Snowflake vs Databricks: The AI Data War | CEO of $SNOW

0:00I was fortunate to be part of another trillion dollar company where a product that I was personally leading search ads spent from$1.6 billion in revenue in 2003 to making$100 billion the year that I left. Eric Schmidt, our CEO at the time, made me and a few other folks write a$100 billion revenue plan. This was like 2007, 2008. We all thought that this was the funniest thing ever because the idea that one company could make$100 billion just sounded so preposterous. So we're in a bit of a wartime environment and one of your competitors recently raised, I think it was their Series Q. Rumors of IPOs coming around.

0:39How do you view your strategic positioning against someone fast moving like that? There's no question that we have to move fast. We have had pretty good quarters, but bluntly, we should be asking ourselves why we are not growing a whole lot faster. What's the biggest secret you can tell me right now?

1:03Straight Art, welcome to Sorcery. Excited to be here. Thank you, Ma 'am. Thank you for having us in your glorious snowflake palace. It is a nice office, but I'm always very quick to point out that we subleased it. It's not as expensive as it looks. And Meadow was here before. Meadow was here before. I think only partly. I also heard that there are rumors you have a secret ski slope.

1:33We do? Okay. They say we do. I'm not that much of a snowboarder, so they must know better. So I also noticed that there's only snowboards here. There's a lot of snowboards. There's a lot of snow gear. We are a very blue snow. You're just good with any kind of, yeah. Light blue. So since you are the data expert, I want to know what's happening with data centers in space. to be honest i'm not convinced that i understand the economics of that thing or whether the economics are really going to work if you went and do as you know you can do a lot of like fun calculations without spending a lot of time by asking chat gpd questions if you ask gpd questions like hey how big does my sail have to be in space if i have to use solar power to power my one megawatt space data center I want to run.

2:30It's really big. So I'm not completely sure of how this thing is going to work out. Maybe the same people that want to send it will also send like a little nuclear reactor, which does not sound that attractive, but who knows? We went from data centers on the moon to data centers in space. And I think the next evolution might be data centers on Earth. What do you think? We have them. And it's precisely because we can't get enough power in countries like ours that you have all of these other ideas coming. Wonderful. So part of what's driving that is people think that it's because of SpaceX's IPO and they're going to need extra money and that kind of thing.

3:14What do you think? I think it's more to do with influence capacity. What is pretty remarkable about this, moment is that there is more demand for AI and AI products than people can reasonably produce. That's the reason why there is so much investment in things. Similarly, people are doing calculations like if every human being had a chat GPD-like product that they could use, how much inference capacity will you need? And what's also cool about this moment is that the the previous generations of chips are still getting used because those chips are super useful for running smaller models and people are getting utility out of that for simple applications like Weiss transcription or doing classification on large volumes of text.

4:08So the remarkable thing about the current moment is that old models are useful, old chips are useful, new models, the frontier models are letting people do stuff with things like coding agents that they could not dream of doing before. So there's a lot of demand for the new stuff. So it's a time of plenty. In turn, that is leading to people feeling very optimistic about just having more AI capacity. That's what's driving the boom. That's great. I guess to take a step back a little bit further, where are we exactly in the AI super cycle? It's a great question. And my humble answer is, I don't think anyone really knows.

4:49It is early, and it is useful to draw from history about similar kinds of large changes. And if you look back, even things like industrialization, which one tends to think of as like some discrete thing that just happened within a few years in, I don't know, 1800s. something like that actually is a phenomenon that started closer to the 16th century and just kept going for a very, very, very long time. I think AI represents the beginnings of us being able to run truly thinking algorithms, to industrialize thought in a very, very big way. And I think it's very much the beginning. And I think there are lots and lots of applications for things like this.

5:43Everything, as I said, from being able to pretty much create, for example, what's called a classifier or a sentiment detector. You have a piece of text. Is this text angry or happy? Being able to do problems like that without really needing to know any programming. Two, let me think about a complex analysis that I am going to do when I'm given a new piece of information. We get revenue information every single day. We often want to know, why did revenue go up? Did it go down? Is there something else? These are all things that a human with expertise in the area needed to be able to do. Now we can begin to write down some of these things in plain English.

6:23Have models interpreted go take action on that behalf. So I think we are very much in the early cycles of how do you actually use AI to partially assist how we as humans think in every sphere. I think it's just super early. Another way to kind of ask this question is where the value is accruing. So in 2025, value really accrued to a lot of models and hardware. Where do you see the value accruing in the next year and afterwards? You used an important word here, right, which is accruing. I think we should first start with where is value being created at scale. Value is being created by all the users of ChatGPT, nearly a billion of them.

7:15And it is getting created in completely surprising ways. My wife and I moved into a new home. We don't know anything about it. I'm trying to reprogram the gate because I can't get it to open. and normally it's like a super painful process where you like brush off things, try to see what the model name is, go inside, look for the manual, hope you can understand it, come back, try and program it. Instead, I took a picture with ChatGPT and said, hey, how do I program this thing? Tells me what the model is, what button I'm supposed to press, what I'm supposed to do on my remote. Huge amounts of value being created right there.

7:56Value is also being created with all of the people that are using coding agents to write software faster. Absolutely, there's a huge amount of value being created there. In the world of enterprise, big value is being created in areas like customer support. Interactions that are particularly amenable to an AI model saying, I'm going to go take the context of hundreds or thousands of conversations that have happened before, and I'm going to use them to drive and help new conversations. So there are these kinds of applications that are driving the bulk of where value is being created on the AI side.

8:38We are also beginning to see the power of AI in things like Snowflake Intelligence, where my sales team, 5000 plus people, rather than use a series of dashboards to get our data, they just ask questions. before I show up for a customer meeting, it's much easier for me to ask like a six-sentence question about what do I want to learn about this customer than it is for me to somehow stitch that together with eight different dashboards. So these are all places where value is being created. In terms of where is value accruing, definitely value very firmly is accruing with NVIDIA. Just look at their market cap.

9:18It's accruing with the hyperscalers. They simply cannot build enough data centers, buy enough chips, and satisfy all of their demands. And value is accruing in areas like the model makers themselves, as in their valuations, are going up. But I think this is where one has to then think about the second and the third order effects of will open source models catch up? How much of a lead, a moat, as it were, will the model makers have? It's very clear that it's not all that easy to make chips like NVIDIA. It is very clear that running operations as large as what the hyperscalers run for renting out compute, that's not that easy.

10:08It's very hard for a fourth or fifth hyperscaler to come up. I think beyond that, it is clear that OpenAI and Anthropic have created a ton of value in creating these frontier models that are better than anyone else's. But I think an open question there is how quickly will others catch up? I think that's part of what is quite open right now. In 2025, over 75, like 70 to 75 percent of the gains in the stock market were from AI and AI adjacent companies. Yeah. Around six trillion dollars. Yeah. And so given that those were mostly NVIDIA, Meta, big name stocks, Alphabet, some people are pointing to the application layer, some of the categories that you just mentioned.

10:58Are there any categories within that that you think are underrated? I think there's still an open question about things like the modes that traditional application providers have. I think the jury's out. I think they have absolutely a lot of value, but there is also going to be a lot of disruption for many applications as we know it. What happened in the first decade of the cloud computing revolution is that things like enterprise SaaS software had a Cambrian explosion. There's just an application for every single niche that you can think of. And there is an amount of fatigue in the enterprise when trying to manage several hundred applications, their own licensing models and things like that.

11:46I think even in the world of data, there was a very large explosion of different kinds of tools for every stage of the data lifecycle. So at Snowflake, for example, absolutely, we want to be there for our customers at every stage of the data lifecycle. so that they don't have to stitch together lots of complicated individual applications in order to get some job done. I think what AI does to the application space is very much something that is open. I think there's lots of opportunity and lots of disruption. Of your customer set, who have you seen just outperform in their embrace of AI? Obviously, they're using you with data and increasing their capabilities.

12:33So what is a good case study there? Tech companies are often ahead when it comes to using AI. It makes sense that Snowflake is using AI because we are a tech company. I think we have seen very good results with customers like ServiceNow, with DocuSign, with Zoom. These are all folks that are well-versed in tech. unsurprisingly the place where AI is making a big difference is in different kinds of financial services companies. Now there are different categories of financial services companies subject to different amounts of regulations. Asset managers are less heavily regulated compared to banks and so they tend to be leading when it comes to adopting new tools because it can make a real difference.

13:25They are always trying to create that little bit of alpha because they manage assets on behalf of other people. So we have published case studies of companies like TS Imagine, which provides software for asset managers, fully embrace AI. Their CIO waxes really lyrical about agents. Yes, yes. Cool names like Taya that he created on top of Snowflake Intelligence that do a lot of 24-7 support for their various customers. We are also pleasantly surprised by how much the healthcare sector is adopting AI. They drown in a sea of data, and so a lot of those companies are actually adopting AI because it can make a real difference.

14:12Imagine cases like this one. A lot of medical providers have a huge amount of clinical data, let's say notes written by a doctor whenever you go visit them. And previously, if you wanted to answer questions like what fraction of people had flu-like symptoms in a particular month, that would be a project. Right now, it's a SQL query that you can run that's going to rip through all of these notes and kind of give you an effective answer. So we are seeing a lot of adoption. I'm often shocked that even now when I go visit doctors, they are very happy and proud about taking notes on their phone with spoken voice, which is pretty cool.

14:56So that's a sector that's actually leaning in pretty heavily. Sorcery is brought to you by Brex, the financial stack trusted by more than 30 ,000 companies, including one in three venture-backed startups in the U.S. Nearly 40 % of startups fail because they run out of cash. Rex is literally built to help founders avoid that. Unlike traditional banks that let your money sit idle, chipping away at it with fees, Rex's designs help you spend smarter and move faster. Their all-in-one solution combines checking, treasury, and FDIC protection into one powerful account. You can send and receive money globally at lightning speeds, get 20 times the standard FDIC coverage through their partner banks, and even high yield from day one.

15:39with same day and even same hour liquidity. Access your funds anytime. Companies like Scale AI, DoorDash, Service Titan, HIMSS, Anthropic, Flexport, Robinhood, and Plaid trust and use Brex. Start today at brex.com slash sorcery. That's B-R-E-X dot com slash sorcery. So what is Snowflake's role in the AI super cycle? Snowflake at its core is a data platform. We make it easy for you to bring data typically from other systems of record and be able to analyze them. And so people do all kinds of analysis. For example, in our own case, we use Salesforce as our CRM solution, but we bring that data into Snowflake so we can figure out what's the rate at which we are acquiring new use cases.

16:32what are characteristics of new customer logos that we are winning. We bring all of the data into Snowflake. So we are fundamentally about make data easier to analyze. And when we think about AI, we think of AI as a massive accelerant to this data cycle. So what this means is instead of bringing in data into Snowflake and going through what is a fairly painful and laborious process of setting up dashboards, of basically needing an army of analysts in order to provide you with an insight. We can provide you with a conversational interface to this data that lets you ask questions in natural language about any aspect of the data that you have in Snowflake and be able to provide that for you.

17:22VCAI as an extension of our data platform capabilities. We define our mission as a company to let every enterprise realize its full potential through the use of data and AI. And so that's very much what we aspire to. A surprising side effect of AI and Snowflake is AI is making it much easier for our customers to set up Snowflake to be able to do things like configure agents. It's very meta. You want to create data agents on top of Snowflake because it can let end users get to data faster. But the analyst or administrator can use AI now to make the process of setting up that agent be a whole lot faster.

18:10That's something that I'm really excited about because I think that coding agents can massively accelerate this very painful process of bringing in data, stitching it together and moving it from raw to consumable insights. Coming into this role as CEO, you've been here almost over two years now. What has been the biggest surprise for you leading through one of the most volatile environments? I'm very pleasantly surprised by how quickly a pretty large team in dozens of countries has adapted to a very new environment of operating. Two years ago, AI wasn't nice to have, but no one thought that it would affect their day-to-day life.

19:00And yet, what we expect, for example, I expect our sales team to be able to show off Snowflake Intelligence, ideally on their phone, every single day. That was not something that I thought would happen all that quickly. Similarly, solution engineers went from people that clicked things on screen and maybe wrote a SQL query or two to actually being really good and proficient with coding agents in order to deploy actual solutions to problems for our customers. Similarly, our engineers have gone from, we are a database company, we make 24-month plans. Database companies love making long plans. Two, I guess we'll figure out what we do next week and act week on week.

19:45I think that's been a very pleasant surprise for what I said. It started as a database company. Companies don't change that quickly, but I think the company has transformed itself very, very quickly. It doesn't come easy. It is hard. But I think that's been very positive. And personally for you, what has that been like? How has your leadership changed? Is it harder? Is it, you know, easier than you would imagine? Change is always hard. I think we...

20:20Snowflake, through an act of sheer conception from Benoit and Thierry, the founders of Snowflake, created a product that was years ahead of what anyone else had imagined.

20:34And that made Snowflake a very, very confident company. Two, three years ago, it was clear that data was going to be affected in a very massive way by AI. And so two years ago, there were doubts. Hey, will we make this transition? Can we also be good at AI? This feels like something that we hadn't done as well on. But in a number of these situations, you have to try and you have to prove yourself. And I think that's been positive, but it's also been very hard. And right now, for example, getting the whole company to think iteratively. because the way of deploying products today is completely different from what we did at the beginning of last year.

21:28At the beginning of last year, if you ever wanted a product to get deployed, you needed to have a perfect web UI that would walk people through how to configure something, help them with mistakes. You'd have to write long manuals for how you go about configuring something. Today, the expectation is that there's going to be a smart person that knows how to use coding agents, is very comfortable doing it, that can just deal with APIs, which are super abstract concepts. It's how you talk to some server, but they can get a lot of work done. It's a completely different way of developing product. Making changes like that to how people think, to how people act every day is really hard.

22:07It's hard for me because I went from somebody that had a pretty good grasp of how tech worked, of how cloud computing worked, of what was fast and what was slow, and what are qualities to look for in engineers and leaders, to this is a whole new world. And if I don't pay attention for three months, I'm really behind when it comes to what is possible with AI. Honestly, I think it's a little bit of a terrifying moment for every technologist that there is, because there is a lot of Alice in Wonderland here, which is you have to run as fast as you can to kind of stay at the same place. Turing is training the next generation of AI with tasks that require real expertise and real world judgment.

22:54That's why companies like NVIDIA, Anthropic, Salesforce, and Gemini partner with Turing. Turing builds realistic reinforcement learning environments and data systems based on real operational traces, the kind of infrastructure frontier labs need to train superintelligence. Visit Turing.com slash S-O-U-R-C-E-R-Y. You come from a venture background, a tech background. Snowflake was incubated and created out of Sutter Hill. That's right. How involved is Sutter Hill to this day? Well, Spicer's still on our board, Still as opinionated as ever. Love that. You know, Mike is a technologist. Super passionate about how, where the world is going.

23:39Super passionate about the opportunity in front of us. This is a moment where it feels very much like a company has only two outcomes. It's either is going to become a trillion dollar behemoth or disappear without a trace. and so he is super passionate about the power and impact that things like AI are going to have. And so we have lots of fun discussions in that very room. That's our boardroom. Wow. Should we go in there? Should we interview in the boardroom? You can pretend to be Mike and ask Mike-like questions. All sit at one end of the table, you sit at the other. No pressure. um and then so i guess you brought it up how are you going to become a trillion dollar company um there's a formula i'm joking i think um the answer is honestly compound growth compounded growth i was fortunate to be part of another trillion dollar company where a product that i was personally leading search ads went from a billion and a half,$1.6 billion in revenue in 2003, the day I started as an engineer in that team, to making$100 billion the year that I left, 2018.

25:07And so I've watched that transition. In fact, funny story, Eric Schmidt, our CEO at the time, made me and a few other folks write a$100 billion revenue plan. This is like 2007, 2008. We all thought that this was the funniest thing ever. Because the idea that one company could make$100 billion just sounded so preposterous. But the magic of Google is 35 % odd growth compounded over and over again. And so that's what it takes to become a$100 billion revenue company, to become a trillion dollar company. those things are better off as kind of visions. I tell our team here, you're a great company. We want to be an iconic company.

25:59And that requires a lot of hard work year in and year out. For pretty much the entirety of my stay at Google, we took what's called an OKR, an objective, to increase RPM, which is how much revenue we made for 1 ,000 queries by 5 % every quarter. Every single quarter during my time there, we took that as the objective. And that's the reason why you get to have these compounded gains of 35%. You do the math, 1.35 raised to the power of like 12 or 15, very big number. That's what it takes for Snowflake to be an iconic company. But you need to leave it as like the distant North Star over there and focus on getting stuff done today.

26:47I'm really curious, like, because we're in such a competitive environment, you alluded to this a bit, like, culturally, it's totally different, and even from, like, the beginning of last year. So how do you keep the teams motivated and kind of able to weather the storm with you as it changes so much? You have to embrace change. And you also have to find people. There are a class of people that thrive under change. You have to make sure that they are in positions of being able to have an impact. When we wanted to roll out coding agents, for example, to our solutions team, we identified 35, 40 odd people that were naturally inclined to want to go learn, to want to go tinker.

27:39You hold them as exemplars to the rest of the team and help spread the message, help spread change through. Winning over iconic people with extra effort makes a huge difference. I joke to people that I could scream from the rooftops to all my engineers about how they really need to use coding agents. It's not going to have that much impact. They're going to go, like, what does he know about software engineering? On the other hand, Benoit, our iconic founder, fell in love with coding agents. And Benoit is a truly religious figure. When he believes in something, trust me, you're going to hear about it.

28:19And every engineer that he met heard about the impact that these coding agents had on Benva's own day-to-day work. That had a huge impact. So you also need to be strategic about who do you have representing change. And the more you can have naturally influential people represent change, the easier it's going to go. You mentioned a couple characteristics for talent. What are the key traits you look for? in terms of like qualities that I look for in people that I want to bring into the team I look for I mean you need a certain amount of subject matter expertise before our conversation I talked to you about how hard it is to just become skilled at something sometimes you have to spend 10 plus years to get like really good at something stuff is hard and after a certain age you cannot get good at something.

29:16It's like, I'm not going to become a concert pianist like ever. It's just, it's gone. But even when you have the background to get really good at something, you need to be spending a lot of time at it. So I look for drive. I also look for malleability. I'll often ask people, tell me things that you have changed personally in your life. Everybody will give you a good story about what they did with work because they know that that question is coming and they'll say oh Srila I did this this project to make these people more efficient and I use this chatbot over here so I'll practice answers I don't like them I'll ask them questions like tell me how you change yourself like show me that you're malleable in how you think in a professional setting in a personal setting to me the combination of drive and malleability those Those are the prized qualities that set the truly amazing people apart from everyone else, especially at a moment like this.

30:17My kids are two years apart, and when they were growing up, even in their teenage years, you could tell the difference between the older one and the younger one. The younger one always used voice to ask questions off of then chatbots like Google Home, while the older one just like never quite wanted to do that. There are all these age characteristics that go on, but the truly amazing people, they like break through all of these kinds of just learned behaviors and adapt and change. So drive willingness to change. One of the iconic figures of Snowblake is Frank Sweetman. What are the biggest lessons you learned from him?