Okay, well, let's get started. Nice little crowd, Sridhar. I think when you put up a bad quarter, you don't get this crowd, so you must have done something right, I think. Maybe we start there, talk about what went right that's drawing this interest, and maybe I'll keep it high level, but do you feel, Sridhar, that just the environment maybe incrementally improved, or do you feel the upside that you put up was a lot more Snowflake specific? A little bit of both, maybe.
Absolutely, it's a little bit of both. I see this quarter, honestly, as the culmination of investments and extra effort that the team has been putting through the year.
Okay.
As you know, we beat both the first quarters, but we were less unsure about the future. I think part of what laid the stamp on Q3 was not just, you know, it was a really good performance that exceeded expectations, but also, you know, just insights into things like broad-based growth, the strength in the core business, plus new products, you know, objectively and quantitatively working, plus a confidence within me, within the company, of how do we have scaled motions for taking new products. You know, folks have been seeing us launch new things, but it's a different thing to get your customers to adopt, get value from those. I think it's all of these things coming together that added up to this is a meaningful step forward for Snowflake.
Okay. So we're going to dig into some of the more Snowflake specific drivers, but maybe before we do that, just on some of the broader macro tailwinds and headwinds. One of them that's historically been very good for all of the cloud-based data firms, obviously, is when we're in a period where large entities like UBS are migrating more of their data on-prem to cloud. And I'm just curious what you're seeing on that front. Any change in the last quarter or two? I know it's been a more constrained budget environment in the last 18 months. It feels to me as if maybe that migration activity moderated it a little bit. Any sign of life that migration activity might be picking up, Sridhar?
It's been a steady stream for us. I mean, just so you know, a large migration from an on-prem platform onto Snowflake can sometimes take a year. So these things, you know, these things take a while to come. And when it comes to things like contracts and customers renewing with us, there's usually years of effort that have gone into developing that customer, into getting them to realize value from the spend that they have on Snowflake. So by the time, you know, we get to talking to them about, hey, you need to renew your contract, they're already spending at a pretty healthy clip. But more importantly, they're getting value from it. You know, from what we hear from customers, there is not a, oh, I can't do anything because I'm in a budget constrained environment.
It's more of a, as long as we continue to deliver value the way we've been doing together, then things like renewals are a matter of course, which of course, in my mind, is indicative of a positive environment.
Okay. One of the other macro factors, which lately has been a little bit of a headwind to all of the cloud-based firms, has been around spend or cost optimization. You and Jimmy and Mike and the team have called that out over the last 18 months. Where are we on that cycle? Do you finally feel like that it's always an ongoing effort, for sure, but is the bulk of it in the rearview mirror now, Sridhar?
I would say that, but we're also trying to set a new standard for the Snowflake team in terms of how they should think about cost optimization, which is I tell my team, if your customer has to initiate cost optimization and they get a meaningful reduction, you're probably not doing your job right. I don't want to be in the business of cost optimization. I want them to be iterating much more continuously on customers are not wasting their money. Anytime a customer gets to spend $1 million less by putting in two months' worth of effort on the part of two people, that doesn't speak that well of us. It's more of that continuous mentality. And this is why we've also started doing things like releasing features that give visibility into how efficiently are our customers using this.
We expect things like AI to have a bigger impact because the same things that can get at your financial information faster by being able to ask questions can get cost optimization information faster to the people that are looking at Snowflake. I would say, yes, it's not like a dominant theme in our accounts anymore, thank God, but I also, like, again, want to be out of the business of needing to do cost optimizations with our customers because that's indicative of like us not leaning in heavily into optimizing them in the first place.
Okay. So it sounds like the bulk of it's behind you, which is great. Let's talk a little bit about more Snowflake specific drivers. One of the things that struck me is since you arrived, Sridhar, you've made a point to remind investors that one of your key tasks has been to sort of reignite that innovation inside Snowflake, to reignite that product roadmap and take Snowflake well beyond a terrific cloud-optimized data warehouse and be a broader data platform. It felt to me that one of the key takes that investors had in this last quarter was that it felt like this was finally like clicking, where you went out of your way to make the point that these quote new data engineering workloads were already $200 million in annualized revenues.
So I'd love to talk to you a little bit or let you talk to us about why that's beginning to inflect now. Was it just a matter of time? How would you characterize that new workload expansion that seemed to help you a lot in the October quarter?
Again, these are multi-quarter efforts, and sometimes these are multi-year efforts. Like the effort to support Iceberg has been long in the making. It's like a year and a half to two years. The team presciently invested in things like that. But to drive any change from a company perspective, it's one thing for you to try, but the outside world has to be ready for it. There has to be something that triggers the change. One of the, you know, like I used to work at Google, working for Larry, who was like a notorious taskmaster when it came to innovation. One of the things that he drilled into my head is simply aspiring to be someone else that's really good is a terrible aspiration because they're also moving, they're changing, they're evolving, and so I think I would point to two external factors that significantly aid us.
One is the rise of open formats. Boring as that sounds, it's changing the world of data. More and more companies are electing to store their data in formats like Iceberg because they want their data to not be beholden to any company, including Snowflake, mind you. So that's a big change that is sweeping the world of data, and then on the other side, there is AI that is promising to revolutionize everything from how people construct data pipelines to how they get value from unstructured data. Used to be, if you wanted to extract like numbers from your contract, that used to be a custom software engineering project. Now you can take a PDF and stick it into a really good model and it'll spit out numbers for you, so things are getting easier.
And so AI is driving a lot of change, both for data transformation, but also in terms of what it can do, say, to things like business intelligence tools. And so it is really us writing both of these that then manifests itself into products that we can release and products that we can drive the adoption of. We see a new world with open formats permeating with AI sweeping in from the other side as a very different world from the previous world that everybody lives in. And that's where we see us as being much more of an end-to-end provider for data ingestion to data engineering to analytics to machine learning to how it finally gets consumed with AI and BI. That's the opportunity in front of us. You know, you can try all you want.
Sometimes you also need the external winds to blow your way for you to truly take advantage of it. And you saw it manifested in numbers, which again represents a willingness on the part of the Snowflake team to adapt and think about how do you take new products to market in a world where you have to compete for business. And that was a useful muscle for the team to learn as well. So it's a bunch of different things coming together.
You said on the call, Sridhar, that these data transformation workloads that constitute that $200 million quote previously would not have been addressed by Snowflake. And I want to ask what you meant by that because my understanding when you said that is that these were workloads that were probably addressed by other data software firms. And so I'm just curious, is that suggesting maybe that your customer base is beginning to consolidate more onto Snowflake? And to the extent you can be specific, which firms are they consolidating away from?
You know, like a bunch of complicated questions in what you just asked, but the broader point that I would make is we are still in the midst of a secular migration to cloud. We are going from something like $450 million-ish spent annually on cloud to, it's going to be over $2 trillion 10 years from now. So there's lots of business. There's going to be lots of, you know, lots of winners. Snowflake, by choice, elected to play in like in providing analytics on the best data. In the world of data architectures, this is called like the gold layer of data. This was the one that had the most business semantics, and we provided amazing analytics on top of that. We left the other stages to other companies. Some of it is open source, some of it is like EMR from AWS and so on.
What we are positioning ourselves to do is to be much more of a full service provider along all of the different stages that data goes through. Quite a bit of it comes from customers that already work with Snowflake that look at it and go, I can use exact same architecture now on Iceberg data that is sitting in cloud storage. This is a natural extension to what they do. Underlying all of this is the fact that Snowflake is an amazing and loved product that is easy to use, that is low maintenance. We build on all of those strengths and add on these additional capabilities. That's the magic of why we are able to extend out. We're also taking decisive steps. We bought a company called Datavolo. I'm super excited to have that team.
They make 100 plus connectors that let us pull structured and unstructured data into Snowflake, increasing the value that we provide to our customers. It's a set of steps, but with a simple mission that we can be an end-to-end data platform with interoperable formats, with AI sort of powering all of these change cycles.
Okay. Sridhar, you mentioned earlier that there's two important market tailwinds. One is the adoption of open formats, and the other is AI, and they're probably intertwined. But maybe we touch on each one of these because they're both key variables that we need to be monitoring. So on the open format side, what is driving that ultimately? It can't just be that companies woke up yesterday and they don't want to have all their data in a particular vendor. They would have probably said that a year or two ago. So something's acted as a catalyst to this desire to move to a data lake open format. What is that?
It's a lot of things coming together. You know, obviously nobody wants to, no customer, including you, including Snowflake, wants to feel that their data is trapped in applications. And the analogy that I give people is that compute for most of humanity over the last 50 years meant buying boxes. Sometimes they sit in your pocket, but it's a box. And we put our data into it. And whenever you want to move to a new box, it's sort of a pain. And much of early cloud computing was the same way. It's a virtual box you put data in. You can't actually take anything out. And so there is this underlying, you know, current of resentment for it's my data. It should be, I should be able to use it how I want. You know, but it's a little bit like social movements.
It's hard to pinpoint exactly one thing and say that's the one that changed it. Certainly there were early, you know, Parquet was an open format. Lots of people could read it, but it was also, it was not particularly tight. And so it was very inefficient. People struggled with it. There were earlier attempts at it. There's another format called Delta, which started as open source, but became more closed source, and a bunch of people felt burnt by that format. And so Iceberg, in some sense, is like generation three or generation four, but it had a lot of things going for it, which is it was started by consumers of data, not by a data company. These, the early engineers that contributed to this were folks from Apple and Netflix, the original big data company that said we need something that is interchangeable.
Of course, Snowflake leaning into it in a big way helped drive adoption, but it's a lot of these things plus the, you know, ill will around previous formats that has made the industry coalesce around that. I know this like sometimes life is about taking advantage of opportunity. And one of the things that I pushed the Snowflake team very hard on was to open their eyes and see Iceberg as the opportunity rather than the headwind of, oh my God, people are going to move data out from Snowflake into Iceberg, but turn it more into there's 100 times as much data or more sitting outside Snowflake. How do we bring the power of Snowflake to all that data? And that's also a little bit of an aha moment that you can seize on.
So we probably collectively in the investment community, I do anyway, could use some help in understanding whether this architectural shift out of putting data in a data warehouse as good as Snowflake's is to an AWS- or Azure-hosted data lake and querying that data on an open format, whether ultimately that's good or bad for Snowflake because one can come up with a scary scenario, which is that it's an architectural shift. So you're going to need to pivot. And who knows if you'll be able to pivot well. Are you going to have the same query performance on Iceberg as you do in the traditional data warehouse format? So it creates uncertainty. So how do you help us all get through that uncertainty and convince us that this can actually be a net positive for Snowflake, Sridhar?
It's all about execution, and then demonstrating the execution, whether it's in terms of winning logos, in terms of customers sort of betting all in on Snowflake. This is why even the numbers that we released about data engineering and just the effort that we will be making to drive Iceberg adoption across the board. These are the things that, and certainly we've published benchmarks on Iceberg performance compared to FDN performance. You absolutely have to demonstrate it, and you know, I would say like two, three was early proof that we are leaning into it. The thing that I told, you know, some previous analysts that we talked to, certainly that I tell my team is this is a history you can write. You don't have to wait for it to be written.
You know, I've lived through many transformations, like moving Google Search from being a desktop-oriented company to the biggest mobile ads company on the planet. There's both. There's an art, there's a science, and there's also the sheer will involved in migrations like that. We feel good about where we are.
Okay. And today, Sridhar, is it the case that incremental workloads that have arisen because of this move to open formats have been equal to or greater than the runoff of the storage revenues that so far it's actually net neutral or better for Snowflake? Is that the case?
We've talked about it in our earnings. As far as we can see, it's been strictly better. There's no mass migration of data out of Snowflake. And there are lots of intermediate stages. Quite a few of our customers want their data stored in Iceberg, but want us to manage that storage anyway because people don't particularly care for managing S3 buckets. They're scary in their own way. And so there's a lot of nuance. Again, this is a history that we can help write. It doesn't have to be written. And as I said, it is a big opportunity in terms of what we can do along the entirety of the data life cycle, many pieces of which we just abdicated to others. And so it's a stronger, beefier, aggressive, in a positive way, strategy.
Okay. That's good to hear. Let's talk a little bit about that second tailwind outside of open table formats and that's AI. Maybe you could summarize how Snowflake wins in an AI-centric world. How exactly does it lift your growth over time?
Yeah. So one lens that I apply to AI, not the AGI of five years or four years from now, but to the tech that is there today, is that it's like this amazing transformation engine between structured and unstructured data. So for example, you can speak something to a language model and it can extract structure from it. A 17-year-old kid can write a little like voice application that can answer weather queries and say, "Which location did you mean? What time did you mean?" Stuff like that. And so at a core level, we see like AI models, language models for sure, but multimodal models also, that's just making data much more flexible. Let's make it come real. Let's say you had clinician notes sitting in a table. These are just notes that people have written.
Three, four years ago, if you said, "I want to like have a list of symptoms for each patient or I want to have the top symptom for each patient," you'd have to go like start a project. You'd need a bunch of software engineers that would like go work on this for several months, and then they would need to set up servers and run it and monitor it and stuff like that. Something like that is a single SQL statement that you can write. You can turn that into a task and just let it keep running. It's sheer magic. It makes the power of your unstructured data so much, so much more. And this keeps going. You can do the same if you want to extract structured data from PDFs, if you want to extract structured data from images, if you want to do a voice transcription.
At a core level, we see AI as elevating the five million odd people that know how to write SQL into being legit AI scientists without needing to ever look at a GPU or touch it. At a core level. So it just makes like access to data, doing things with data so much more. But we are adding important primitives on top of it, which is like if you think of the prototypical AI applications, all of us know a chatbot. And all of us complain about things like hallucinations. What we really want is a grounded chatbot that speaks the truth to you. We make that trivial to create within Snowflake. It's two commands and out comes a grounded chatbot.
Similarly, if you want access to structured information, you want to know, like I want to know how much did my GVP, Jonathan Billier, how are they doing with respect to their like revenue target over the past 14 days? That's like one English question and out comes a table or a chart that answers that. It makes access to data so much easier, so much faster. And you can use this same tech to make things like the act of writing SQL queries, the act of writing Python for Snowpark that much faster. It just makes data a whole lot more useful and a whole lot faster. And those are the applications that we're seeing. We have over 1,000 production deployments of AI with our customers. And this is why we feel good about the future.
And you know, very soon you'll also be able to do things like AI take actions on your behalf on the basis of things that are happening in the external world or being interactive with it. This is where things like the open data formats feeds in so powerfully because the more data comes and is accessible to Snowflake, the more utility and faster utility you're going to get out of it. So I see it as a massive tailwind for anybody that has the relationships with customers like we do that have the access to data that bring on things like, you know, we have solved permissioning. It's not something that you have to worry about. We've solved things like replication. You don't have to worry about your data center going down. We have a backup one running for you at a different place.
It's the enterprise-grade nature of Snowflake combined with AI that we think massively accelerates the value our customers are going to get from Snowflake.
What does the Anthropic partnership that you announced recently bring to Snowflake?
Anthropic and OpenAI have emerged as the clear leaders when it comes to making the best AI models on the planet. What this partnership lets us do is run their models natively within Snowflake. Something that a lot of our customers worry about is their data leaving their Snowflake security perimeter. They know that we will never let anyone else look at that data. They, you know, are very leery of doing custom integrations where their data is sent out because they always worry about who gets access to it. With Anthropic partnership, we get among the best models in the world running right inside Snowflake, accessible to all our customers with no contract, no additional work needed. It's just there inside Snowflake.
We think of this as a big opportunity both for us and for Anthropic because they get access to the 10,000 plus customers that we have that use Snowflake every day.
Sridhar, what do you think of the debate that might be relevant as well for investors in NVIDIA in the audience around scaling laws? And you can't just throw more compute, more data, general purpose models that they're beginning to flatline. First of all, do you believe that? And secondly, if that's true, does that mean anything for Snowflake? And I guess what I have in mind, the subtext to this question is that if model performance is starting to flatten, is there an opportunity that the next big leap in terms of model performance might actually be enterprises exposing their proprietary data to the models to drive model performance? And if that's true, it would seem that there's a role for Snowflake in that. So what do you think about that? I know there's two or three questions in there.
Yeah. I mean, like look, the good thing about AI is everybody has a strong theory at any given point in time. You just need to wait for three months. There'll be a new theory to explain the future. And so yes, current belief is a worry that model training performance is flatlining. And yes, one can argue that the difference between like GPT-4o or GPT-4o-preview and GPT-4 is small by some benchmarks. But on the other hand, when it comes to the reasoning capabilities of GPT-4o, I find it mind-boggling. You know, math questions, algebra questions that previously, or even gaming questions that previously you would have said are never solvable by these models, all of a sudden are solvable.
And even the slow thinking paradigm, which I think people dismiss it because when you use it interactively and the model takes 25 seconds to answer your question, that's just annoying and you're less impressed by the answer. But on the other hand, if you're talking about one of the data pipelines I'm talking about that's crunching through a million documents every night that doesn't care about whether something has to wait for 25 seconds or not if it takes the right action. I mean, think about that. That is actually profound in being able to go through logic like that. So overall, I would say there's a ton of innovation that is, you know, that's left. When it comes to data, you actually asked a pretty nuanced question, which is, is the answer more enterprise data?
Personally, I don't, I mean, enterprise data is incredibly valuable for making AI useful. At the end of the day, if somebody wants to write a chatbot for like, you know, the performance of my sales team, they better have access to that data for them to be able to create any useful application. In that sense, absolutely it's valuable. But on the other hand, if you look at where has innovation in AI models come from, it is from increasingly sophisticated sources of data, some of which is getting, you know, it's synthetic data. Other people are using experts to literally write questions and answers to complicated questions. So I don't feel from a pure model reasoning capability perspective, enterprise data is going to be a magic answer to that one.
Certainly, having access to enterprise data is going to make the AI a lot more valuable and relevant in an enterprise context. And this is why, you know, both the products that we have, the things that we have announced in terms of Snowflake Intelligence, our agentic platform, plus the relationship with Anthropic is such a big deal for us because it lets us address the potential use cases for AI in the enterprise, building on top of all of these things.
Okay. We've got two more minutes, so I'm just going to stir it up a little bit with the final question.
Sure.
So we're going to have Ali at Databricks here tomorrow. So for years now, Snowflake and Databricks have been coexisting and still do in many large customers in essentially separate swim lanes. It does feel to us, and I think most investors here, that those swim lanes are converging as they step a little bit more into the SQL data warehouse space. They've given a number of 400 plus million. And we talked a little bit earlier about Snowflake moving into those data engineering workloads that used to be their swim lane. So how would you characterize this convergence, Sridhar?
Look, both are ambitious companies that want to do a lot, and it's a big market, and so there is plenty of opportunity. The competition makes both the companies try harder. Hopefully, it's a better deal for all of our customers. You know, but I'm not particularly sure there's so much of a competition. I mean, we have a customer base that's been with us for very long periods of time. I have talked to multiple customers who have asked for a notebook, for example, for data scientists right within Snowflake. We feel very comfortable about our position and our ability to grow. As I said, it's a big market and there's lots of wins to be had.
Okay. Awesome. I learned a ton from this, Sridhar. Thanks for coming to this event. Congratulations on the last result. Jimmy and team, thanks for coming to our event and have a great day, everybody.
Thank you.