation model companies and GPUs, et cetera. We are just getting to the point where we are seeing enterprises start to mature into managing their own models. That translates into a combination of use of foundation models and open source models, and put in production environments. This is beyond training, this is beyond internal use, LLMs in their client-facing applications. But we are still very early in that.
Yeah.
I think we're getting into the production phase of it, but we have a long way to go.
Yeah. It feels like there's a long way to go.
Yeah.
There's some dirt earlyâ
Yes.
âto what we are seeing here early in this AI revolution within the tech.
Durability, but everyone trying to figure it out.
Yeah.
Trying to figure out how do you create sustainability or how do you have secure applications?
Yep.
How do you manage the cost? How do you manage the models? How do you differentiate yourself with the models? All of that, I think, is being discovered right now in regulatedâ
Yep.
âtraditional enterprises.
Yep. This is probably my least favorite question that I get asked, and I'm sure it's probably your least favorite, because I think we all wish we could just sort of move on from this.
Yeah.
Your biggest customer outperformed in 3Q. You brought them down to the minimums in 4Q. As you've reflected on the quarter and the 4Q guideâ
Yeah.
âfor a lot of us, we're just like, "Let's just pro forma it out." Let's just like , because you look at the AI native acceleration, the non-AI, it's like it's all goodness.
Yeah.
How would you tell investors to think about that? Because to me, on my math and you are not endorsing this, this customer, assuming it is at the minimum, is going to be pretty small next year as a percentage of total revenue. What is the message to investors?
Yeah. I think stepping back, this goes back probably eight quarters now. Oli said that we have a very diversified business, and then we have one larger exposure. We call it theâ
Yeah.
âthe large guy. Everybody knows what it is. And we said that they are using us significantly above the commit, that they are a customer whose behavior is maybe to try to insource. Given that we have gross retention rates that are very highâ
Yeah.
âand in large customers just under very upper 90s, it is a fringe case. We have kind of said that.
Yeah.
Then, it becomes, they renew, but it becomes hard or impossible for us to be able to calibrate the consumption over the commit.
Yep.
When they started to narrow that, not that they've succeeded in doing it completelyâ
Yeah.
âyou can tell from the last quarter, we said, "Let's just take it off the tableâ
Yeah.
âand do that." I agree with you, it's not really that important to the long-term investment thesisâ
Yeah.
âgiven it is sort of a fringe case. Yes, that is why we have been giving the metrics of the company excluding this, growing at 36% in Q2.
Yeah.
Growing higher than that in Q3. That is why we put in this comment of implicit in our guidance, meaning we are being conservative, right, f or being low to mid-30s.
Yeah.
In order to focus, as you have appropriately been saying, on what is going on at the core of the businessâ
Right
âwhich is goodness.
Yeah.
Yeah.
Yeah. It's been this awkward elephant in the room.
Awkward. We encourage , we don't know, so we're trying to enable everybody if they want to de-risk it and take it out.
Yeah.
Who knows? It could be that they never go away, and they âyou don't know.
On that pointâ
Yeah.
âI think one of the things that we've been talking about this year is, you've done a lot on the inferencing side.
Yeah.
But it feels like training optimization, GPU optimizationâ
Yeah.
âwas what this customer was going to do. But just broadly speaking, that's a huge growth opportunity for you guys that really hasn't even impacted numbers yet. Talk about kind of trainingâ
Yeah.
âGPU optimization, model trainingâ
Yeah.
âwhere are we in that S-curve?
First of all, I think if you look at what we've been saying about the AI cohort, and you look at the number of customers that are making a decision not to insourceâ
Yeah.
âbut to use Datadog, that's what was the factor we'll get into in the customers growing very rapidly and increasing their commitment and lengthening their commitment with Datadog. There's just a huge opportunity in monitoring these workloads. They're modern workloads. They don't have legacy infrastructure. They're all in the cloud. So, that's a very, very substantial opportunity. Then on top of that, there's , and that's in production environments. There's incremental opportunities, we don't know how big they'll be, in the sort of pre-production trainingâ
Yep.
âGPU, Agent Observability, MCP calls. All of this are parts of this equation where there's lots of opportunity because the clients, in order to put applications that work and service clients, need to look at all this.
Yeah. It feels like, even from some of your AI natives, this could be a significant growth opportunity and eventually trickle down to the enterprise.
Could be.
The other, and a lot of the big customer was on the sequential 4Q guide. The other element were these early renewals that you were seeing. After our call back, I have got a very clear sense for what is going on. Maybe just for those that are not as familiar with what is a good thing that customers are renewing earlyâ
Yeah.
âbut it is a near-term headwind to growthâ
Right.
âtalk about that dynamic as it relates to 4Q.
Yeah. So to put it in context, just to make sure. With the sequential being flat, two factors really. One is obviously if you are going to guide to a minimumâ
Yeah.
âin the large, and you have had more than the minimum, that is going to be compressive. Now, obviously, if they spend in the same way they did in Q3, there will be a bigger beat, right?
Yeah.
The second thing is that we have always had a model where clients buy credits. They generally have stayed short because these are new workloads. So let us say they bought $2 million. When they are spending more than that, in this case, let us just say $4 million, it is in their interest, if they are committing to Datadog, to get a better price point. We have, just like AWS and the hyperscalers, we have volume-based pricing and term-based pricing. So this happens all the time. This is what our business model is. And generally, the next minute. They are committing. We only do it if they commit more. So they are spending $4 million or $5 million, but at a discount. So the next morning you wake up, the ARR has gone down. It generally has taken about a quarter to make that up.
You can see that that has not been a headwind, but that has been a good thing because of the stability of the net retention. In this case, this is a tremendous endorsement for Datadog and the outsourcing observability by these types of customers, because what they did is they did early renewals. They committed more, and they committed longer, but it creates this headwind. This headwind is, in the scheme of things, very small. If you think about what our raise would have been, our beat was $0.50 a [share].
It is a very small thing, and it is a very good thing, but it creates a little bit of headwind. It probably had happened in retrospect, in a more concentrated way because of this trend of the growth of these customers, mostly AI, but some non-AI, and the fact that they made these economic decisions to commit bigger and longer with Datadog, which is goodness.
Yeah, it is great for long-term growth.
Yeah.
Maybe I just missed this before, but you are basically saying it is about a quarter lag until you start to see some of that up.
That has been the history, right?
Yeah.
That has been the history, right? Generally, if they stay on that growth pattern, which has been the vast majority of cases, they grow past their discount within the next quarter.
The next quarter.
Yeah.
The ones that, if they were late in 3Q, you might start to see a positive benefit late in 4Q, and then more so in 1Q of next year.
That would happen. If you see it in, I think we said it happened in Q3, later in Q3, and then one of them we said happened in Q4. You would see that benefit in terms of they would grow past weighted average.
Yeah.
Some of them grow past right away.
Yeah.
Some of them it takes longer, but weighted average, it takes about that, so you would see that in Q1.
Okay. From a guidance perspective, how do you sort of, I mean, it's probably, you maybe have better telemetry on when customers are getting closer to that. Is that something you sort of factor in?
Yeah. We factor in. We've had a long history of discounting our organic growth rate or our net retention, and that's what is implicit in our guidance, and that's what's caused that's one of the factors that's enabled us to be a reliable beater. This is nothing new. This is something we look at. Yes, we're not always going to get perfect the timing on it, but as we go out more in time, as you go out past the one-quarter guide, the models are able to digest this and risk manage it effectively.
Okay. Yeah, so I guess the more these happen, the more confidence you have in 2027 numbers, especially as they'reây eah. Okay.
We've also gotten better. I think that when we had the bubble burst, I think we've said this to everybody, I think we've developed as a company much more constructively in working with customers. Meaning, we go in the customers who are overusing or using in a different way, or we go in the customers that are using a lot more than their commit, and we work with them to try to get them to be, this is the long-term life value of the company, try to get them and to extend, to take advantage of the price, to use the product correctly. So I think we've risk managed this betterâ
Yeah.
âas we've learned from the bursting of the bubble the last time.
Infrastructure Monitoring crossed $2 billion ARR.
Yeah.
It's growing mid-30sâ
Yeah.
âwhich is remarkableâ
Yeah.
âfor a business of that scale. Talk to us about where we're at, because I mean, that'sâ
Yeah.
âyou guys got, we're sort of known for thatâ
Yeah.
âinitially, yet it's still growing mid-30s.
Yeah.
Like, what's been the catalyst for that business?
Yeah. I think one is the, what I talked about originally, which is the investment in replatforming, and sort of the movement of workloads accelerating.
Okay.
That is the most proximate metric on workloads. The increased complexity, so more information, the broadening of the product line, which includes things like Database Monitoring, Network Monitoring, Cloud Cost Management, GPU Monitoring. All of those things adding up, more workloads, more complexity and need for the metrics around that, broadening of the platform, and adoption have created that persistence of that growth rate.
It's incredible.
Yeah.
I mean, we see other companies on a similar scale where the core is deceleratingâ
Yeah.
âor weak or shrinking, and it certainly doesn't seem to be the case.
Yeah. I think it's also, we sell the platform. So in times when you're going to have different dynamics within this portfolio of products. But if you step back, our customers are buying an observability platform. Sometimes they'll lean more into workloads and CPUs, GPUs. Sometimes they'll lean more into logs or et cetera. We're reporting that, but from the client side, they're using the platform.
Yeah.
Yeah.
One of the most exciting growth opportunities that I don't think it's a tonâ
Yeah.
âyou've factored in a ton in your guidance is Bits AI.
Yeah.
It's now been repackaged fromâ
Yeah.
âper investigation SKU toâ
Right
âmore of an AI credit model. Maybe just talk about where you are seeing adoption of Bits AI, and how do you think about feathering that intoâ
Yeah.
ânumbers as we think into the future?
It is a good question. We do not fully know the answer to that. We have had, I think, we have a third of our customers using one of our AI products, and a majority of the larger customers. It is good in that we have, it is the fastest growing adoptionâ
Yeah.
âwe've had. We have a lot of customers who are using and experimenting. We think there's a lot of value. We get a lot of good feedback, but it hasn't become a very large monetization yet. I think we don't know. We don't know whether this will be something that will be a very large product, or it might be a magnet to bring more uses to the platform and might be monetized through higher market share gains because no one else might have it, or more workloads. It's important. We don't know yet how exactly. I think moving it to the AI credits is a very important move towards reducing the friction and letting it roll, which is what we're good at. We'll have to see and report on that.
I wanted to talk on security. It's been an ongoing focus for you guys. I don't remember the last time you disclosed how big the security?
We did a number of quarters ago. I think we said over $100 million, and I think we said it's been growing at or higher than the rate of growth of the company. We have not disclosed a new number. You can do the math.
Yeah.
It's becoming a meaningful part of the business. I think we said that our competitive advantage so far has tended to be attaching to Cloud SIEMâ
Yep.
âas the first to places where we have a significant log business, since logs are the DNA. There is a market which is in some disruption, but it's a market that still has barriers to exit and is sticky for the incumbents, and we're making progress there. From a small base, we're also seeing, and I think this has a lot to do with what's happening in AI and code securityâ
Yeah.
âbecause code security is becoming more and more important as you have coding being created by humans and by agents.
Yeah.
W e're seeing a very good uptake in that, again, smaller products. So, there's good progress, but there's a lot to be done here in terms of, we talked about channels, go-to-market, branding, all the things that takes to penetrate the market. We're still in the middle of executing on, but there's some good signs.
It certainly is competitive, to say the least. And Dynatraceâ
Yes, it's competitive.
âjust announced a partnership with CrowdStrike, where they're kind of taking logs and SIEM. You guys are, I don't know. Do you have any strategic partners from the cyber side that could be additive?
I think we have invested in our own product and have made substantial progress.
Yeah.
So, I think it's probably an indication that we've already made those investments.
Yeah.
And so, we have been able to get our market share through our product organically.
Yeah. So, I guess, and there's lots of areas, and obviously the cyber vendors are kind of pushing down into observability, and we see what Palo Alto Networks is doing, reports tonight. So, how optimistic are you that your cyber business, your security business could become an even more material driver off of a larger base? And what are some of the logical products that could drive that?
I think if you step back, it's harder than the top line to penetrate each other's business. There have been a lot of companies that have said, "We have observability." They may have observability of some sort, but it really isn't application real-time observability. And I think the same goes true for the other side. So, I think it's a combination of investment in a number of areas, including go-to-market. And then the wild card is AI going to shake all this up? So, we've tended to do better when there's been a very large technological change, which might change the incumbency, and it may change the silos that have been created around security. And we don't know.
There's signs that that's happening, meaning with agents and the pace of this, security will have to be designed more proactively in code development, not caught on the back end with firewalls and other types of diagnostics. So, we'll have to see what happens. I say we're making good progress, but it is challenging to enter other people's markets.
Yeah.
We have been most successful in the DevSecOps world, where people are more, our clients might be more predisposed to what I just spoke about. There are different markets out there, and there are different nuances that have to be overcome both ways.
Yeah. Okay. I am curious, you are running such a large organization that is very AI-native yourself.
Yeah.
Are you guys funding tokens today? Are you trying to keep headcount more flattish? Are you trying to look at maybe enterprise software vendors that you can replace or consolidate? I am just sort of curious, just because you are sort of tip of the spear for a lot of AI.
We are doing it a couple different ways. We are, probably the biggest area is with coding agents, we are at the margins. We are moving more of the expense into tokens versus headcount.
Yep.
We have so much to do, we're not reducing our headcount or anything. But when we look at our headcount plans versus our token and coding, we are moving it and seeing productivity gains. That's one thing. Two is we're doing the same thing when it comes to. We've always been really efficient in G&A support and sales and things like that. We're being challenged and finding ways to trade off tokens for headcount. Then there are a couple other ways that we're using tokens. One, we made the acquisition of Adaptive ML. We, like a number of other companies, are developing our own research lab. That means we're developing our own models. Scout was one model we put out.
Yep.
We think that'll be a competitive advantage. In order to do that, we have to have the research talent and the GPUs. So that would be incremental investment we're funding out of our incremental profitability, like we might do in any other product. Then we're talking about Bits AI. We're monetizing it in both the revenue lineâ
Yep.
âand overall. So there's a number of different ways, so it depends. I think there's four different areas we're consuming tokens, and the way we're looking at it is slightly different in each of those.
Yeah. From a capital allocation perspective, you have $5 billion on the balance sheet and you're generating a lot of cash.
Yeah.
You've never done big M&A.
Right.
Is that sort of how we should think about, from an M&A perspective, still more of the tuck-in variety that's more talent or technology than larger scale deals?
I mean, we've always looked at our product roadmap and talent and done tuck-ins, and you can expect that to continue. We are open to and have looked at some larger M&A. Our bar is pretty high. I think we want to be in a position from a cash perspective to be able to do that if it fits right. Our bar is high, so I wouldn't rule it out. But you look at our track recordâ
Yeah.
âand how economic we've been, and I think some of the other things are, we need to acquire teams that are going to stay with usâ
Yes
âand develop Datadog, and that's where we've been successful. I would say it's the same. We're going to do the tuck-ins, and we are open to, for the right situation, to look at a larger acquisition. Have not done them to date.
Yeah. So, I mean, what is that bar? You said a couple things that I think are intriguing.
Yeah.
You said, one, if it's a larger deal, you have to be confident that the people are going to stay.
Yeah. Mm-hmm.
It probably wouldn't be a revenue buy in that case. It's still more of a larger scale tech purchase.
Probably leaning more that way. We're not trying to consolidate markets by buying weak techâ
Yeah.
âand then bringing it on our platform. I think it could go more towards buying revenues, but I think it's always going to be that we have to have. We're not going to do it unless the team is either from a tech or in some cases it might be go-to-marketâ
Yeah.
âbecause it might be a different channel type of thing. They're going to stay with us, and that involves reinvesting and things like that. Those are barriers to doing acquisitions.
Yeah. Maybe just two more. I just wanted to circle back on I want to kind of end withâ
Yeah.
âa thing Oli said that it is justâ
Okay
âsort of breathtaking.
Yeah.
But when we think about you guys reporting your 4Q, and one of the questions I ask a lot of CFOs is like, it is kind of hard to guide close to the pin because you are always seeing upside.
Yeah.
Is the right way to think about your 4Q guide, like large customer, there's no downside, there could be some upside?
Yeah.
If we take that customer off the table, should we just consider sort of like more of a normalized beat pattern like we've seen?
Yes.
Maybe not like Q3, but that was sort of like there was a one-off element for that.
I think if you want to look at what is sequential, if you want to put that, you can either take the beat from the Q3, from that out, or add it in.
Yeah.
But yeah, of course, we have not changed our methodologyâ
Yeah.
âexcept for this thing we cannot control.
Yes. Okay.
I think that if had it not been for that, you would have seen probably a more normal sequential development.
Yeah. Okay.
Yeah.
To wrap this up, Oli said something.
Yeah.
He said, "Observability is potentially the last software category standing in a world where AI writes and runs most code," which is kind of a crazy thing to think about. Can you unpack how you guys seeâ
Yeah.
âthe world evolving and why he would say something to that effect?
So much more visionary.
Yeah. It was a powerful statement. I am like, "Oh, that is bold."
But what I think it, a nd I always think sort of is, no matter what happens, no matter what creates software, humans, agents, et cetera, I think we are proving that you need to observe and diagnose it. You need to observe and secure it.
Yeah.
Any which way the world goes, you are going to need observability and security as well there in order to make functioning and safe applications.
Yeah.
And yes, it may be more automated. We may go to auto-remediation. We may go to less humans doing it, but we are going to need to do this act and this functionality.
Yeah.
And I think that is what he means. And we have seen that. We have seen as things have gotten more complex, as the world has changed, observability has been there. And in some cases, observability may have come out from the back room to the front room, because I think it is pretty well accepted in order to speed up the pace of introduction of both AI and modern software, you need to observe it and act on it.
Is the internet being rebuilt for this new agentic era? We may be in the very, very early innings of that.
Right.
The best days of Datadog, and I'm sure if you ask Oli, are still in front of you.
We think, we've always said, even without this, that the percentage of modern applications, given the legacy out there, is just small right now and huge. This probably speeds it up.
Yeah.
This is probably an accelerant. Totally agree with thatâ
Yeah.
âand understand what he was saying.
Yeah.
Thank you.
Well, we are out of time.
Yeah, thank you.
AJ, thanks for-
Thank you very much.
Thanks for showing up here. Everybody, thanks for joining us. Best of luck.
Thanks a lot. Thanks.
Take care.
Thank you.