Please welcome Head of Investor Relations at Snowflake, Jimmy Sexton.
Good afternoon, everybody. Thank you for joining us here at Snowflake Summit and our 2024 Investor Day. I hope all of you had the opportunity to attend the opening keynote last night and the platform keynote this morning. Clearly, a lot of exciting things happening here, and we hope that you stick around the rest of the week and can spend time with our customers and partners, hearing directly from them all the exciting things they're doing with the Snowflake platform. We've spoken with a lot of you in recent weeks, and I think it's been clear on a number of questions and themes that we want to address today. I think first, you'll hear from us about a massive and growing market opportunity that we're addressing, the pace of product innovation, and the differentiation with which we're delivering that product through a single, cohesive, unified platform.
And then lastly, we really want to discuss how we're going to invest in this new opportunity through an efficient growth framework. Before I jump into the agenda, I'd like to point everyone's attention to the Safe Harbor. All right, so the agenda. First, you'll hear from our CEO, Sridhar Ramaswamy. He's going to discuss the growth opportunity that I alluded to. Next, you'll hear from our Executive Vice President of Product, Christian Kleinerman. I think we've announced a number of new initiatives over the last few years, and we want to address each of those to help you better understand when those are coming to market and how they'll impact the business.
Then, our Chief Revenue Officer, Chris Degnan, will spend time with Andrew Curry from ExxonMobil, who manages the central data office, to discuss not only why they chose Snowflake, but what the future roadmap looks like at ExxonMobil with Snowflake and our new initiatives. And then lastly, as I mentioned, Michael will touch on not only the growth opportunity within our core business and how we view that going forward, but really how we're balancing investment with these newer initiatives and how that impacts responsible growth within Snowflake. So then we'll invite Christian, Mike, and Sridhar up for Q&A, and we have a couple of mic runners. So looking forward to spending the day with you, and as always, I'm sure we'll spend time after discussing any follow-ups. So with that, I'll invite up our Chief Executive Officer, Sridhar Ramaswamy.
Thank you, Jimmy. Hey, folks. Thank you all for the time. Excited to be here with you. I've roughly spent a quarter as a CEO. Met with over 100 customers, lots of quality time, both inside and outside the building, so to say. I want to start with how positive all of my customer conversations have been, how much confidence they have in us, how much trust they place in us, and the strength of the core business. Absolutely. We'll be talking about the things that we're going to be doing, our product strategy going forward, how we are going to operationalize things, our metrics. There's lots to come, but it's important to understand just the core strength of the company and the platform. Changes take time to come out, but I'm very happy to be at this place in terms of where we start.
I thought before we get into this, into sort of our product strategy and our go-to-market strategy, I'd just spend a couple of minutes with a brief introduction. Some of you might know me, others might not. I've been working on data pretty much all my life. Started with a PhD at Brown University in databases, went to work on query processing at research labs like Bell Labs and Bellcore, worked for, ironically, a marketing analytics company in the early days of the dot-com boom before joining Google as an engineer and building up one of the largest
businesses and teams ever. So I pride myself on being able to handle scale and business while being in tune with massive technology currents.
We went through everything from the 2008 meltdown to the end of desktop as a meaningful contributor to Google growth, the need to really think about a holistic mobile desktop platform, but also things like AI and their impact on advertising. It's really this combination of technology and business driving great outcomes that I love, that I enjoy. During my 15 years at Google, I spent the last five of those years running, among other things, the ads and commerce teams, which were, of course, the monetization engine of Google, everything from
search ads, display ads, YouTube ads, but also teams like payments, shopping, and travel, effectively pretty much Google monetization pre-cloud. Run a team of over 10,000 people. In the way Google was organized, I was tied at the hip to two amazing business leaders.
First, Nikesh Arora, who is now, of course, the CEO at Palo Alto Networks, who was my counterpart on the business side, and then with Philipp Schindler, who is the current Chief Business Officer at Google. It is really in this combination of product and go-to-market and partnerships that we excelled at. These folks who continue to be close friends were also amazing at driving business at scale. So that serves as, in some ways, a blueprint for many of the things that need to happen at Snowflake, because ads was uniquely a consumption
business. While the business teams brought advertisers in with an intent to spend, it was ultimately the product teams that decided how it was going to get spent, how much monetization actually happened.
And the big unlock at Google was when people realized that marketing spend on Google, advertising spend on Google, should really be thought of as a cost of goods sold, as a way to drive business rather than as a separate expense item. And we see this happening very much at Snowflake with our consumption-based model, where we are all about driving business outcomes. And as we talk about applications and us becoming a delivery vehicle for monetizations, I think you will also see us become actual contributors to revenue with some
of the partner companies that are developing on top of Snowflake. But it's really that formative experience at Google for over 15 years that shapes a lot of what I am as a business and technology executive. And now, as I said, my presentation is in two parts: product and go-to-market.
Snowflake's core product strategy is to have a unified, cohesive product. The thing that we are really, really very good at is the data engine that's at the center of everything that we do. It started as a very pure analytic engine, but over time, it sort of evolved to do just a whole lot more. We now support data engineering workloads. You can write Python. We announced notebooks, which are going into public preview today. Obviously, there's a lot of excitement around interoperable formats like Iceberg, which can do a lot.
The core of our strength is the data engine. And around that, I'm going to talk about how AI collaboration, which I think of as really a huge capability of Snowflake, as well as application and distribution, strengthening and building on that core.
We will get more into the stats behind these, but we have close to 10,000 customers, big, small, across the globe. There are folks from Asia, like Canva, for example, amazing customers from Europe, like Sanofi, but of course, many, many other impressive customers in many different sectors in the Americas as well. As I said, the core strength of Snowflake is the data engine, is the central part. What we are doing here, in addition to what we do with analytics, is we are massively expanding the scope of what you can bring Snowflake to bear
with interoperable formats like Iceberg, but also with Polaris, the catalog that we announced and brought to market very, very quickly.
What this really does is it now opens up Snowflake's ability to act on very large data sets that typically have sat outside Snowflake, especially a lot of things like unstructured data, where we have core strengths, but Iceberg and Polaris are going to massively amplify that. While traditionally, in this community, for example, we have looked at Iceberg as a downer of storage moving out, I really see this as a massive opportunity because what I hear from some of our best customers is that there's hundreds, sometimes thousand times more data sitting
out in cloud storage than what has been brought into Snowflake. Many of our capabilities around data engineering, around AI, around what more you can do with the data is going to become available to Snowflake, and strengthening that core is very much an important aspect of what we do.
Christian will cover things like Hybrid Tables in a little bit more detail. It was an incredibly ambitious project. No one has honestly built a database, a data engine that combines analytical as well as transactional capabilities. That's what Hybrid Tables is. It's taken longer than we had hoped, but it's close to shipping. We have production workloads that are running on it. So making that core strong is part one of our data strategy. And collaboration is a big, big deal for us.
And really, you should think of that as us building network effects. There are companies like Fidelity. Mihir was here at last summit talking about this. He's their Chief Data Officer. That essentially mandates that not only is Snowflake the data backplane of the company, but any partner that wants to share data with Snowflake does it via Snowflake data sharing. It is real-time.
It is cross-cloud. It is just transparent. This leads to a web of interconnected relationships between companies. We increasingly see ourselves as powering enterprise communication, enterprise business, enterprise data exchange. The number of what we call stable edges, edges that are between customers that are exchanging information, has gone from something like 24% a quarter to close to a third just in the space of a quarter. We see that as really amplifying both the power of Snowflake and the network effects. In the earnings call, if
you recall, I talk about how we actually started acquiring customers because one of our customers, Pfizer, created a data product to share data to all of their customers, banks, big and small, via Snowflake sharing. We see that happen time and time again, whereby sharing data, we actually get access to many more customers.
So there's these transitive relationships that end up happening. I would say collaboration is very strong. It's a few years in the making and only getting stronger. Data applications is nascent from a business perspective. Now, folks have been building on top of Snowflake for a very long time, whether it is BlackRock or DTCC or other partners like Blue Yonder that platformed on top of Snowflake, in other words, use Snowflake as the data platform underneath.
We are introducing a new kind of applications, which we call a native application, which run in a sandboxed environment. And you might have seen some of the examples that Christian talked about. This makes it very easy for big customers, AT&T in this case, to be able to bring the power of a relational database from this company, RelationalAI, right into their Snowflake instance.
AT&T benefits because they can now use this graph database without going through a complicated procurement security kind of process because, similar to what happens with Apple and the iPhone, we sandbox what these Native Applications can do so that customers can be very confident that they can only get access to certain things. They can exfiltrate data and so on. And our partners love it, the people that create these applications like RelationalAI, because we become a go-to-market vehicle for them.
As I said, it's nascent from a business perspective, but early results are very promising. And I see this as yet another way in which we continue to create customer value and build a strong connection with customers.
These native applications enable classes of applications that you can't even create in a standard SaaS model because a provider can bring a very large, meaningful amount of data, combine it with a proprietary data set from the customer, and be able to act together on it in a way that you can't even do in a SaaS model. In a regular SaaS application, there is no such thing as a customer giving access to 8 TB of data to an application so that you can get something done, while our unique model of being the underlying data store for both the provider as well as the consumer makes things like this possible. And of course, no topic is complete without talking about AI. I'm pretty sure you've heard lots and lots of this, but we really do see this as a foundational capability.
This is why, in fact, we have a foundation model team, but we also have folks that are experts in inference. We are not in the business of competing with the likes of OpenAI and Meta, that are potentially spending billions of dollars. We look at this much more in terms of how can this turbocharge our business? How can this make data engineering, for example, much, much better? That's what I mean when I talk about pervasive AI. I see this as helping everything from migrations from legacy platforms onto Snowflake, but also making it easier to create data engineering pipelines, for example, in a notebook, or to generate better documentation, which Christian showed in the keynote demo today, but also in terms of making things like chatbots super, super easy to create.
There's a little bit of a cottage industry of different things you have to stitch together if you want to create these things. Our power really is in delivering everything here in a simple, tightly integrated way so that applications like being able to talk to a set of documents and have them answer questions, but also things like unlocking data for direct business use. We see this as a massively enabling technology. What's, in my mind, unique is the speed at which we've been able to bring high-quality products that are super, super tightly integrated with everything else so people don't have to worry about governance, for example. That just works out of the box and to create meaningful value.
What's exciting for me is that with applications like Cortex Analyst, remember, this is the talk to your data, ask a question, and underneath, if you're a business user, you ask a business question, and underneath, the model generates the right SQL and gives you back the answer while having a guarantee that it's not hallucinating, that it is high reliability. Those are the kinds of things that we think can drive broad AI adoption, but much more importantly, significantly increase the number of users that directly interact with Snowflake, that get value from
Snowflake. Obviously, in a consumption model, there's a very direct alignment between user value created and consumption on top of Snowflake. We have customers that have over 700 use cases that are being developed. We have several dozen that have hit production.
This i s everything from being able to index hundreds of thousands of pages of PDF manuals so that people can get answers to questions instead of painfully looking through with Command F, like we've all been forced to do for the past 20 years. But we have production applications from customers the likes of Zoom, Airbnb, Kraft Heinz, Canva, Bayer, and many others. And we expect this to be an area of focus, but also a brisk area of us being able to generate value.
And as I said, the power of Snowflake is really that all of these go into a single platform. A cloud service provider that I talked to recently, it's not that hard to guess who, basically has 300 independent business units that sort of act on their own. And what you get from Snowflake is a single product that has these capabilities.
There's not a separate SKU. There is not integration. Things that need to work out of the box, like through-and-through governance, just do indeed work out of the box. And you can get a lot done without having to deploy independent servers, independent services, and so on. And so that's really sort of our aspiration in terms of where do we want to go. We see ourselves very much as an enterprise data backplane where the most interesting data transformations happen, but increasingly, it's the place where insights, where applications,
both ones that you build as a customer, but also applications from partners run. And I think we are well on our way there. And especially at the core, we've been strengthening it much more going on the offense. And this is where things like Iceberg and Polaris come into play. It is this unified platform that is simple, easy.
That's really our strength. It especially stands out when you deal with enterprise customers and partners that clearly understand that hiring armies of software teams is not where they want to be spending their time, is not where they want to be investing their money. That's where Snowflake stands out as a simple product that just works. We see collaboration and applications as driving another wave of growth. We see us being multi-cloud out of the box as a significant advantage, especially for our more sophisticated customers that want things
like disaster recovery to work out of the box. Some customers are required by regulators, for example, to not only be reliant on a single cloud. Setting up Snowflake replication so that another cloud is there as a backup is a job of a few hours.
Most people that are writing complicated services on the cloud generally would not even contemplate the possibility of uprooting and actually having that thing work on a different cloud. That's the value that we bring. And then the final point on this slide is a larger theme that I'm passionate about. And the larger theme is that over time, higher-level abstractions emerge in every ecosystem. Right now, we are at a place where cloud spend is roughly call it $400 billion.
And our TAM within that, in the realm of analytics, is considered to be roughly $150-$160 billion. But over the next 10 years, we expect cloud spend to grow to $2.3 trillion. This is based on publicly available data that we extrapolated a little bit. But a huge amount of cloud spend today goes towards things like buying compute, buying storage, buying network capacity.
I joke to people that I want myself an EC2 Kubernetes cluster, said no CIO. It's like that's not their expertise. Our value prop as a data platform that abstracts the things that should be abstracted. Our position as a platform as a service, I think, places us incredibly well to capture a much larger portion of the cloud spending market. The analogy that I would have is when cars first came, everybody knew how to open it up and repair a carburetor or the first early computers.
I'm sure there are folks here that have actually assembled a computer. Used to be fun. But no such chance with an iPhone or a new Mac. Why? Technology has moved upstream. It's the right level of abstraction. That is why I think we are uniquely positioned to win.
If you're playing the game of here are 200 different services, and that is all you ship, you're leaving the job of integration. You're leaving the job of making things work to the customer. And this is where our foundational model of things are tightly integrated, things work out of the box, will, I think, serve us well in the years to come. And that, I think, is the real strength of Snowflake and a place for us to offer value. And this ecosystem of collaborating enterprises, this ecosystem of partners now writing applications on what is truly a data platform is, I think,
the key ingredient for our long-term success. And in terms of how we go from here, I would actually say Christian's keynote covered a lot of it.
Part of my value prop to the Snowflake team that I would say is unique is I push the pedal very, very hard on innovation. I ran a team that was making $10s of billions, but getting products to market very, very quickly. Having the framework to be able to drive it without causing problems is something that we were incredibly good at. There's no way we would have survived all of the Apple changes for privacy or the shift to mobile or many of the other sort of mini disasters that we had to deal with without that incredible product speed and
agility. And you can already see the difference in the team in terms of the speed at which we have gotten things out, the number of things, for example, not just in AI, but beyond that get to market.
And so there's a framework that I have for not just how you get things done faster, but on the other side, how you make sure that you have the right harness for reliability, where you want to make sure that you never bring customers down. If somebody is writing an application on top of Snowflake, what used to be a two-minute glitch in performance for somebody consuming your analytics is essentially an application going down. And so driving that higher level of reliability, of excellence, while also having product velocity is something that, again,
I've done for 10, 15 years. And I think that is an important aspect of how we are changing how we deliver things. And then the second part is around how do we get our business teams to actually thrive in this new world of d riving value, driving consumption. This part predates me.
I'm just leaning heavily into it. As you folks know, we went from a hybrid model of rewarding our sellers on contract growth and some on consumption to a much more streamlined model of paying sellers, paying our salesforce only on either new logos or on consumption. And there is, again, a whole machinery that you have to put in place to support consumption at scale. I used to have a multi-hundred-person team that was simply focused on algorithmic optimization of how customers spent money.
And those are the techniques that we are bringing back to play in order to drive customer adoption.
But again, I think in terms of just driving that sort of efficiency, how you drive education in a product portfolio that, let's face it, is a little bit more complicated than what it was, these are all things that I've driven at scale before. And I think it is going to be a significant factor in how our close to 4,000-person business team is going to drive value for Snowflake. And I want to leave you with this, which is I really think of Snowflake as becoming the nervous system for enterprises globally.
This is where business data is. This is where business data gets exchanged. And we are part of a very large ecosystem. And by the way, customers absolutely can and will ask me, "Do you aspire to be the only data player?" And the answer to that is absolutely not.
This is a massive market, and there are going to be lots of players. We are proud both of our heritage and excited about where we are going as a team and company. We think we have a really key role to play. In many, many areas, we are way ahead of everybody else. In areas like AI, where we have had a late start, we have shown the capacity to significantly accelerate and get to be world-class, whether it is our search product or the analyst product. I feel very good that these are truly, truly world-class.
So this aspiration is, I think, the right one, both in terms of stating where we are and in terms of where we want to go. Two final points on acquisitions. We are typically opportunistic about acquisitions when teams fill a need.
I don't think of this as the primary way in which Snowflake is going to grow. Truera, for example, was mostly a talent acquisition of a local team that specialized in things like LLM and ML observability and truthfulness. So we will continue to do acquisitions like this. I myself, as some of you might know, came from the Neeva acquisition. I would roughly say we will continue to be opportunistic, but really, the strength of Snowflake is in the core data platform, its expanding scope, and our ability to create sticky and value-add services on top
of it, whether it's powered by data sharing and collaboration or by partners writing applications on top of Snowflake or by the incredible value that AI can bring to the table. We have an incredibly strong team.
You're going to hear from two of them, Christian and Beneva—sorry, Christian and Mike—today. Our founders are super, super involved. They actually sit next to me. I see them every day. Beneva was actually one of the key engineers behind the Polaris announcement. But we have a very strong team beyond these folks as well. We have a very deep bench in terms of talent across the board, whether it is marketing or engineering or product. We recently had a product leader move over to head solutions engineering because we saw the product had
become very technical, and we wanted somebody that was intimately familiar with the product to be over there. So we have options like that, which is incredibly important and good for any company.
With that, I am going to hand over to Christian to talk about product delivery, and we'll be back to answer more questions. Thank you.
Hi. How's everyone doing?
Good. How are you?
Good to see so many familiar faces. We did establish that most of you caught the keynote this morning. Okay. So we have at least a level set on information. As Sridhar said, a big part of the context for us on what we want to share today is how we're accelerating product delivery. We put together this summary of major announcements. Don't try to take this as the most comprehensive list of everything we launched, but the big items and big investments that we've done.
And I don't know if this slide conveys it, but hopefully, for sure, the keynote this morning gave you additional evidence that we are delivering at a very different pace and incredible amount of capability for our customers and for our partners.
Sridhar just covered a big part of the reasons and how he thinks about this and how he's working with us to go and accelerate that product delivery. But here are some of the reasons that contribute to not only what I think you saw this morning, what was in the slide I just showed you, but most importantly, what you can expect from us in the future. Part one or reason number one is we had several foundational infrastructure projects that needed to happen and come together.
Beneva used two concepts that I liked very much this morning. He talked about, "We are data complete. We are compute complete." Data complete, you needed to have semi-structured or structured, semi-structured, unstructured data, and the transactional data. And we'll talk in a second about the update on the Hybrid Tables, as Sridhar just said, but also compute complete.
There was no way we would be a credible platform for applications if a customer could not just say, "I want to host my own React-based front end or my own UI or my own business logic and go." All of that runs through the integration of not only runtimes, but Snowpark Container Services. And these have been projects that you've seen us mention in the past, work through the adoption or rollout phases. And as you got evidence this morning, many of them, the vast majority of them, have gotten now to general availability ready for prime time with
our customers. Reason number two is extensibility of the platform. And extensibility plays two roles. One, when we build these capabilities like Snowpark or Snowpark Container Services or the integration with Streamlit, they're amazing for our customers. They're amazing for our partners, but they're also amazing for ourselves.
It gives us the ability to go and build new capabilities and integrate new areas of the product without having to go and rebuild infrastructure. So these two are important aspects. But the third one, I could have just written here, Sridhar, literally. I think that that's a big part of it. He just shared with all of you a lot of the philosophy. But Sridhar is working with all of us and the different engineering teams, bringing a pragmatic view on things. And I could give you dozens of examples.
We were in a conversation a few weeks ago, and we had a very, very clear signal from customers, "We need this now." And our traditional mindset would have been something like, "Yeah, this will get sorted out when this happens and that happens." And current course and speed, it'll get sorted out. And that's not an incorrect assessment.
But Sridhar said, "Hey, have you all really internalized how important this is? It can drive business. It's slowing down use case deployments, go live dates." And interestingly enough, that mindset got the team to think about it and get creative. And we have a path to do things way sooner than initially we thought. And this is starting to happen. I think you're starting to see the beginning of that, and there's so much more to come. The other example, and Sridhar just mentioned, Cortex didn't exist six, seven months ago, and we just showed customers and
partners building solutions and going live on that type of technology. So definitely, there's an aspect on some foundational work needed to happen. Some extensibility is going to work, is going to help us, but also how we approach solving problems.
This one, chatting with both Katherine and Jimmy, I describe it as the greatest hit of all the questions that we get from you on a regular basis. We've curated some of our most significant product innovations. We've sorted it left to right based on when we expected to have contribution to revenue. That's the bottom row. I'll walk you quickly through each of them because I think this is what you want to know. Is it available? How's the adoption? When can I put it in my model? Is that accurate?
Yes.
Let's start with Snowpark. I think we're largely on the same page here. Generally available now for 18 months. The adoption number is great. You saw it, 50% of our customers have more detail in a few slides. And we've guided to include some amount of Snowpark revenue in this fiscal year. So that's probably the easy one. Cortex AI, there are multiple aspects to it, and I'll cover more detail, but the core functions, the core foundational model hosting is generally available.
There are challenges and questions on models continue to evolve. So how do we keep refreshing our offering? There's challenges on GPU availability, different regions, different countries. But as I said, we have customers and partners building on this. Adoption is starting, again, for a product that did not exist 8 months ago, expecting a large percentage of our customers to have adopted it.
We'll take a little bit of time. By the way, the color coding on the circles, it's at the footnote. Less than 10% of our customers adopted it is yellow. More than 0, less than 10, and above 10 gives you a green. That's sort of the high-level rubric that we're using. But what we're seeing is so encouraging on Cortex AI that we think we will see meaningful contribution this fiscal year. Companies have budgets for AI. Everyone is trying to do something. And from the early results that we have, we're very encouraging. We're not yet guiding on it. Our principle is all
based on past consumption helps us forecast, but we are confident that there will be AI consumption this fiscal year. The next one is unstructured data.
I think that whenever I have meetings with subsets of you, this is one of the most common questions on the, "Well, but Snowflake doesn't have unstructured data, so you guys are going to miss out on the bulk of it." And no, Snowflake has had unstructured data for a couple of years. It does monetize, by the way, very related to Snowpark and AI. But we have a good number of customers using it. I mentioned at the keynote this morning, thousands of customers are storing unstructured data in Snowflake, and we're bringing more capabilities
to help accelerate. Then Iceberg Tables, GA as of this morning, all three clouds. So it's early, but the signal and the interest from customers is obviously palpable. Native apps also went generally available at the beginning of this year.
The piece that is interesting is the integration with Snowpark Container Services, and that we have it for contributing more next fiscal year. Streamlit in Snowflake, this is our hosting of Streamlit independent from the open-source framework. It went generally available in January. So strictly speaking, it's FY2024. It was last month of FY2024. The adoption is quite strong, but here's the pattern that we see. An organization goes, tries Streamlit, they like it, they deliver one or two applications.
And then there's an aha moment on, "Ooh, I can do so many cool things. I can maybe in some instances replace some dashboards or replace some custom application." That's what we're starting to see in some instances. There's an organization in Japan that just told us every new data experience is going to be Streamlit.
So that's great, but there's some time that is needed for the adoption to take place. And then the last one on this list is UniStore. I have to own it in many ways because I was here or we were in Vegas two years ago. I remember saying, "This is the Holy Grail of databases." There were some skeptical questions from many of you, and they said, "Well, how is this going to work? This seems to be very difficult. Nobody has pulled it off." My answer at the time, for those of you that were in the room, was we made enough progress because it had been
going on for some time, and we've seen positive enough results that we are confident in announcing it. Yes, all of that was true. There was also a lot of additional work that we hadn't accounted for.
But most important, we're in the tail end of this. We're in public preview, and we do expect this to be a very important part of our application development stack. Now I'm going to go through these same product capabilities with a little bit more detail on how we think of what's helping, hurting, and how do we go and drive the adoption? Snowpark, I shared this that this one, for the most part, is going quite well. But you can ask, "Okay, what is going to give it a tailwind?" I have no problem saying also that we should have realized sooner the importance
of notebooks to the adoption of Snowpark. In many instances, what we heard is, "Okay, pound for pound, your engine processes data 2x, 3x, 4x, 5x better than Spark.
Pound for pound, we like the ease of use, the hosted model, etc." But the people that own these use cases, my data scientists, my data engineering team, they're addicted to their notebook, and they're not going to do it until you have something like that. That's sort of the reality. The announcement this morning, Snowflake Notebooks going to public preview, is a meaningful tailwind into the conversation we have with customers on this topic. Iceberg Tables, even though I have it in a separate slide on how we think about it, from a pure
Snowpark perspective, we have customers saying, "I have a lot of raw data already sitting in cloud storage. I would be open to leveraging Snowpark for a transformation, but I'm not going to ingest that data into Snowflake only to transform it." This is an additional opportunity.
And then something that we've shared with all of you in earnings calls and other forums, many of the use cases, not all, but many, are migrations from Spark. And we have dozens, if not hundreds, of migrations either completed or in progress. And with all migrations, this is true of Teradata and Oracle and everything else. It takes time. There's an iteration cycle on how do we find gaps, items that we need to develop. But there's also an aspect of tooling. We are looking in the same way that we did SnowConvert for database conversions and
migrations. We are looking at here, how do we help accelerate customers' goal migration? So very optimistic on pretty much every aspect of Snowpark. Then Cortex AI. I assume since most of you have heard it this morning, I'm not going to spend time unpacking the three layers.
But the big message or the most important message is we are delivering. Not just aiming and this is a vision. No, we have delivered on a fairly comprehensive AI stack that is built for the enterprise use cases. And most important, is built the Snowflake way. Ease of use, simplicity, etc. The demo this morning that did make me lose a few pounds on what was going to happen with what this lady was going to do. But that was the purpose. We had a chatbot built live by someone that has never done it.
And that is the aspiration that we have with Cortex in general. We have lots of early feedback. "Hey, I compared it to the solution from one of the cloud providers, and it was simpler.
It was easier." Many other people are building RAG applications, and it's a little bit of tools here and a vector database from someone else and something else. And simplicity, simplicity, simplicity. And we're starting to see the momentum. Let me see if I should cover anything else here. No, the notion of expanding the reach of Snowflake matters a lot to us. I do think that if you look at the potential for Cortex Analyst and Cortex Search, Cortex Analyst, again, is the one that helps talk to structured data via SQL.
Cortex Search is based on talking to text data. They start to go into a much broader boundary for Snowflake. How do I give access to business users, non-technical users? If you squint, I like to be careful on how soon you're at odds with BI. Excuse me.
But many of the use cases that BI are used for is this type of thing. If I have a business user that can go directly and ask a question and get an answer, that's what you would want. Let's talk about unstructured data. There are two parts or two prongs of investment that we're doing. One, how do we make it easier for more data to be available for Snowflake in terms of unstructured? There's connectors, there's ingestion, and partnerships, like what Sridhar showed at the keynote yesterday.
What we're doing with CODA, lots of data being made available to Snowflake. And once you have that, then you have Cortex and everything else. But also is there, how do we support more use cases? How do we make it easier to get value out of data? And that's very important.
Document AI has the ability to generate so much value for customers that that's where the value is going. Sridhar just said that value is pushing up. I ask questions in natural language of a document and extract. Think of your legal departments, how easy or how much easier it would be if they could just ask questions of a contract. Same thing of, I don't know, financial documents. That's what that type of technology does, and it should make it easier. Document AI is one that we're working on, but we have lots of other ideas in how do we keep pushing up
and creating value. Now let's talk about the opportunity on Iceberg.
I know that we've had lots of uncovered conversations on the headwind that Iceberg represents because we said, "Hey, there's some storage and ingestion revenue that may shift from our current trends." But there's so much more data available to us. CTO of one of the large telcos in the U.S. told me, actually, he drew it in a paper. He said, "Here's how much data we have in Snowflake. And we're happy. We close the books with you. Life is good." Here's how much more I would be interested in being able to query and transform with Snowflake,
but I can't because I'm not going to ingest it today. That's where the importance of Iceberg comes in. And this person was very emphatic. By policy, we're a very regulated industry. We cannot use preview features. So GA, that happened today. So that's how we're unlocking an opportunity.
Let me recap a little bit of what Sridhar was talking about, native apps, because it's very material to the opportunity for us. The evolution of collaboration for us is, "Hey, how do we get organizations to collaborate based on apps?" And what is most interesting to me for all of you to think about and reflect on is it has the characteristics of a cycle that has benefits for all parties involved. Think of it as I am an independent software vendor. Today, I built some interesting app.
And my go-to-market process, especially assuming an app that is powered by data, is I need to go and convince customers to give me their data so that they can do something interesting with their data.
The moment that you say, "Give me access to your data," you're talking to legal, you're talking to procurement, governance bodies, and that thing can take one quarter in a company that is small and nimble and eight quarters in many other organizations that folks in the room work at. That's not for a bad reason. It is high stakes. Data governance, data security is very important. For an app developer, for what here is called a provider, the value prop is imagine if you can build an app that can run closer to the data and your engagement with security and legal
teams is a fraction of what it would be otherwise. I can't promise that, "Oh, it's 0 ." No, no. There will always be some amount of scrutiny.
But if we can reduce it and we've started looking at, "Hey, this is what you need to do to achieve SOC 2, and this is what you need to achieve other controls." And we can optimize just by the guarantees we can give, double-digit percentage, something like 30%, 40%. We'll give you formal numbers at some point, but we can accelerate the time to value to the provider. But now think about it from a consumer. Any customer says, "Hey, I like this app." You mentioned RelationalAI.
So okay, I want RelationalAI. And if me as the business owner or sponsor of that deal, now I need to wait the one quarter or the eight quarters to be able to use that app, it's painful.
So the faster time to adoption creates value for both parties, for someone that is building software, but also for the consumers of the software. We benefit from we're the party that enables all of this. Rarely you find those types of cycles where everyone is getting value out of doing the work. There are additional benefits. The provider writes once, we can run it on all three clouds. That is the interest we see on native apps. It is going well. Any of you can go to Snowflake.com, search for listing the marketplace, and you see the apps that are being
created. There's all sorts of interesting value-add use cases. Partners, providers are participating in this because they see this acceleration of value, and consumers are starting to see that. That's why we had three customers at the keynote this morning.
Because it's not just a thesis that, "Hey, this should be good for everyone." No, it is proving out to be good for everyone. Recapping what we had at the keynote this morning, Jean-François from CN Rail, Canadian Rail, was on stage saying, "We get faster value from being able to use this app, Maxa." Maxa does ERP insights, and they were able to, without having to copy out their ERP data, which was already in Snowflake, they can get insights and value add from the data that is in place.
I mentioned container services as the ultimate aspect of being the compute complete for Snowflake. And for the same reason of what I was describing, customers or partners want to be able to run inside Snowflake, want to be able to leverage our security and governance perimeter.
They're very interested in this, and these are all companies that are investing heavily. I went out of my way at the keynote this morning to thank them because some of them started building when this thing was a little bit more than an idea. But now Snowpark Container Services is generally available in AWS, public preview on Azure, and these companies are starting to see the value. We're starting to see early positive signs on consumption. Streamlit, I mentioned a little bit the cycle that we see: tested, like it, small use case, and then rapid
adoption. Maybe at this point, the ideal organization we could aspire to in terms of Streamlit adoption is what Snowflake itself looks like. We went through that cycle.
We went through, "Hey, this is a cool app or experience that we built." These days, the go-to reaction for most folks inside of Snowflake is, "If I need to present something based on data to any user in the company, for legal, for marketing, I go build a Streamlit app because it's the fastest way to deliver value to business users." Same is true for machine learning. There used to be this often repeated statement on, "Oh, how many machine learning models get trained and never get put in production?"
Because delivering them and making them available to business users was very difficult. Streamlit makes this trivially easy. Same thing for AI. We're seeing a large number of AI chatbot-based applications being built on Streamlit. The adoption here from a percentage of users that are using this, it looks quite phenomenal.
I am just caveating it with everything I said on the credits will, on a customer-by-customer basis, follow after that initial exploration. And last but not least in this list is UniStore. I already told you it's a very unique project. It's a very ambitious undertaking, but we're very, very close. Our expectation is to be generally available before the end of this year. And as I shared at the keynote, this is the architecture from Mutual of Omaha, who tried it, loved it, got their legal team and security team comfortable with asking us for permission to go production, and they
are in production. And we have maybe a dozen other use cases where the value was so compelling that customers decided to go and do it. The thing that I like the most is the full value prop of Hybrid Tables of UniStore. I have an app.
It's a call center app. It is writing and reading. The call center app needs to be responsive, so the latency matters a lot. But they're also doing analytics, campaign analysis on the same data. No ETL, nothing like that. Some of the use cases, they're joining data from the data sharing data shared via the marketplace. That is the sweet spot. What they're doing is not rocket science or unique to them or their industry. This is what a lot of people do. That's why we're excited about this.
I want to end exactly where Sridhar started, which is we see many aspects to the technology we're building. These are the elements of the AI data cloud. Customers are already telling us this is compelling. Customers are already leveraging our core engine. Customers are doing data sharing. You heard the stats. The native apps is real.
AI is happening. Applications in general as a platform will continue to evolve, inclusive of AI apps. We're very positive on not only the pace at which we're delivering product, but at the pace at which customers and partners want to leverage the technology that we have. And as Sridhar also said, we, at least in the department organization, are also very focused on ensuring the successful go-to-market and the successful adoption of the different products. With that, I'm going to turn it over to Jimmy. Thank you.
We are just going to take a couple of seconds to get the chairs set up. Christian, you've clearly been busy. Appreciate all the hard work. Now I'd like to introduce our Chief Revenue Officer, Chris Degnan.
I'd like to thank Andrew Curry for taking time out of his busy schedule to spend time with us and helping you understand how ExxonMobil is evaluating Snowflake and kind of their plans for the future. With that, I'd like to invite the both of you up here. Thank you.
Hello. Chris Degnan.
Andrew Curry.
Andrew, so thank you for doing this. Give us a little bit of overview. I'm sure everyone knows who ExxonMobil is, but maybe what you do at ExxonMobil.
Sure. In case you haven't heard of ExxonMobil, a Fortune 5 company, you can look it up. Lots of public data about the company. I run the central data office. So I've been in this role for almost three years. I was actually probably one of eight people in the think tank to help form the central data office and a ctually built out that group.
So a global team with staff in 12 countries.
Awesome. So obviously, as you formed the CDO office and the think tank, you obviously were going on this digital transformation journey. Where is ExxonMobil in this journey?
Yeah, I think we're right in the depths of things. It's probably a great way to put it. In 2001, or sorry, 2011, we really updated our corporate digital strategy. And that was part of the precursors, "Hey, we need a central data office." And a lot of that strategy was saying, "Hey, can ExxonMobil better leverage our corporate scale?" So we used to run a supply chain for many different divisions, and we said, "Hey, could we possibly run one supply chain for the entire corporation?" That's going to drive data needs and data demands and things that have
those types of changes.
So a really in-depth change of how the business operates driven by digital transformation.
Great. So in your digital transformation journey, why did you choose Snowflake? And kind of what was the initial use case and how have you adopted us so far?
Yeah, it actually started fairly small. I think we had actually our chemical division was the first one to start using it as a data platform. And it was probably a period of five to six years ago when I'll say everyone seemed to be building their own data platforms. So projects were kicking off. Multiple platforms were being built. Over time, I think there was a general gravity towards using Snowflake. So suddenly the upstream said, "Hey, what's chemicals doing? I want to use that technology."
And it started kind of taking over. And we eventually did a really strategy plan.
So do we want to make this part of our strategic technology stack? And we did make that decision. And then we've sort of moved to a centralization strategy to say, "I don't think we need 15 data platforms." In fact, that's actually producing more data silos. It's blocking some of the initiatives that we want to do. And so we really started saying, "Hey, can we start collapsing some of these data platforms, bringing our data together and getting that full use, getting that full corporate scale, that capabilities of ExxonMobil size?"
So on your journey in terms of, obviously, there's a ton of use cases that you're using us for. What's an example of a good use case?
Sure. So last year, Exxon generated $55 billion in cash. It's kind of a public data point to go off of.
But we started saying, "Hey, can we optimize our cash flow better?" And so we operate in 60 countries. It's not a simple task to look and predict the cash flow of ExxonMobil. But as you have that centralization strategy, as we brought our data into Snowflake, we said, "Well, I've got all the vendor data. I can predict when cash is going out the door. I have all my customer data. I can predict when cash is coming in. I've got our trading department's data. I've got our financial data.
I can start building this." And that data used to be disparate. It was very hard to pull that data together. Now that it's together, I can actually start doing that. Now we have a cash flow prediction model, which again allows us to leverage that cash in a more effective, efficient manner.
Yeah, that's awesome.
So obviously, a big question that I think a lot of people would ask is, "How do you budget for Snowflake?"
That's a great question. It's an interesting time. In data and analytics space in general, the annual budget cycle is not working the best, I'll be honest. I'll get hauled in front of senior management at some point and say, "Why is this spend up? What's going on?" Not just in Snowflake, in this space in general. We've really had to show, I'll say, two things. Data is an investment. So what is the value? What is the ROI we're getting on that investment? It is a very important conversation we have to have to defend that investment.
I think the other part is, are you doing it efficiently?
I think that's one of the things that's actually helpful with Snowflake is, what is the cost per query? And so we can say, "Yes, my spend is going up, but my cost per query is actually dramatically going down over time." So we're getting more efficient so that saying, "Hey, we can use this tool smartly and efficiently," and it's getting more efficient over time, even though that maybe total spend may be going up.
Great. So in addition to Snowflake, how do you evaluate vendors when you bring them into your stack at ExxonMobil?
Yeah, I mean, it's a tough question. I mentioned we operate in 60 countries, so it's not always a simple answer. So being cloud agnostic is something that is helpful for us.
When you're in that many countries, it's hard to say, "I'm only going to be in one cloud," and things like that. So something that has that agnostic and that can maybe sit on top of different cloud providers was something that is appealing for us. Another key feature, I think, for us is interoperability. We have a data ecosystem. We have other tools and things like that. What works well together? What fits maybe in a modular architecture so I can kind of have a core technology but bring in pieces of technology easily when I do that?
So again, part of that centralization strategy we have is, what is that core technology? What fits? What works well together as a collective ecosystem? That's a very structured approach. We don't identify strategic partners easily. That's a structured process and evaluation by our engineers and our architects.
Great.
Well, you heard Christian talk about Iceberg tables. We think it's obviously super powerful for reducing different data silos, as you kind of indicated earlier. What's your thoughts on Iceberg? Is this something that you're evaluating? Are you looking into it?
Yeah, this is an interesting one. When I actually first heard the announcements on Iceberg, I missed the impact to it. I went actually back to my team and said, "Hey, what does this mean to you?" I'll say the engineers and the architects were actually fairly excited. Again, ExxonMobil's a 140-year-old company. We do have some silos of data still. We do have some legacy data platforms and things like that. And the cost to just move it all at once isn't a realistic investment for us right now.
So being able to tap into those existing assets and being able to pull and tap that data is a real appeal. And our architects are excited.
Great. Well, we're excited too. So in terms of obviously, AI is super popular to talk about. How are you at ExxonMobil looking at AI? What are you thinking about it? How are you looking to adopt it?
Yeah, we have kind of multiple views of AI. I think one of the challenges we have is it's at peak hype a little bit. And so I always get a little fear for when the business comes to me and says, "Hey, I want to use GenAI to do something," and I don't know what yet. And so a technology looking for a problem to solve is never a good thing. We really try to root things in. What is the business case?
What is the opportunity? And sometimes we'll come back and say, "We do have a solution," and maybe it's AI. It may not be GenAI for this particular one, but it is AI. So sometimes we have to do a little bait and switch and say, "Hey, here's the right tool for the job," type of thing. So I think that's kind of a key thing for us. But I'll say beyond that, we see a lot of opportunities, and we're certainly evaluating and actually doing a lot in the AI space. We have a whole team of data scientists and lots of expertise in there.
The other challenge I think we have is this is your data ready question. And we've had to tell the business that sometimes the data quality of this data set is not ready.
And it's actually very important for us to say, "Hey, don't make that investment. Don't place that bet right now because the data quality is not ready." And again, having that data centralized, having an understanding of what is the data quality is really important and really critical to say, "Yes, if you want to do that, here's the pre-work we have to do. There's some homework. There's some data quality. I need the business subject matter experts. I need to do some things to clean that data up first."
Great. So we've kind of ripped through the questions pretty quickly. But kind of the final one is, where do you see Snowflake going forward as a part of your architecture?
Yeah, I mentioned to you, it's definitely one of our strategic partners. I think it's something where we're looking at the roadmap.
We're trying to give our feedback and influence the direction of Snowflake. So I think we're excited by it. It's a strong partnership. We see it continue to grow over time. I think we're very keen on not moving our data around. So if you can have one copy of your data as much as possible, that's a huge advantage. It's a huge advantage for multiple reasons, by the way. There's a cost component to that, right? But there's also, if you're trying to comply to GDPR, if you're trying to comply to Sarbanes-Oxley and different things, where you have to audit and
you have to have traceability and things like that, the more copies of my data has, there's a lot of internal overhead that that brings in as well. So again, we have this centralization strategy. We have this vision of leveraging our corporate scale.
And really, that puts Snowflake at the center of that for us.
Well, we like being at the center. So thanks for the support. So Andrew, thanks for being a customer. Thanks for doing this investor day for us. And hopefully, you'll have a great summit. Thank you.
Thank you.
Yeah. Thank you, Andrew. Thank you, Chris. All right, for our final presentation, the moment you've all been waiting for, our Chief Financial Officer, Mike Scarpelli, will come up and discuss how we're evaluating our efficient growth framework. So with that, I'll invite Mike up to the stage. Thanks, Mike.
Wow, did you guys clap when Christian or Sridhar came up? Thank you. Good afternoon, everyone. I appreciate you guys being here today. So I'm going to talk a little about investing for efficient growth, but I kind of want to go over a few numbers as well first.
So I really w ant to talk about our core opportunity. Our core opportunity is strong. We are in a large and growing market. According to Gartner, this market is going to grow 2x over the next five years from $152 billion-$342 billion. As Sridhar mentioned before, there's going to be many players in this market. Large markets attract people. But we're going to have, we feel we have the right products at the right time to capture more than our fair share of this market. And we've been doing that, and we will continue to do that.
The potential for Snowflake spend within our largest customers, this just keeps going up every year. If you go back in looking at our top 25 customers, this has beaten our expectations from the early days. The average of our top 25 customers was spending $12 million in 2022.
It's 21.9 million today. 21.9 million today or last year on a trailing 12 months. We expect that's going to continue to grow. We are very, very focused on large companies. But that doesn't mean we're not focused on the mid-market and the smaller companies as well. That is, as I've said many times, the beauty of Snowflake. The exact same product can service a company of two or a company, as you've seen Siemens today, 311,000 people. It's pretty unique to have a product like Snowflake.
Significant expansion opportunity. You can see the green. That's the customers I just talked about. But if you look at our million-dollar-plus customers, these still continue to grow. On average, our million-dollar-plus customers are spending $3.7 million a year.
You may say it was $3.7 million last year, but we added so many more into that cohort in the last 12 months and will continue to add. Our G2Ks, these are spending $1.6 million a year, up from $1.2 million last year. And there's no reason, and I've said this before, that a G2K can't spend $10 million+ a year on Snowflake. It just takes time to roll them out. Like Andrew, thank you for ExxonMobil. I'm not going to say what you guys spend, but I remember when you guys started, it takes time in a large organization where there's many different people involved to
actually ramp that spend. But when it does ramp, it can become very, very significant. What are the growth drivers for Snowflake? There's two things that we really focus on.
We focus on landing new customers, and we focus on expanding within our customers. That expansion happens in many different ways. It could happen from a migration and then another migration. It also happens with new features coming out, new workload opportunities we find within customers. One of the things that we've done, we've consistently added these new customers. Sridhar mentioned earlier, we're close to 10,000 today. This is as the end of the last year.
Our net revenue retention in 2024 was 131%. These high net revenue retentions are an example of how we continue to expand within our customers. We are very focused on the sales side to get our customers to expand. If you look at our Fortune 2000, the Global 2000, these guys are actually growing faster.
We expect that this will continue to outpace the overall growth of the company in terms of net revenue retention. Investing for growth, what we've done, we talked about sales comp plan. This is something if you go back five years ago when I first joined the company, they were just paying reps on booking, a customer buying capacity. Reps were never really focused on getting customers to consume. We started gradually changing so that reps were paid on the growth, whether that was landing a new customer or an upsell to an existing customer.
And then we started on consumption. This year, we made an even bigger change. We said, "No, going forward now, reps are only going to be paid on landing." About 35% of our reps are focused on landing new customers.
And if they land a follow-on deal within that 12 months, they get paid as well too on that. And then 55% of our reps are focused on expanding within our customers. They only get paid if customers are consuming. And that forces them to be close to the customer to identify new workloads, to work with the customer to get them to grow more. But the other thing we found too, and Andrew talked about this when he was up here, they look at, and customers look at different things, but ExxonMobil looks at price per query.
As price per query becomes cheaper, customers are willing to move more workloads to Snowflake. They look at that. I look at that stuff internally. Are we getting cheaper? When I look at my own consumption of internal cloud spend within customers, every customer does this.
About 10% of our reps are in this hybrid model that we talked about. We're also trying to get our partners more aligned with this as well too, to get our partners to be incentivized to get customers to consume more on Snowflake. That is a big focus of what we're working on now. I want to talk about what drives our expansion. We've talked before about migrations. Migrations historically, and they will continue to be one of the drivers of our expansion. This is an example of a Global 2000.
They've done over the last few years four migrations, and their product revenue went up 4x on Snowflake. I guarantee you their price per query went down though in that time. Here's another example of a telco over a two-year period. They went from $2.3 million in revenue to $19.3 million.
That was three migrations, and they're 8x the product. There's more work to be done there. Then when you layer on top of this all the new features that we have coming out that have nothing to do with migrations, you start to see further expansion within customers. This is another example of a financial service customer that started using Snowpark. I will say they're one of our early ones that we learned a lot on the product of what we needed to do from an engineering standpoint and made the product better.
But you can see now they're contributing to roughly $500,000 of their revenue the last quarter came from Snowpark. We think, and I know there's a lot more work with these guys as well too. New features are driving revenue growth within our customer base as well. Financial model.
I want to really go through some modeling points with people because you guys are asking a lot of questions on modeling. And I know you guys know that I really don't like when people ask me modeling questions, but I'll answer a few today. So there's this big misconception about queries and what does that really do. And queries is really an important signal of engagement within our platform. As customers do more work on Snowflake, they run more queries.
But you have to understand how do things relate to one another. And so price per credit, there's a lot of people saying, "Oh, Snowflake's discounting more and the price per credit is going down." Well, actually, our price per credit went up 2% year-over-year in Q1.
If you look at product efficiencies, these probably have the biggest impact on our customers, which, by the way, ultimately drives more revenue to Snowflake because we become cheaper. They move more workloads. But as we have performance improvements, and these are either in the form of software improvements that our engineering team works on or hardware improvements that the CSPs put out there, they're neutral to query growth, but they're negative on our revenue per query.
Optimizations. We've talked about optimizations in the past. There's probably more than there's been a lot of talk in the last few years on optimizations. Optimizations will always happen in the Snowflake environment. But optimizations are good too. They don't impact query, but they do have an impact on revenue. And that makes the price per query cheaper too, which drives more revenue in the future. But this is the big thing.
There are different types of queries within our customers. We call them heavy queries or light queries. Heavy queries tend to be very compute-intensive. As you run more of those heavy queries, it drives revenue per query up. You have a lot of these light queries. Light queries will drive your number of queries up, but they actually drive your revenue per query down. There is a mix. You can't just take our query growth year-over-year and apply that to our revenue. It all comes down to what is the mix in that.
As I mentioned before, our margins are very strong in our core business. If you guys go back to before we went public, we were running product margins in the high 50s%. We probably got to 78% faster than we even thought we would get to.
Yes, we guided to 75% product margins in Q for the year we talked about at the end of Q1. That's driven by a number of things, which principally it's around GPUs. But it's not just GPUs. There's other things too. We have to invest in GPUs for our customers. We talked about Cortex going GA right now. We've already been paying for those GPUs that are in our COGS that aren't factored into any revenue associated with those yet. You can see our non-GAAP operating margin.
We were 8% in 2024, probably faster than where we thought we were going to get to. Maybe we were underspending. But one of the things that we are doing is we are investing more heavily. We had Sridhar and Christian up here talking about delivering product faster, more innovation.
We are investing for more innovation right now, for which we don't have the revenue factored in there yet. I'll talk more about that in a minute and what we're going to do there. Then on the adjusted free cash flow, no one ever thought we were going to be at 29% free cash flow in 2024. I never thought we were going to be. I actually thought if you went back a few years, we would have been around 25%. So much faster there. Once again, we're sitting on $4.5 billion in the bank as of the end of last quarter.
We are going to invest in this business, but we're going to invest efficiently in the business. We're still very much focused on continuing to show margin expansion. This year, though, we decided we'd invest a little bit more heavily.
You will see in the future margin expansion. Now, what's the framework around our investments for AI? I don't think anyone will dispute our core business. We've shown you can have good margins. If you look at our 25 revenue guidance, we have zero revenue in there for our AI initiatives, even though we talked about products like Cortex coming to market and Snowpark Container Services is coming to GA as well right now. That is why are we making these investments?
We think long-term it's going to have a huge revenue contribution. But you need to spend the money upfront, as I talked about, that are going to impact your product margin and your operating margin. There are huge opportunities for efficiency in the product gross margin line. We're dealing with a few things right now, hardware availability.
Right now, with our CPUs, we can get what's called an on-demand model with the CSPs. They do not offer on-demand models right now for GPUs. So we're having to pay for the GPUs, whether we're using them or not. We give them to our customers in an on-demand model. And we do expect over time there's going to be more hardware availability. All of the CSPs are working on their own GPU formats. And as supply chain comes up over the next few years with NVIDIA, we think there's going to be more there.
And there's going to be better pricing that we're going to be able to get out of the CSPs. There's a lot of ability that we have to leverage native models that do not require revenue sharing.
Right now, we're assuming we signed some deals with people like Reka and Mistral where we have a minimum revenue commitment in there that we don't have the revenue in there, but we have the costs associated with those things in there. But we want to make sure those models are available to our customers. There's other efficiency things that we can do on the software side that we're working on, such as traffic pooling. You don't need to run everything right away.
And these are things that customers can work on. And we can do things cheaper for customers, give them better pricing if they're willing to trade off the timing when they want these things. And as we gain more visibility and we need more visibility into these things, we'll provide you guys with additional details on our longer-term margin potential.
But I'm not comfortable now to talk about what our margin is going to be in 5 years from now. All I know is we are very committed to continuing to show expansion in our model. And I will tell you, we pretty much are making these investments right now. And if we don't see the return on these investments, we will cut those investments. It'll probably drive us to do other things, but we feel pretty confident we're going to see the return on these investments. Now, capital allocation.
I talked about we have a lot of cash. Everyone knows about our stock repurchase program. We still had almost $900 million remaining at the end of last quarter. When we put this program in place, we said we would spend up to $2 billion on that. We are clearly on pace to do that.
Probably we may have even bought some more already this quarter. Organic growth, we're going to continue to invest in R&D, but also on the sales and marketing investments. In Q1, we saw really good returns at certain teams. In particular, we saw really good on the commercial side. And when we saw that, we've given them more heads to hire more people. We've seen opportunities in certain regions in Europe and other places. And as long as we see the productivity, we will hire these people.
And I'm looking at productivity. These are mainly on the acquisition front. We haven't necessarily seen the revenue yet from those, but I feel confident the fact that customers are committing to contracts, they are going to consume that. Sridhar talked a little about M&A. Nothing has changed because I get a lot of questions.
Is anything changing your M&A strategy because Sridhar is on board here? Well, our M&A strategy has always been about accelerating our product roadmap. We're not looking to buy product. We're looking to buy really good teams. They typically are small teams of really good domain-specific engineers that can help advance our product roadmap. Now, some of them have some interesting technology as well too, but not looking to buy an installed base of customers or anything like that.
Just a cleanup item on cash management. We just are forecasting about 4% interest. It is a significant other income line. I want to stress this too that today, through last quarter, 80% of our billings are paid annually in advance. I've been saying this for a while.
I do think more and more of our large customers, when they get to these big bills, are going to want to do more how the CSPs charge, which is monthly in arrears. And so I do think that will have an impact on our free cash flow, and that is factored into our guidance there. So with that, I am now going to invite Sridhar and Christian back up on the stage once they move these chairs to get another chair up here, and then we'll go into Q&A. Does that mean you'd like the first question, Kash? Well, wait, there's a couple of mic runners that will come up here.
Matt, right up here. We got an expensive mic runner. He's one of my core data scientists on the finance team.
I'm honored. My question is not going to be as good as your data, but.
I don't bite, Sridhar.
Well, congratulations on Sridhar. Your first USA conference. And my name is Kash. I'm at Goldman Sachs. Nice presentation. Short, sweet, but to the point. Really liked it. Sridhar, one question for you is right now, at least the investor community thinks of unstructured data. Structured data is two separate domains. Data science, business intelligence being two separate things. You are bringing all this together under one common platform, one common technology.
Talk to us about how you see the future. Nobody has done it, even Oracle, structured data. These silos have been largely separate. You want to bring this together and at the same time become a platform for third-party companies to develop applications on. There's a lot that's going on here. So how do you think this all plays out? And who's the role model for success?
Who do you see in the technology world as having done something like this before that can be a template for success? Or maybe there isn't one. You're trying to chart one on your own. Thank you so much.
Thank you. I think that's a great question. I would separate these in sort of layers and time. When it comes to being a great data platform, for example, you need a position of core strength, which we have with our compute engine. The original magic of Snowflake was the separation of compute and storage, which led sort of produced independent scale so you could rapidly ramp up compute. But that compute engine was still and is still the magic behind Snowflake.
I think what is pretty unique at this moment is this rising wave of data interoperability, where a number of companies, including your own, have said, "Hey, data really needs to operate freely." And I think that will end up breaking down sort of a lot of silos.
What has happened is that even in the world of cloud computing, people have had just different chunks of data that were governed by, say, individual products. And I think you're seeing a very strong trend of data being interoperable. To me, that creates unique new opportunities in terms of who can act on this data. So the old formula of needing to ingest data into Snowflake before doing anything gets replaced by, yes, for your most important data, you want to extract it.
You want to make sure that it's actually gold standard because that's the thing you're going to run your business on. But on the other hand, there is backup data. There are all kinds of other things like videos, images that everybody has, presentations that are going to be on cloud storage. I think that is what creates this opportunity.
I would say us, for example, saying that we want to do things for data engineering and analytics and data science, there is a commonality to it, which is really it's about scaled data processing, but just with different ways to access it. If you're a data engineer, what you want to do is use a dbt or perhaps write some queries in order to stitch things together. And if you're a data scientist, you're probably living in a notebook, but you're running scale computation.
There is that commonality. I would say that's where our strength overall comes in. And in terms of platforms extending out, I think there are lots of examples I can point to. They all start with you need a core strength. So for example, if you think about, I'll give you two examples that I'm pretty familiar with.
One is how Google became a powerhouse in advertising. It was always based on being incredible at search. It drove a lot of demand, a lot of advertisers, lots of users. And then that gradually became the display network over time, augmented with acquisitions. Similarly, Salesforce, for example, started in an incredible position of strength as the place for CRM. And then over time, they had a marketplace. And in terms of our investments, just so I'm very clear, the bulk of our investments go to the core data platform.
And something like AI is a horizontal, but it is not like a massive amount of Snowflake's engineering is working on AI or is working on native applications. We see an opportunity to create value. We thoughtfully invest in it. And depending on sort of the returns that we come back, we will double down on it.
And so I would say the broad message is the data platform is very strong. And expanding that and making sure that we capture that is one of the most important things that we need to be doing. But there are opportunities that come, for example, with partners wanting to create applications, which they have, by the way. There are plenty of people that have built apps on Snowflake. We have just added on this framework for making it even easier to create and distribute these apps.
You should think of that as a step forward, but the core is the core. And I think that is going to drive a huge amount of our growth.
Right up here. Mark, you can have the next one, Kirk.
Thank you. Mark Murphy, JP Morgan. Wondering if you could help us trace out the impact from Iceberg Tables over the short term, medium term, and long term, potentially either activity levels or revenue impact because we do understand it's a net headwind on revenue this year due to the loss of some of the storage revenue. But we've heard you talking about, in a pretty tantalizing way, Sridhar opening up some data volumes that might be 100 times, 200 times larger.
And those data volumes were previously inaccessible to Snowflake. So I think we're wondering, is it something where you see it as a headwind this year? And then maybe there's a time frame where Iceberg Tables kind of transitions over and actually becomes a revenue tailwind.
Look, the way I think about this is Snowflake is a pretty large company. And so there are many things that we can absolutely do in parallel. I can assure you, for example, that the breakneck speed at which we've been getting quality products in AI out has not come at the expense of being really, really good in the data core. It's a multi-thousand-person team. On things like Iceberg, the use cases that it can create for us, we absolutely have an internal effort that prioritizes, for example, data engineering use cases, unstructured use cases, all of the
enablement material that needs to go to our sales team, to our partners' team, and mechanisms to track these things so that we can go back to the mode of very tightly integrated teams up and down, like the product engineering business stack, in order to drive these opportunities.
We're certainly not waiting for these transitions in time. But to Mike's point, we have modeled out the tailwinds from Iceberg. They are modest. But the effort to take these capabilities to market and turn them into customer value is already well underway. And definitely, I hope to be able to talk more about it in the coming quarters.
Can I just add in this year, for modeling purposes, we view it as a headwind. We do think, obviously, that it could be a tailwind. And there is the potential this year, but we don't know. As you know, we don't model revenue upside until we see historical consumption patterns with customers. So it is a headwind this year, as we discussed earlier on the calls. Right. Go to Kirk, and then we'll get to the back. Sorry. You're next.
Thanks. Kirk Materne, Evercore ISI. Yeah, Christian, I want to build on the slide you showed in terms of product sort of revenue contribution. And I know you're referring back to what Mike just said, that you're not going to really guide anything or put anything in the guidance. But when you think about customer adoption, can you take on multiple products at once if you're a customer? Meaning, historically, an enterprise, it's been difficult to take on too many new products at once.
And do they compound on each other, or do they add to each other? Meaning, if you take on Snowpark and Streamlit, does the consumption revenue go from being linear to being more exponential? I'm just trying to get a sense on you're giving customers a ton of new product.
I'm just wondering about sort of their ability to adopt them and then what that means in terms of sort of a curve around the consumption. Thanks.
Yeah. So it's a great question. In reality, that slide makes things look more discrete than they are in reality. I mentioned the opportunity with unstructured data is intricately tied with the Snowpark opportunity and the AI opportunity. The AI opportunity is intrinsically tied to the applications opportunity. Same with the Streamlit opportunity. So the beauty of what we're offering as a unified platform is that there's fluidity between use cases, and there's a lot of synergy.
So I don't think that there's eight columns, and each one needs to have a different strategy and approach. Customers and partners will be able to leverage things in a more cohesive way.
The only thing I'll add on to that is, look, we don't sell eight products. So you might be mistaking that slide as one in which we have to go sign a new contract for something. That's like the mindset that comes when you see, "Oh, this is something that is distinct." They buy one product from Snowflake. All of the things that we offer are just included. It's also equally important to understand that we are in the business of solving business problems for our customers.
We are not in the business of selling Snowpark or selling unstructured data. And so a customer looks at any place that has a search box. Let's say they have help text, and they offer a place to search.
They go, "I want that to be an interactive modern experience because it can make my customer experience better." And underneath, that might involve Iceberg. That might involve the Polaris Catalog. That might involve our search index. But that doesn't matter. That's not what the customer cares about. They care about the value that we are delivering to them. And this is where I think we are uniquely positioned to use these to create that value. And remember, we often work with teams that have several hundred people that are working on
sort of various instantiations of Snowflake within a particular enterprise. So it's not the case that one person has to all of a sudden become an expert on all of these. But the best answer that I have for you, we focus on business value and the technologies underneath that are going to be used to serve something.
That's the means to the end.
Next question, then there.
Okay. Great. Karl Keirstead at UBS. Sridhar and team, I'd love to give you the occasion to address the competition question that I think a lot of investors have. I think there's a perception that when Snowflake went public, it was a mile ahead in terms of having a cloud-optimized data warehouse. But today, the hyperscalers, maybe because it's a natural upsell to their AI stack, appear on the surface at least to be leaning pretty heavily into their alternative offerings.
You've got Databricks, obviously, that's approaching your scale. So Sridhar, what would your rebuttal to those concerns be? And when in a head-to-head, what are the one or two key attributes of Snowflake that you hit on? Thank you.
I mean, first and foremost, when it comes to our sweet spot as a data platform, I would say we are still world-class. This is not to say that there's not going to be competition. There is going to be competition simply because of the scale of numbers that you're talking about. I would argue that we are structurally advantaged when it comes to winning these workloads, both because we are very easy to use, we are very cost-efficient, we lower the total cost of ownership, but also we come with things like a plethora of business relationships that people
want. And in fact, collaboration is now an entry point for us in terms of winning customers over. And my sort of broad rebuttal would be that Snowflake is not going to get replaced by a set of 50 individual services that are being offered to other people.
Does this mean we have to compete hard in new areas? Absolutely. This is the reason why we are getting a notebook ready very, very quickly because that is important to capturing the entirety of data use cases that our customers have. And that's the reason why we press the gas really hard on making sure that we are world-class when it comes to AI. But we do it in a way where if you're using our AI, you don't have to worry about governance. You don't have to worry about taking your data, putting it somewhere else, recreating the governance that you
need. It is really that. I think that's the unique differentiator for Snowflake in that it is a tightly integrated product that works exceptionally well. And I think it's the product philosophy that separates us.
I would sort of go back to the iPhone versus Android analogy, which is that the iPhone ecosystem was a tightly controlled ecosystem, was a tightly integrated ecosystem. Everything just worked. That doesn't mean that there is not an Android, for example, but it's really our product philosophy that I think of as the biggest asset that we have that is not going to be taken away in a hurry because that starts at the beginning of how various companies look at it.
Who has a? Pat?
Hi, Pat Walravens at Citizens JMP. So Sridhar, I want to talk about culture a little bit. And I thought from your keynote at the beginning, you were trying to share a little bit with us in terms of who you are, like playing the drums and then getting a PhD instead. What was the culture that you found when you got at Snowflake, when you got here? What is the culture that you're trying to drive? How is that coming along? What's the hardest part?
It's three months. So it's a 7,500-person company. So these things take time. But my sort of two consistent qualities that I bring to the table that I like all of my teams to have, first and foremost, is a relentless sense of urgency about getting things done. Opportunities vanish, competition shows up. And just having that sense of urgency, yes, there are other things like work ethic and strategic thinking and so on, all of those, but having that sense of urgency to get things done.
And the other one is competitive paranoia. We are in a large market. Large markets always beget competition. And I think it is important that we learn from the best of what everybody else is doing and make sure that we are better than them.
I would say pressing both of these really hard has been an important part of what I bring to the table. There are other, I would say there are other elements, like figuring out these are less cultural. It's a combination of sort of culture and experience. Things like how do you take new products to market? What are the special skills that you need to bring back? Or what are extraordinary things that you need to do if something is not going as well as expected? When to really bring on, it's sort of an overused phrase, when to bring on a wartime mentality as
opposed to driving an expansion peacetime mentality. As I said, there are plenty of times during my Google career where lots of colleagues were willing to say, "Oh, we are at a great place, and that sounds kind of good.
I guess it's a lower growth rate." But a number of folks, including me, would never accept that for an answer. I would say that I have the history to show what it takes to drive this sort of sustained excellence over a decade plus. And it's the drive and paranoia that I'd like to see more of. Look, the other thing that always comes up in these is you're going to turn it on and ask me a counterfactual of, is that to say it wasn't there? That's not what I'm saying. I'm saying it is really important to have these two qualities because I see them as being essential for
continued success.
Derek? No, right there. He's right there. That's Derek.
Thanks. Derrick Wood at TD Cowen. I guess, Christian, for you, you mentioned that you've got customers that are either in production or ready to go to production with Cortex AI and that you will see some real consumption revenue this year. Maybe too early to tell, but just curious how you're thinking about the impact on the core when customers adopt AI. For instance, if you built a chat app and you're interacting in a conversational way, maybe that shifts the query from kind of a traditional CPU query to a GPU query.
Or do you see it all additive, or do you even see a multiplier effect? Just curious how you're thinking about AI and the impact on the core.
Yeah. So two thoughts. One is there are aspects of Cortex that are 100% complementary to the core. So Cortex LLM functions, examples like a translate operation, a summarize operation, that is what people want to do as part of the core business. And that's where I think we're going to see the lowest hanging fruit adoption because it's targeted to our audience, our users. It can seamlessly expand existing use cases. So that's one part which is completely additive.
There's another one where you talk about chatbots, where products like Cortex Analyst that we announced today, where I have a chatbot that is effectively answering business questions. But the beauty of that model is the chatbot is going to understand what the question means and how to answer it. And at the end of the day, it generates SQL, and we run the SQL statement.
So I would say if we had a strong BI product ourselves, which we don't, I would say, "Ooh, there's some confused mix shift in there." But from what we have, I think it's just additive. Some of it with more synergy, the other one with more like a shift from how queries are coming to us. I don't know if you want to add on to that.
I think this world of AI is pretty exciting from an analyst perspective because it vastly broadens the set of users that can get at business data and get meaningful information. Like any other sort of game-changing technology, I think the effects on what's going to happen are a little bit hard to predict. But we especially like being in this position of now Snowflake being directly accessible by everybody in the enterprise, but still in this governed way. So I see that as a huge positive for us.
But again, I'll stress again that this all has to be in the context of creating value for our customers. And so rolling out use cases thoughtfully is an important part of how we are going to approach this.
Alex.
Hey, guys. Alex Zukin with Wolf Research. I have a multi-part question. I'll try not to scream. My favorite slide was the green light, yellow light slide. There was a lot of products. But most importantly, and this might be a tough one for you to answer, Mike, but if we think about Snowpark as a gauge of revenue impact of early adoption, but once it's generally available, if you look at fiscal 2026, and I use that as a gauge, it's like double digits of your growth potentially are coming from these new products.
So can you help frame kind of what that could look like? And then have any of these products that have been on the roadmap that you just launched, can you kind of stack rank the gating factors of people have been waiting for this to really start driving consumption?
And then the third one, given since you've reported there have been some kind of wobbles from other consumption players in the marketplace in terms of just a macro backdrop, and you usually have a pretty good gauge, just give us a sense of kind of how May kind of was trending or ending into any things we should be thinking of on macro demand.
Well, I'll start with that one first is we just guided 10 days ago. Nothing has changed in my guidance. I was looking at consumption patterns through, what was it, the 25th of May or whatever the date was. So that's all I'll say on that. In terms of what it could be, I'm not in the business of giving out what it could be. We like to see data. As we've said before, we don't forecast revenue until we see consumption patterns start within our customers. So stay tuned when you see a little bit more of that before we do that.
But Christian can tell you exactly what customers are asking for the most and what are the gating factors to adoption.
Yeah. I would say for at least the items that we had in the yellow-green slide, the high order bit for everything is we need to be generally available. Our largest customers, all of them have very strong policies that maybe they get to play with it and do some evaluation, but nothing goes real without it, which is why what we did this morning was significant. There's another aspect, which is cloud parity. Even though a big part of our business is on AWS, we have a large growing customer base on Azure.
I highlighted in a few of the announcements, not everything is on par right now, and we're working fervently to get to that parity. And maybe a third one that is at a different level, but still is coming up from customers, especially in the context of those investments, is support for private connectivity.
That holds back a certain set of use cases, and we've started the rollout of supported private connectivity on AWS for those products. I would say within the next quarter or so, we'll have a much more rounded experience. But the good news is that for pretty much everything on this slide, we have customers that tried it, liked it, and they went and said, "We want to do it." For me personally, the best indicator of product-market fit is when we see customers starting to ask for that, "I want to go production in preview." And for pretty much everything in that slide,
that has been true. But that is not the majority.
Okay. Brent Thill up here. All right. Right up here, Matt. Brent, put your hand up so he sees you. There you go.
Thanks. Just to follow up, Mike, on Alex's question on demand.
I guess I shouldn't have asked you.
No, I'm just curious. What do you think is going on? Some have suggested that the software stall is because of AI investments. There's a lot of work and not ready to move forward. A lot of the products aren't ready. What do you think is happening when you talk to customers about what our industry is going through right now? Because it's pretty consistent from apps to back office to all over. We're seeing it. So I'm just curious if you have a view, having spent a lot of time in the industry, and what your customers are telling you right now at the
conference about.
Yeah. I'll let Christian and Sridhar, who talk to customers a lot more than I do, talk about what they're hearing from customers. But listen, things can always be better. And I do think, in general, things are, but it's factored into our guidance. They're maybe a little slower than where we thought. As I said, February and March were very strong relative to what we were forecasting.
April came in there. May is about the same as what we were forecasting. But we feel good about what we're seeing. We don't see this massive slowdown in any way. And I'll let Sridhar or Christian talk about what they're hearing from customers.
Yeah. I would say demand for the core product is very healthy. It's very strong. And 100%, the appetite for things like Iceberg, which several of our customers do see as a game changer for them, is very strong in the core. And AI conversations run the gamut from people that have already tried language models. They're able to build a chatbot, but they're not able to build an analyst. So there's levels of sophistication. And so, but with a lot of those customers, the conversations inevitably go to what are the things that we should be working on where we
can create value. That's why the 700+ use cases that are in active implementation right now. I would concur with Mike and say that our core business feels very strong to me. And I'm out on the road pretty much every other week meeting with customers.
Generally, as I was saying earlier, sort of the core product is much loved, and people are very open to having conversations about what other things could they be doing. I generally focus on what are the business problems they would like us to partner together on.
Let's go to this side of the room. Right back there.
Great. Thank you. Mike Cikos from Needham. Wanted to circle up on, I guess we're talking about macro here, but more the go-to-market. With the shift in comp plan towards driving more around new business and consumption, can you talk about the mechanisms that you guys have in place to ensure that Snowflake is acquiring the right customer who still has the same potential for expansion? There's been some other infrastructure names, and consumption is different, especially at these different layers here.
But we don't want to turn around following some of these comp plan changes and find out that the wrong customer might have been introduced, and there's an air pocket to growth as a result when we turn back a year from now.
Yep. No, good question. Just to be clear, yes, we made a change to our comp plan this year. We've made changes for the last five years every year, and we're getting to where we wanted to go because you can't make it that dramatic. We do have people that have when we say we have people that are focused on landing new accounts, there are people that are focused on that have named Global 2000 accounts are going after. They tend to be large organizations that we're targeting.
Yes, there is a commercial team that just focuses on the commercial segment, but we have an acquisition team within our enterprises. We have an acquisition team within our large verticals that has a lot of the Global 2000. People are incented very much to land the right customers. Right up here.
Thanks. Ittai from Oppenheimer, and I appreciate the presentation today. Maybe I want to tie Mike, your comments to Christian, your comments. Mike, you talked about heavy query and light query type of use cases with customers. Christian, with all the features and capabilities that you're introducing, I think that everybody knows ultimately it comes down to CPU cycles. How many more can you generate? One of the challenges I've had talking to investors is figuring out which one can actually drive a lot versus less, or maybe in Mike's words, heavy
versus light.
Help me think about, of all the things that are coming up here that you have introduced just in the past six months and coming in the next six to nine, which ones would you anticipate a year or 2 from now will have more of a heavy query load drive, much more CPU versus light query volume?
Two parts to it. One, under the premise of your question, I would say Iceberg, Snowpark, unstructured data, Cortex, all of them have great consumption potential. Maybe what I was caveating was on Streamlit where because it's an app and it's a lot of interactive experience, that's where you see more of the lighter queries. But I also would highlight that we will evolve our business model as needed. And the specific one is on Hybrid Tables and UniStore. Those queries are going to be tiny and super fast, like 10 times faster than the fastest we run today.
But we've priced it in a way that in that world, shorter queries does not necessarily mean less CPU, means less money because the value delivered matters a lot. So we want to be very judicious that in the current context, your question, the premise is correct, and I answered it.
But also, we will see a lot of queries coming with Hybrid Tables, and we will make sure that we capture the value that we're giving to organizations.
I'll add to that, Ittai. By the way, every query, whether light or heavy, are incremental revenue. It's just the revenue per query, which a lot of you try to do the math equating total revenue with number of queries. That's why I was trying to show you that. Get on this side over here.
Patrick Colville from Scotiabank. I guess, Sridhar, I want to just circle back to this kind of walled garden versus open approach. I mean, when I speak to your customers and partners, that is the biggest difference that they highlight that Snowflake is taking versus peers. So I guess, why do you think this kind of walled garden curated approach is the right one? And I guess, what proof points can you share with us thus far that this is the right direction to travel?
Can you just clarify the walled garden part? Because I would assume that with our support for Iceberg and Polaris, at least when it comes to data, we're really saying that there is a larger role. Data is going to be interoperable. So what do you mean by walled garden?
I guess like Streamlit, container services, the new announcement this morning around governance that as a customer, you have your data in Snowflake and you don't need to leave.
I would say it's much, much less about that and in more having a unified platform. Absolutely. One way to look at how, let's say, the data stack should be organized is to think in terms of there should be one vendor for storage and another vendor for, let's say, the processing layer, and then one more vendor for, let's say, a lightweight application layer, and then a different vendor for providing GPUs. That's the kind of architecture that tech companies full of engineers love.
That's not the kind of architecture that you want if you're a bank out in the Midwest or if you're a healthcare services company that wants to drive business outcomes. We are clear about where we add value.
With things like Cor tex AI, for example, we bring AI to pretty much anyone that can touch Snowflake, whether they are an analyst or a data engineer or a data scientist or with things like Cortex Analyst, even a business user. Absolutely, we see that as a strength. But it's also important to understand that we don't box people in. For example, you can absolutely use our search services to set up a semantic index. On the other hand, as a customer, you feel particularly passionate that you want to use that from an application that you've developed
where you're using GPT-4 as the model. We 100% make that possible. We even make it possible for people to use GPT-4 from within Snowflake as an API integration point.
So we are very careful to make sure that we interoperate and understand that we are part, an important part, but a part of the data ecosystem of our customers, many of whom like they spent $1 billion on just data and data infrastructure. Of course, we want a big part of it, but we are also careful to position ourselves as, as I said, playing well in this. So I don't see our desire to create services end-to-end as a walled garden. I see that as healthy options that our customers love because I can tell you if you use Snowflake that you can set up a chatbot in
five minutes and have a front end for it without needing to spin up even a single new server. I see that as a huge positive for the product and much less an exclusionary walled garden type approach.
Matt over here. Brent. Oh, okay. Brad, I don't care. One of you. I'll get both of you.
Thanks very much, Brad Zelnick, Deutsche Bank. Congrats on another great summit. My question, Sridhar: I think one of the messages that's pretty clear from this afternoon and from the keynotes is accelerated pace of innovation. Absolutely. And it's very clear that you have a point of view in how to achieve that. Yeah. Can you maybe double-click for us on how that happens from an engineering standpoint? And what, if at all, are the trade-offs or costs that that comes with? Thanks.
Yeah. This is a great question. And as I said, we start with the things that are absolute must-haves, which is whatever we design has to work within the context of Snowflake, meaning that the investments our customers have made with governance, other things that need to work out of the box. We have a well-defined, basically like engineering framework for how we think about how a new feature plays within Snowflake. That's part one. Part two, Christian touched on it a little bit both in the summit and here.
We made a series of investments with respect to extensibility within Snowflake. Snowflake started as a core data platform that was super tight, so tight that it meant a limited number of binaries.
But what the team did, this is roughly two years ago, it came to fruition about a year ago, is this extensibility framework that lets us bring on new capabilities. So a lot of things that you're seeing now, whether it is Document AI or Cortex AI, really run in this extensibility layer. And so the way I think about scale and speed is, first of all, have the most important thing is have the safety scaffolding in terms of how quickly you get releases out, how you can ensure that you don't create production problems, and how you can also ensure that you can quickly recover
from production problem while maintaining an incredibly high SLA. That's like the safety net. Having done that, you make sure that you have this extensibility framework that lets you deliver stuff very quickly.
And the rest of it is also there's a lot of product magic that goes into how do you think about products that you're delivering in incremental fashion so that you can add onto it. And it is that thinking and a more iterative delivery style combined with a look, it is important that everybody in a company be asked in a nice but firm way to pull their weight in terms of speed and execution. 100%. I push that very hard. I'm pushing that very hard with respect to go-to-market for all of the new things that we are delivering.
There's an incredible amount of just operational scrutiny into what does the pipeline look like, what's the health of the various use cases. So it is really like this approach that goes across all the layers of the stack and the business.
As I said, having done this at a scale that's 10x that of Snowflake gives me both the experience and the confidence to be able to say, "Sorry, these things are not, and neither are. We can absolutely get it done.
Okay. Brent.
Good afternoon. Brent Bracelin, Piper Sandler. We are that Midwest bank that probably needs to use more Snowflake for the simple architecture.
Happy to connect you.
I wanted to go back to this existential debate a little bit that's kind of come to life with a billion-dollar acquisition of Tabular. Maybe it's $2 billion. We don't know what the exact fee is. But it feels like for a 50-employee firm, that's a lot of money to pay. And so.
We didn't do that, did we?
You did not. Your competitor did. But in the context of what you want to do, where you're going, are there limitations to what you can do in AI if the data is stored externally in an Iceberg table format? Are there limits to UniStore functionality where you need compute and storage together or not? Just walk us through the roadmap as you think about what you can and can't do if data and compute are together.
I mean, first of all, as I said, tsunamis are different from little acts that any company, including Snowflake, can do. The tsunami that is right now washing over companies and data is that data needs to be interoperable. We also realized that cloud catalogs, which is really it's your directory for how you get to data, was going to be a new choke point. It is really our investment in Tabular that made us realize that something like Polaris was going to be pivotal. Similar to what we did with Arctic, our LLM, which was developed from scratch in three
months, Polaris had an even faster genesis to shipping kind of experience. That is an important part of what we need to be as a company, which is to seize opportunity.
I would say that the world is very, and by the way, I think it was a team of six or something that worked on Polaris for three months to ship it. So, you're not talking about, you can do the math on that one. And so I see a world very much in which data is in open formats. It's interoperable. It is easily accessible. And companies have to win on the quality of the products that they build. And I am very happy with where we are and just how quickly we can get things done.
And I would add one quick thing. The big, large industry players are all aligned behind this world of interoperability. I don't know when was the last time I saw a press release that had Amazon, Microsoft, and Google on the same one. And they're behind what we're doing. And it's all in the name of interop. We'll go to the back there.
Hi. It's an ask for John from Brent Bracelin. One question, two questions, two part questions. First one for Christian. Iceberg tables, you mentioned that that opens up a lot of data, which is not ingested in Snowflake today. But longer term, isn't that a risk to Snowflake that you have customers who can now move their engines because the data is not in Snowflake?
A little bit. Yes. So it levels the playing field and it lowers to a degree switching costs for customers to find the right engine for the right use case, who has the best experience, the best price performance. And this is where we have confidence that that level playing field is going to play in our favor. But yes.
But data also does not exist in a vacuum. It exists in the context of other groups that are using this data, other companies that are using this data, the code that's sitting on top of data. We think of data interop generally as a good thing, but there's a lot more that we bring to the table. Yes, there will be some use cases that we will not handle well. This is why we have the partnership approach with folks like RelationalAI. And as I said, Snowflake's primary value prop was an incredibly efficient compute engine, and we can and will stay that way.
I will add to that, though. A number of our customers like the simplicity of having everything in Snowflake. Actually, one of our largest customers has said, "I'm not interested in Iceberg. I want everything in Snowflake." So yes, some will, but we still think a lot of our customers, and for our small customers, we give them storage pricing. We have tiered storage pricing with our buying power, which in many cases is cheaper than they could go if they went directly to the CSPs for their storage.
So I don't think that's a massive financial risk. And to remind you, 11% of our revenue is associated with storage, and the amount of data being stored in Snowflake continues to increase.
Thanks. Part two, Mike, your net new logos you're adding every year has been going down for the last two years. You made those go-to-market changes this year to incentivize salespeople to start landing new logos. Is there a time frame or, I guess, a timeline when you think your net new logos will start, I guess, moving in the positive direction again?
Well, that was the whole one of the thesis for that change. We are expecting that this year will be a better year, and we see a very good pipeline. So you'll see as we progress throughout the year.
Thanks.
By the way, I think we got time for two more questions because I have to go to another keynote. We're over already.
Thanks for letting me get in here. This is Simon Leopold with Raymond James. I wanted to come back to your comment about AI revenue becoming meaningful within two years because I want to understand how much of that time frame is reflective of the availability of your products and how much of that is reflective of your expectations for the market development. If the market is developing sooner than your products, how are you thinking about the potential business that might be lost given the timing of availability?
Well, first of all, I think I'll let Sridhar talk more about it, but I think our products are pretty much there today. We have Cortex out there. One of the biggest things, and I've talked to a number of customers and I know internally, people are really trying to figure out their policies internally, especially large organizations around the use of AI, what data do they want to allow to go into these models. And so I think that's what's going to take a little bit more time for our customers to figure out how they want to use it.
Yeah. To Mike's point, absolutely. We have Cortex AI. We have Document AI. These are already in GA. And the other products that we want to get GA, which are Search and Analyst, are I would say at most a few months behind. We feel very good about where these products are. Of course, it is a rapidly moving space in terms of what is possible, and we absolutely will keep up with what is coming. But in terms of code capability, whether it is in data pipeline that bridge between unstructured data and structured data, but it is not just text.
There's also images and videos and so on. We feel very good about where we are. And then in terms of our customers being able to use higher-level functionality, whether it's Search or Analyst to create additional value, again, we feel good about where we are.
And I would say the emphasis internally, like over the last three months even, has shifted from, "Let's ship these products," which we have, over to, "Let's make sure that we figure out how to create customer value, how to create those playbooks and POCs that can be repeated at multiple places to drive scale revenue." Lots of really exciting partner conversations as well because they see the opportunity. So I would overall say we are in a good place when it comes to AI. I wouldn't have said this a year ago. So I think it's a sea change in a year.
Last question. Tyler, you can have the last question.
Thank you for the honor. I'll try to involve you in the question too.
No, I don't need to be. That's okay.
Going back to the spending side of the equation, last quarter, you increased spending projections for the year. The first part of the question, I wanted to direct it at Christian and Sridhar. What specifically did you observe since Q4 into Q1? What did you ask for? Do you feel like you have all the resources you need? And then secondly, Mike, as you're thinking about the returns on those investments, can you just talk about the time frame? Is this an 18-month, two-year payback period?
I think I saw something that said two years on one of your slides, but just what's the time frame that you're evaluating those investments in terms of seeing that return? Thank you.
A quick framework that I would put on how to think about our spending in AI is in terms of training and inference. As you know, we trained a model called Arctic, which was pretty good for the amount of compute that we put into it. But essentially, we rented out an amount of GPU capacity to be able to run experiments, to be able to train meaningfully large models, not gigantic models, but still pretty impressive models. Because as I said, I think of that as a core capability.
Now, that is not really going to increase unless there's a breakthrough that we think we can achieve with it, which is, that's not my inclination right now. So that's one part.
And then the other part, and this relates to Cortex AI and operational deployment, is really around essentially renting GPUs around the world close to all of in our deployments to be able to handle inference traffic. It's a little bit like setting up a shop. You need a certain amount of inventory to be able to take on customers. And in the future, as I said, I only expect this inference part to go up, but it'll very much be revenue-giving. We have the capacity that we need for our customers to be able to run experiments, to be able to bring on actual loads.
And if a customer comes in with a massive new load that they want to run, that's fine. We'll go and spend more on those GPUs, but there is revenue to show for it.
Over time, I expect sort of inference spend to dwarf any training spend. But as I said, it's revenue-given, and we have multiple layers of optimizations. It is whether it's in hardware, whether it's in the new world of CSPs plus a CoreWeave, for example, which is a new player in GPU capacity, but also in things like more efficient models, which is why having a research team helps a lot. But also finally, in optimizations. Can we pool together GPU traffic? Can we offer a different pricing for batch workloads, encouraging customers to run workloads when
loads are low? Can we better share inference capacity between the training cluster and the inference clusters? These are all things that will make sense as our revenue scales up. And absolutely, we'll get the right folks on it.
But it's important for you to understand that this is a pre-commit for making AI revenue, and any further increase will be driven by actual business progress.
You should probably repeat the second question for Mike.
No, generally, we look for product traction within two years of starting something. Remember, this is our R&D. There's a lot of research, and there's a lot of money spent on there, and not everything turns into a product. But we feel pretty good about, as I said, a two-year kind of window we're looking at for these things.
An early indication or that there's ample demand, which is not surprising for sort of the AI product offerings that we have.
Okay. With that, thank you, everyone, for coming here today. I look forward to talking to you after our next earnings call.
Thank you all.