Snowflake Data Cloud Summit 2024 | John Adams | VideoAmp & Jeff Hollan | Snowflake

Clips
News
More from Snowflake Data Cloud Summit 2024

John Adams

VP of Architecture

VideoAmp

Jeff Hollan

Head of Applications and Developer Platform

Snowflake

Four insights you might have missed from theCUBE’s coverage of Snowflake’s Data Cloud Summit

The Snowflake Data Cloud Summit in San Francisco last week offered an in-depth look at how artificial intelligence, data and open source are continuing to reshape the enterprise world.As Snowflake Inc. moves from its Data Cloud into the AI Data Cloud, the company is seeking to write its next chapter under the leadership of new CEO Sridhar Ramaswamy, supporting an AI-driven environment where customers are demanding data simplicity and value for the business

Unlocking the power of data: How Snowflake and VideoAmp are transforming media measurement technology

The data management landscape is undergoing a revolution, driven by advanced technologies that promise unprecedented efficiency and security, especially when it comes to media measurement technology.Companies are now capable of processing enormous datasets swiftly and securely, paving the way for more informed decision-making and enhanced operational capabilities. The rise of platforms that integrate seamlessly with popular programming languages and containers is redefining how developers manage data, emphasizing ease of use and scalability.Snowflake’s Jeff Hollan and VideoAmp’s John Adams talk to theCUBE about media measurement technology

play_circle_outline VideoAmp Leveraging Snowflake: Ensuring Trust and Accuracy in Data Measurement for Media Agencies and Advertisers

play_circle_outline Snowflake's Revolutionary Approach: Expanding Developer Attraction, Evolution, and the Game-changing Benefits of Snowpark

play_circle_outline Benefits of running processing within Snowflake, including governance and security

play_circle_outline Enhancing AI-Centered Enterprise Applications: Snowflake's Leap Forward in Data Clean Room and Security Governance

Info

John Adams | VideoAmp & Jeff Hollan | Snowflake

John Adams

VP of Architecture VideoAmp

Jeff Hollan

Head of Applications and Developer Platform Snowflake

search

Rebecca Knight

>> Welcome back everyone to theCUBE's live coverage of the Snowflake Data Cloud Summit here at the Moscone Center in San Francisco. We are wrapping up four days' wall-to-wall coverage on theCUBE. I'm Rebecca Knight, your host, sitting alongside my co-host and co-analyst, Dave Vellante. It's Developer Day. The hoodies are out.

>> Been quite a week, hasn't it?

Rebecca Knight

>> It has indeed.

>> Yeah. We went from suits to hoodies. I love it.

Rebecca Knight

>> I like this crowd, I've got to admit. I've got to admit. So with that, I'd like to introduce our next two guests. We have John Adams, the VP of architecture at VideoAmp. Thanks so much for coming on theCUBE, John. And Jeff Holland, head of developer tools at Snowflake. Thank you both for being here.

>> It's great to be here.

John Adams

>> Yeah, thanks for having us.

Rebecca Knight

>> I want to start with you, John. Tell our viewers a little bit about VideoAmp and how you work with Snowflake.

John Adams

>> VideoAmp is in the media measurement, so content and ad measurement space. We leverage a lot of large data sets to go ahead and create something that's actionable for our publishers and advertisers to transact upon. They want to know who's watching what, what those demographics are, and they want to be able to trust the data. They want to be able to know that when we say this many people saw it, this many people saw your ad, that there's not overreporting, there's not underreporting. And we leverage Snowflake for that. There's no other technology we've been able to find that would allow the volume of data that we have to be so actionable, so easy to process. So yeah, we're good partners of Snowflake.

>> Can I give you my card? This is the third guest-

Rebecca Knight

>> I know....

>> that I'm a prospect of.

Rebecca Knight

>> He is doing... yes.

>> Seriously.

John Adams

>> Yeah, absolutely.

>> We do video. That's what we do all day every day. And it's a nightmare.

John Adams

>> There we go.

>> The reporting of one company and another company, different platforms... you never know what you're looking at. There's no standard definitions. We need a semantic layer for video.

John Adams

>> I would agree.

Rebecca Knight

>> John, he's there for you. No, but all kidding aside, at a time in the attention economy where there is so much pressure on these media organizations to get eyeballs and then to make sure that those eyeballs are, in fact, recounting them once. It's really true. So you're doing some important work. So talk a little bit about the work you're doing.

John Adams

>> So as VP of architecture at VideoAmp, I'm over our security, I'm over our infrastructure. I'm over platform like APIs, things that we're going to use across all of our products. So the products we're building... we're building planning tools for media agencies. We're building planning tools for advertisers. We're building measurement tools for both sides. And a lot of the work we have to do is... there's been a lot of regulations, there's been a lot of work around privacy. We want to make sure that you're not sharing individual's data with people that the individual hasn't opted into sharing. So we're heavily using Snowflake for data clean room technologies where two parties can come in, they can transact on their data together, but not actually have access to each other's data. So they're able to apply our methodology in the middle to actually surface measurement, surface... what's the reach, what's the frequency? What are you actually seeing? Who's viewing your content? Who's viewing your ad? And that's all being built on Snowflake.

>> So Jeff, what's your perspective on the evolution of Snowflake? Went from hardcore database people and now you're attracting application developers, you're getting more folks in the company that understand the application environment. How have you seen that journey? Actually, what's your background? Because you're one of those folks that has an affinity developers, and how is the company's sort of culture evolving to support developers?

>> Along with the same types of thing that John's been doing in VideoAmp, started obviously with this core of data, making sure that, "Hey, we need to be able to process these massive volumes of data so powerfully," but in the world we're moving into and have been in for a while, data on its own needs so much more around it. And so Snowflake has been expanding for many years now into things like the Snowpark platform to say, "Okay, let's let you bring Python, let's let you bring containers, let you do more both in terms of giving you more flexibility in the processing that you can do and bringing those applications to the data." And what we see and what gets us very excited is that because those volumes of data are so massive, if you can bring those apps to the data, you just get these massive price performance improvements. And then at Snowflake, one of the big things we care about on product is easy to use. So it's, "Hey, bring those things there." You mentioned the background, what brought me to Snowflake. Prior to Snowflake, I actually spent about a decade at Microsoft in their developer cloud services, so things like serverless functions, serverless containers, working closely with GitHub and VS Code and those things. So I was much closer day-to-day working with app devs. But even then, at the time, this was a few years ago now, we're seeing the types of apps that people were building, they're bringing serverless functions to, it wasn't about generating the data as much as, "Hey, we have these massive amounts of data that exist. How do we make sense of it? How do we provide insights around that data?" So I was like, "Oh, there's clearly this shift happening." We were talking even then at the time of data apps and this growth of data apps. So that drew me to Snowflake of like, "Okay, I got to go to where the momentum is going and help developers be productive in this world around data applications, as well."

>> So what's your brand promise to developers? How do you think about that?

Rebecca Knight

>> It's very similar to the one that we've talked about with the pieces on AI all up. It's the easy to use, efficient, and secure. Those are the three big pillars that even the developer pieces are built on. So it's, "Hey, you're doing development? We don't want you... you've got your cool Python script working, your JavaScript working. I don't want you to have to worry about fiddling with infrastructure, managing VMs, patching VMs. We're just going to run it at scale." That's part of the efficient piece. "You're working with these massive volumes of data. Don't worry about your data having to leave, go somewhere else. Do the processing here, deal with the latency and the cost of pushing and pulling that data, different pieces," and then of course secure. You want to make sure both the code that you're writing that you can trust that code running next to your data. So there's a ton of effort there. And then just having those security promises that Snowflake is known for. So it aligns a lot with the messages we've been talking about all week.

>> And John, is Snowflake from your standpoint keeping its promises?

John Adams

>> So historically, we ran a lot of workloads on Spark. We've been around for a long time. We were using Snowflake and Spark, but a lot of our processing for actually taking the data that was being stored in Snowflake or being stored on HDFS or other storage systems, we would go ahead and process this outside of Snowflake. And I spoke earlier this week about actually moving that processing into Snowflake because of Snowpark. So we were able to take a lot of our processing that we would be running on other systems, moving it in there. And the reason we did that is governance. We don't have to move our data around. We don't have to have exfiltration of our data. I make my security guys very happy. They don't have to worry about data leaving the controlled environment to be processed. Instead, it gets processed right there. It's being done cheap, fast. It's kind of a win-win for us.

>> Okay, I'm going off script. Last year at Summit, we heard some customers say, "Well, we're doing a lot of the hardcore data pipelining, data engineering, data science work outside of Snowflake." Really? Why? "Too expensive." Okay. I talked to Frank Slootman offline, I'm like, "We're hearing this a lot." He goes, "They're wrong."

John Adams

>> They are.

>> I'm like, "Well explain that to me, Frank." He goes, "Think about it. We do streaming, we do really sophisticated analytics. We're bringing in AI. We're doing data engineering and data science work all in one place. And when you add all that up, it's going to be less expensive." And we started talking about it and we're still trying to squint through this. So I want the customer's opinion. Snowflake bundles in the AWS costs, right? When you go outside and you're just using Spark, you're paying AWS separately so it's all not in one bundle. So somebody's got to do the work to add it up and actually do the out-of-pocket costs, but then there's all the other costs of securing and governing. So you said he's right. Explain from your perspective.

John Adams

>> So one of the interesting things. When we had a large data center, so we had two petabytes of data sitting in our data center. We would've go ahead and take a lot of data that we were mastering in Snowflake... we would have to pay egress costs. So a lot of the processing that we were able to do and spend a couple million bucks a year on this data center, we're now running that workload on Snowflake for cheaper than our egress costs from AWS to our data center. So the cost of tokens to actually ship this data was more than what it's taken to actually process it in Snowflake. And one of the big reasons for that, people look at Snowflake's billing model and they say, "Oh, you're paying a lot per credit," but you're paying for the seconds you're using. So if you can actually have something that would traditionally take you 20 minutes on Spark plus the spin up time, plus the spin down time, plus the data transfer time, and you can run that in two, three minutes on Snowflake, I'll pay a multiplier on that time because Snowflake is eating the cost of having these servers available for you. They're paying Amazon to have the infrastructure ready for you to run, but I don't have to worry about paying for that time. I just pay for what I'm using.

>> So because the compute engine is so efficient, you're saying?

John Adams

>> It's efficient and it's ready for you to actually query.

John Adams

>> And John's not alone in this, either. We've actually done an analysis of over 30 customers over the last year who've gone from Spark to Snowpark. And on average across them, we see 35% less expensive. And that includes calculating the infrastructure cost as well and three to four times faster on average. And the reason is, there's the data movement. If you think about it too, the amount of steps geeking out a little bit on developer day, you've got to take data that is stored here natively in Snowflake with our engine. To go process it anywhere else, you've got to deserialize the data, push it across an egress pipeline, serialize it over here, process it there, and then send it back again and back and forth and back forth. All those things are just steps that when you run it natively inside of Snowflake with Snowpark, all that goes away so you get this cheaper, faster, more efficient piece.

>> And I want to understand that. So what you just described is labor-intensive, so it's human labor that is required to do that? Or are there other costs?

John Adams

>> It's even heavier just on tech. It's like laws of physics, almost. There's laws of physics advantages on top of the fact that the Snowflake Elastic engine, one of the core pieces when we were building Snowpark, and the reason we built out Snowpark the way that we did is, "Look, we have this core engine that Benoit and Terry and founders have built together that does this incredible elastic scale. That's the exact same engine that's processing the Python workloads." So just pushing down to that gives benefits, and then there's just laws of physics benefits in terms of... we don't move the data anywhere. We see what you want to do. You described it in Python, and we'll just push that down to the data. So just keeping the data in there, reducing those steps, we optimize the full end to end, you get these massive benefits.

>> So I'm sorry to keep pushing on this, but I really want to understand it. So if I understand it correctly, the business cases would be around time to value is much greater, and it's somewhat soft, but that's why... same thing with the cloud. Not cost, it's the agility. I can do stuff that I couldn't do before. Is that correct?

John Adams

>> I would say I think the one that gets the most people excited... and John, I'd love to hear your perspective though. It's the cost though. All up, we've seen time and time again, you're paying a certain amount of cost to do processing in Spark, wherever you have that Spark to do. There is a human cost to this. Like, "Oh, I've got to go make sure I've ordered this right and do I have my Z partition set up the right way? And I've got to maintain these things," that is a cost, all up. But if you just look at, "How much money are we spending to get our data to a state that we can act on it?" Snowpark can get you just lower costs all up. You'll just spend less money and do it faster. It's a win-win.

>> Weigh in on this, if you would.

John Adams

>> So interesting thing with Snowflake or Snowpark or Snowflake in general is the time it takes to do something and the costs are inversely proportional, but you're only paying for the time that you're using. So if I want to sit here and I'm running, say, a medium warehouse and I'm like, "You know what? I want this to be half the time, half the wall clock time." I'm going to go ahead and make it a large, I'm going to get it done in half the time, and if I'm efficient with how I'm managing my warehouse, I'm going to pay the same amount of cost. And you really can't do that in a lot of other systems. Other systems, you're going to provision new cluster to do new things. You're going to have a lot of fixed costs. So we are seeing that. We are seeing that we can get fast, we can get cheap.

>> I've done a lot of TCO analysis in my day and I'll give an example. So when you do Oracle TCO analysis, the software licenses everything. It's not usually the case with other systems. It's usually labor cost or a combination of things. My question is, in that pie, how much of the blame pie goes to egress versus those others? Is it more balanced or is it... if egress were no cost, would that equation flip?

>> If I look at cost all up, I don't think the equation flips at all. I haven't done the deep... I don't know exactly. You would know more about your bill. I don't know specifics.

John Adams

>> . For us, no. The egress cost was... I mean if we look at what we were spending on egress versus what we were spending for our data center, just the process of the Spark workloads, it was drop in the bucket. So when I say that Snowpark's cheaper, I don't know if I should share the numbers, but I shared it on the presentation. So we have about 90% cost savings.

>> 90? 9-0?

John Adams

>> Yes. 9-0.

>> So 10% of?

John Adams

>> Is what we're spending.

>> Okay.

Rebecca Knight

>> Wow.

>> That's great. You guys are passing my tests here. Since you told me you were a former CISO in a painful life ... so I want to ask you about the developer perspective on securing... shift left, sort of the bromide, and how that experience is different or is it different in Snowflake? So it's like the cloud became the first line of defense, but then the developers were asked to secure the code and they're like, "All right, well, if I have to." And so what's your perspective on the developer experience in the context of having to worry about all that security and governance, and how is the experience inside of Snowflake?

John Adams

>> Well, it's interesting. It's a lot easier in Snowflake than it should be. I think it's because Snowflake follows pretty normal standards. You can lay out a nice RBAC structure really easily and it's transparent to the developer. Then you can sit here and Snowflake introduced the thing a few years back which was secondary roles, which allows you to compose all of your access while you're going ahead and doing reads and writes, and you don't have to worry about what role you're part of. So when I'm managing, let's say, a provisioning of a user and I want to give them a bunch of different roles, the developer doesn't even have to switch between them. They have a composition of all the access and I can take away access very easily without changing their workflow at all. And then when we actually think about securing Snowflake's account, a part of provisioning a Snowflake account, as you're working with your sales engineer, you're spinning this stuff up, they're going to go ahead and walk you through, "Hey, let's enable MFA. Let's make sure we have private link enabled. Let's make sure that we harden your system." It's one of the checklists of onboarding on a Snowflake anyways. For the developers, it's not much. They go, they click into Okta, they get into the interface, they'll probably be on their VPN if they're using private link. They're going to be secure, nothing's leaving their network, and then they're going to go ahead and use the tool. Now if they don't have access to a table, they're not going to see the table. And then you have additional controls that have been added, and I'm going to ramble. I'm sorry. This is my passion.

>> No, this is beautiful. I'm loving this.

John Adams

>> The controls, you start having data masking policies. You have row-level access policies. You have all these additional controls. You have aggregation policies. You start putting these things onto the tables, subsequent views and whatnot down there further right in the pipeline that are built on top of this, inherit the same capabilities. So I go ahead and maybe I have access to a table that maybe I shouldn't, but I definitely shouldn't be seeing the sensitive data in that table. As long as there's an access policy on that or a data masking policy on that, I'm not going to see the data. It's going to be obstructed or obfuscated away, it's going to be encrypted, it's going to be masked. And if two of my account reps are using the same share table and they're going to maybe have different access to different subsets of accounts based off the region, they're going to only see... I put a row-level access policy on there? Any views built off of that table, anything downstream built off of that table, they're only going to see the customers they actually have access to. So not just from a developer perspective, but also a user perspective, it becomes super safe and transparent. And a term we used earlier that you hear from Benoit is easy. And so yeah.

Rebecca Knight

>> Simplify, simplify, simplify.

>> I'm dominating the conversation here, Rebecca.

Rebecca Knight

>> This conversation is getting very geeky, so yes, let's talk about the future. I'm interested in both of your perspective on this new era of generative AI that we're living through and how it will change how developers do their jobs. I'm going to start with you, Josh.

>> Sure. I think there's two vectors to it. One is that generative AI really gives developers a new set of superpowers. This is what I love.

Rebecca Knight

>> They weren't powerful enough.

>> Yeah, I know. They didn't have enough power already. So it's the copilots, the editing, so many different flavors. And Snowflake is embedding AI into all these pieces to make it so that, again, if we talk about some of the time required, we'll just have AI help augment. So that's one side of those investments that I think is very, very exciting and disruptive. But the other part too is just the type of apps. The next generation of applications that are going to be providing value are all going to be centered around AI and especially around enterprises. Those types of apps are always going to need some systems of record, some valuable data to make them powerful. So I think it's just going to accelerate the shift into saying, "Look, it's not data and apps. It's really like these data to the apps piece." There's no other way to do it in generative AI that I see. So that to me is very exciting and why I get very excited to be able to work at Snowflake, to have that ability to have a platform that's so purpose-ready for this new generation of generative AI apps.

>> What would you like to be able to say 12 months from now? So I go back to last year, it's a Summit. Snowpark Container Services was the big announcement, at least I thought that was the big one. And I said, "All right, well, I want to see developers actually using that next year and I want to see more folks in the ecosystem." We're seeing that. What do you want to be able to say 12 months from now that you can't say today? AKA, what do you want to see Snowflake deliver?

John Adams

>> Oh, that's a tough one.

John Adams

>> Let me get ready to write it down.

John Adams

>> I think one of the big ones is, like I said, we're big users of the data clean room. Data clean room is a native application. It's great. It gives us the ability to have this trusted relationship with our partners, and I would like to see data clean room and any security governance capability, private link, whatnot, when something goes GA, it's there. So we already saw that today or this week with Snowpark container services being available in native apps. There's so many people in... data privacy is so important. Data stewardship is so important. I'd want to make sure that the data clean room fully supports all the capabilities of Snowflake, Snowpark right off the bat.

Rebecca Knight

>> Got that Jeff?

>> Yeah, I got it. Yes.

Rebecca Knight

>> All right. We're looking forward to the next summit. Excellent. Thank you.

>> I want to just wrap because we're done.

Rebecca Knight

>> Go for it. We're done. This is it. This is it.

>> We're done. Right. So we have to say, this has been a fantastic show. Thank you guys for helping us wrap. To me, my takeaways are the plumbing is actually really robust here, and Snowflake is building on top of that. The Polaris announcement, open sourcing, the technical metadata catalog, and then enticing people to come back into Snowflake with Horizon and all that wonderful governance is going to be interesting to see how that plays out. I think that building... because they have such robust plumbing and a really good compute engine, they can now build things on top of it. They can bring in gen AI, Cortex, the NVIDIA stuff with NIMS, kind of supercharges Cortex. Then of course the applications piece is really exciting, the vision of iPhone for data apps. There's a new set of applications coming. Data is the new development kit. John Furrier said that 10 years ago, but it's actually now happening. And it brings Snowflake into a whole new realm with new competitors. It's going to be really interesting to see how this plays out. I'm super excited for you guys.

>> It's fun to be on, and I'm excited to share whatever in the next few months of how that journey's been going. So yeah.

>> Great. Thank you for hanging out this week.

Rebecca Knight

>> No, thank you. It's been a great show. As you said, a great show. I've really enjoyed myself here at Snowflake. And thank you for watching alongside of us here at theCUBE. This has been the Snowflake Data Cloud Summit 2024 at the Moscone Center. I'm Rebecca Knight for Dave Vellante and all of us here at theCUBE thank you for tuning in. You've been watching theCUBE, the leader in enterprise tech news and analysis.

John Adams | VideoAmp & Jeff Hollan | Snowflake

search

Rebecca Knight

>> Welcome back everyone to theCUBE's live coverage of the Snowflake Data Cloud Summit here at the Moscone Center in San Francisco. We are wrapping up four days' wall-to-wall coverage on theCUBE. I'm Rebecca Knight, your host, sitting alongside my co-host and co-analyst, Dave Vellante. It's Developer Day. The hoodies are out.

>> Been quite a week, hasn't it?

Rebecca Knight

>> It has indeed.

>> Yeah. We went from suits to hoodies. I love it.

Rebecca Knight

>> I like this crowd, I've got to admit. I've got to admit. So with that, I'd like to introduce our next two guests. We have John Adams, the VP of architecture at VideoAmp. Thanks so much for coming on theCUBE, John. And Jeff Holland, head of developer tools at Snowflake. Thank you both for being here.

>> It's great to be here.

John Adams

>> Yeah, thanks for having us.

Rebecca Knight

>> I want to start with you, John. Tell our viewers a little bit about VideoAmp and how you work with Snowflake.

John Adams

>> VideoAmp is in the media measurement, so content and ad measurement space. We leverage a lot of large data sets to go ahead and create something that's actionable for our publishers and advertisers to transact upon. They want to know who's watching what, what those demographics are, and they want to be able to trust the data. They want to be able to know that when we say this many people saw it, this many people saw your ad, that there's not overreporting, there's not underreporting. And we leverage Snowflake for that. There's no other technology we've been able to find that would allow the volume of data that we have to be so actionable, so easy to process. So yeah, we're good partners of Snowflake.

>> Can I give you my card? This is the third guest-

Rebecca Knight

>> I know....

>> that I'm a prospect of.

Rebecca Knight

>> He is doing... yes.

>> Seriously.

John Adams

>> Yeah, absolutely.

>> We do video. That's what we do all day every day. And it's a nightmare.

John Adams

>> There we go.

>> The reporting of one company and another company, different platforms... you never know what you're looking at. There's no standard definitions. We need a semantic layer for video.

John Adams

>> I would agree.

Rebecca Knight

>> John, he's there for you. No, but all kidding aside, at a time in the attention economy where there is so much pressure on these media organizations to get eyeballs and then to make sure that those eyeballs are, in fact, recounting them once. It's really true. So you're doing some important work. So talk a little bit about the work you're doing.

John Adams

>> So as VP of architecture at VideoAmp, I'm over our security, I'm over our infrastructure. I'm over platform like APIs, things that we're going to use across all of our products. So the products we're building... we're building planning tools for media agencies. We're building planning tools for advertisers. We're building measurement tools for both sides. And a lot of the work we have to do is... there's been a lot of regulations, there's been a lot of work around privacy. We want to make sure that you're not sharing individual's data with people that the individual hasn't opted into sharing. So we're heavily using Snowflake for data clean room technologies where two parties can come in, they can transact on their data together, but not actually have access to each other's data. So they're able to apply our methodology in the middle to actually surface measurement, surface... what's the reach, what's the frequency? What are you actually seeing? Who's viewing your content? Who's viewing your ad? And that's all being built on Snowflake.

>> So Jeff, what's your perspective on the evolution of Snowflake? Went from hardcore database people and now you're attracting application developers, you're getting more folks in the company that understand the application environment. How have you seen that journey? Actually, what's your background? Because you're one of those folks that has an affinity developers, and how is the company's sort of culture evolving to support developers?

>> Along with the same types of thing that John's been doing in VideoAmp, started obviously with this core of data, making sure that, "Hey, we need to be able to process these massive volumes of data so powerfully," but in the world we're moving into and have been in for a while, data on its own needs so much more around it. And so Snowflake has been expanding for many years now into things like the Snowpark platform to say, "Okay, let's let you bring Python, let's let you bring containers, let you do more both in terms of giving you more flexibility in the processing that you can do and bringing those applications to the data." And what we see and what gets us very excited is that because those volumes of data are so massive, if you can bring those apps to the data, you just get these massive price performance improvements. And then at Snowflake, one of the big things we care about on product is easy to use. So it's, "Hey, bring those things there." You mentioned the background, what brought me to Snowflake. Prior to Snowflake, I actually spent about a decade at Microsoft in their developer cloud services, so things like serverless functions, serverless containers, working closely with GitHub and VS Code and those things. So I was much closer day-to-day working with app devs. But even then, at the time, this was a few years ago now, we're seeing the types of apps that people were building, they're bringing serverless functions to, it wasn't about generating the data as much as, "Hey, we have these massive amounts of data that exist. How do we make sense of it? How do we provide insights around that data?" So I was like, "Oh, there's clearly this shift happening." We were talking even then at the time of data apps and this growth of data apps. So that drew me to Snowflake of like, "Okay, I got to go to where the momentum is going and help developers be productive in this world around data applications, as well."

>> So what's your brand promise to developers? How do you think about that?

Rebecca Knight

>> It's very similar to the one that we've talked about with the pieces on AI all up. It's the easy to use, efficient, and secure. Those are the three big pillars that even the developer pieces are built on. So it's, "Hey, you're doing development? We don't want you... you've got your cool Python script working, your JavaScript working. I don't want you to have to worry about fiddling with infrastructure, managing VMs, patching VMs. We're just going to run it at scale." That's part of the efficient piece. "You're working with these massive volumes of data. Don't worry about your data having to leave, go somewhere else. Do the processing here, deal with the latency and the cost of pushing and pulling that data, different pieces," and then of course secure. You want to make sure both the code that you're writing that you can trust that code running next to your data. So there's a ton of effort there. And then just having those security promises that Snowflake is known for. So it aligns a lot with the messages we've been talking about all week.

>> And John, is Snowflake from your standpoint keeping its promises?

John Adams

>> So historically, we ran a lot of workloads on Spark. We've been around for a long time. We were using Snowflake and Spark, but a lot of our processing for actually taking the data that was being stored in Snowflake or being stored on HDFS or other storage systems, we would go ahead and process this outside of Snowflake. And I spoke earlier this week about actually moving that processing into Snowflake because of Snowpark. So we were able to take a lot of our processing that we would be running on other systems, moving it in there. And the reason we did that is governance. We don't have to move our data around. We don't have to have exfiltration of our data. I make my security guys very happy. They don't have to worry about data leaving the controlled environment to be processed. Instead, it gets processed right there. It's being done cheap, fast. It's kind of a win-win for us.

>> Okay, I'm going off script. Last year at Summit, we heard some customers say, "Well, we're doing a lot of the hardcore data pipelining, data engineering, data science work outside of Snowflake." Really? Why? "Too expensive." Okay. I talked to Frank Slootman offline, I'm like, "We're hearing this a lot." He goes, "They're wrong."

John Adams

>> They are.

>> I'm like, "Well explain that to me, Frank." He goes, "Think about it. We do streaming, we do really sophisticated analytics. We're bringing in AI. We're doing data engineering and data science work all in one place. And when you add all that up, it's going to be less expensive." And we started talking about it and we're still trying to squint through this. So I want the customer's opinion. Snowflake bundles in the AWS costs, right? When you go outside and you're just using Spark, you're paying AWS separately so it's all not in one bundle. So somebody's got to do the work to add it up and actually do the out-of-pocket costs, but then there's all the other costs of securing and governing. So you said he's right. Explain from your perspective.

John Adams

>> So one of the interesting things. When we had a large data center, so we had two petabytes of data sitting in our data center. We would've go ahead and take a lot of data that we were mastering in Snowflake... we would have to pay egress costs. So a lot of the processing that we were able to do and spend a couple million bucks a year on this data center, we're now running that workload on Snowflake for cheaper than our egress costs from AWS to our data center. So the cost of tokens to actually ship this data was more than what it's taken to actually process it in Snowflake. And one of the big reasons for that, people look at Snowflake's billing model and they say, "Oh, you're paying a lot per credit," but you're paying for the seconds you're using. So if you can actually have something that would traditionally take you 20 minutes on Spark plus the spin up time, plus the spin down time, plus the data transfer time, and you can run that in two, three minutes on Snowflake, I'll pay a multiplier on that time because Snowflake is eating the cost of having these servers available for you. They're paying Amazon to have the infrastructure ready for you to run, but I don't have to worry about paying for that time. I just pay for what I'm using.

>> So because the compute engine is so efficient, you're saying?

John Adams

>> It's efficient and it's ready for you to actually query.

John Adams

>> And John's not alone in this, either. We've actually done an analysis of over 30 customers over the last year who've gone from Spark to Snowpark. And on average across them, we see 35% less expensive. And that includes calculating the infrastructure cost as well and three to four times faster on average. And the reason is, there's the data movement. If you think about it too, the amount of steps geeking out a little bit on developer day, you've got to take data that is stored here natively in Snowflake with our engine. To go process it anywhere else, you've got to deserialize the data, push it across an egress pipeline, serialize it over here, process it there, and then send it back again and back and forth and back forth. All those things are just steps that when you run it natively inside of Snowflake with Snowpark, all that goes away so you get this cheaper, faster, more efficient piece.

>> And I want to understand that. So what you just described is labor-intensive, so it's human labor that is required to do that? Or are there other costs?

John Adams

>> It's even heavier just on tech. It's like laws of physics, almost. There's laws of physics advantages on top of the fact that the Snowflake Elastic engine, one of the core pieces when we were building Snowpark, and the reason we built out Snowpark the way that we did is, "Look, we have this core engine that Benoit and Terry and founders have built together that does this incredible elastic scale. That's the exact same engine that's processing the Python workloads." So just pushing down to that gives benefits, and then there's just laws of physics benefits in terms of... we don't move the data anywhere. We see what you want to do. You described it in Python, and we'll just push that down to the data. So just keeping the data in there, reducing those steps, we optimize the full end to end, you get these massive benefits.

>> So I'm sorry to keep pushing on this, but I really want to understand it. So if I understand it correctly, the business cases would be around time to value is much greater, and it's somewhat soft, but that's why... same thing with the cloud. Not cost, it's the agility. I can do stuff that I couldn't do before. Is that correct?

John Adams

>> I would say I think the one that gets the most people excited... and John, I'd love to hear your perspective though. It's the cost though. All up, we've seen time and time again, you're paying a certain amount of cost to do processing in Spark, wherever you have that Spark to do. There is a human cost to this. Like, "Oh, I've got to go make sure I've ordered this right and do I have my Z partition set up the right way? And I've got to maintain these things," that is a cost, all up. But if you just look at, "How much money are we spending to get our data to a state that we can act on it?" Snowpark can get you just lower costs all up. You'll just spend less money and do it faster. It's a win-win.

>> Weigh in on this, if you would.

John Adams

>> So interesting thing with Snowflake or Snowpark or Snowflake in general is the time it takes to do something and the costs are inversely proportional, but you're only paying for the time that you're using. So if I want to sit here and I'm running, say, a medium warehouse and I'm like, "You know what? I want this to be half the time, half the wall clock time." I'm going to go ahead and make it a large, I'm going to get it done in half the time, and if I'm efficient with how I'm managing my warehouse, I'm going to pay the same amount of cost. And you really can't do that in a lot of other systems. Other systems, you're going to provision new cluster to do new things. You're going to have a lot of fixed costs. So we are seeing that. We are seeing that we can get fast, we can get cheap.

>> I've done a lot of TCO analysis in my day and I'll give an example. So when you do Oracle TCO analysis, the software licenses everything. It's not usually the case with other systems. It's usually labor cost or a combination of things. My question is, in that pie, how much of the blame pie goes to egress versus those others? Is it more balanced or is it... if egress were no cost, would that equation flip?

>> If I look at cost all up, I don't think the equation flips at all. I haven't done the deep... I don't know exactly. You would know more about your bill. I don't know specifics.

John Adams

>> . For us, no. The egress cost was... I mean if we look at what we were spending on egress versus what we were spending for our data center, just the process of the Spark workloads, it was drop in the bucket. So when I say that Snowpark's cheaper, I don't know if I should share the numbers, but I shared it on the presentation. So we have about 90% cost savings.

>> 90? 9-0?

John Adams

>> Yes. 9-0.

>> So 10% of?

John Adams

>> Is what we're spending.

>> Okay.

Rebecca Knight

>> Wow.

>> That's great. You guys are passing my tests here. Since you told me you were a former CISO in a painful life ... so I want to ask you about the developer perspective on securing... shift left, sort of the bromide, and how that experience is different or is it different in Snowflake? So it's like the cloud became the first line of defense, but then the developers were asked to secure the code and they're like, "All right, well, if I have to." And so what's your perspective on the developer experience in the context of having to worry about all that security and governance, and how is the experience inside of Snowflake?

John Adams

>> Well, it's interesting. It's a lot easier in Snowflake than it should be. I think it's because Snowflake follows pretty normal standards. You can lay out a nice RBAC structure really easily and it's transparent to the developer. Then you can sit here and Snowflake introduced the thing a few years back which was secondary roles, which allows you to compose all of your access while you're going ahead and doing reads and writes, and you don't have to worry about what role you're part of. So when I'm managing, let's say, a provisioning of a user and I want to give them a bunch of different roles, the developer doesn't even have to switch between them. They have a composition of all the access and I can take away access very easily without changing their workflow at all. And then when we actually think about securing Snowflake's account, a part of provisioning a Snowflake account, as you're working with your sales engineer, you're spinning this stuff up, they're going to go ahead and walk you through, "Hey, let's enable MFA. Let's make sure we have private link enabled. Let's make sure that we harden your system." It's one of the checklists of onboarding on a Snowflake anyways. For the developers, it's not much. They go, they click into Okta, they get into the interface, they'll probably be on their VPN if they're using private link. They're going to be secure, nothing's leaving their network, and then they're going to go ahead and use the tool. Now if they don't have access to a table, they're not going to see the table. And then you have additional controls that have been added, and I'm going to ramble. I'm sorry. This is my passion.

>> No, this is beautiful. I'm loving this.

John Adams

>> The controls, you start having data masking policies. You have row-level access policies. You have all these additional controls. You have aggregation policies. You start putting these things onto the tables, subsequent views and whatnot down there further right in the pipeline that are built on top of this, inherit the same capabilities. So I go ahead and maybe I have access to a table that maybe I shouldn't, but I definitely shouldn't be seeing the sensitive data in that table. As long as there's an access policy on that or a data masking policy on that, I'm not going to see the data. It's going to be obstructed or obfuscated away, it's going to be encrypted, it's going to be masked. And if two of my account reps are using the same share table and they're going to maybe have different access to different subsets of accounts based off the region, they're going to only see... I put a row-level access policy on there? Any views built off of that table, anything downstream built off of that table, they're only going to see the customers they actually have access to. So not just from a developer perspective, but also a user perspective, it becomes super safe and transparent. And a term we used earlier that you hear from Benoit is easy. And so yeah.

Rebecca Knight

>> Simplify, simplify, simplify.

>> I'm dominating the conversation here, Rebecca.

Rebecca Knight

>> This conversation is getting very geeky, so yes, let's talk about the future. I'm interested in both of your perspective on this new era of generative AI that we're living through and how it will change how developers do their jobs. I'm going to start with you, Josh.

>> Sure. I think there's two vectors to it. One is that generative AI really gives developers a new set of superpowers. This is what I love.

Rebecca Knight

>> They weren't powerful enough.

>> Yeah, I know. They didn't have enough power already. So it's the copilots, the editing, so many different flavors. And Snowflake is embedding AI into all these pieces to make it so that, again, if we talk about some of the time required, we'll just have AI help augment. So that's one side of those investments that I think is very, very exciting and disruptive. But the other part too is just the type of apps. The next generation of applications that are going to be providing value are all going to be centered around AI and especially around enterprises. Those types of apps are always going to need some systems of record, some valuable data to make them powerful. So I think it's just going to accelerate the shift into saying, "Look, it's not data and apps. It's really like these data to the apps piece." There's no other way to do it in generative AI that I see. So that to me is very exciting and why I get very excited to be able to work at Snowflake, to have that ability to have a platform that's so purpose-ready for this new generation of generative AI apps.

>> What would you like to be able to say 12 months from now? So I go back to last year, it's a Summit. Snowpark Container Services was the big announcement, at least I thought that was the big one. And I said, "All right, well, I want to see developers actually using that next year and I want to see more folks in the ecosystem." We're seeing that. What do you want to be able to say 12 months from now that you can't say today? AKA, what do you want to see Snowflake deliver?

John Adams

>> Oh, that's a tough one.

John Adams

>> Let me get ready to write it down.

John Adams

>> I think one of the big ones is, like I said, we're big users of the data clean room. Data clean room is a native application. It's great. It gives us the ability to have this trusted relationship with our partners, and I would like to see data clean room and any security governance capability, private link, whatnot, when something goes GA, it's there. So we already saw that today or this week with Snowpark container services being available in native apps. There's so many people in... data privacy is so important. Data stewardship is so important. I'd want to make sure that the data clean room fully supports all the capabilities of Snowflake, Snowpark right off the bat.

Rebecca Knight

>> Got that Jeff?

>> Yeah, I got it. Yes.

Rebecca Knight

>> All right. We're looking forward to the next summit. Excellent. Thank you.

>> I want to just wrap because we're done.

Rebecca Knight

>> Go for it. We're done. This is it. This is it.

>> We're done. Right. So we have to say, this has been a fantastic show. Thank you guys for helping us wrap. To me, my takeaways are the plumbing is actually really robust here, and Snowflake is building on top of that. The Polaris announcement, open sourcing, the technical metadata catalog, and then enticing people to come back into Snowflake with Horizon and all that wonderful governance is going to be interesting to see how that plays out. I think that building... because they have such robust plumbing and a really good compute engine, they can now build things on top of it. They can bring in gen AI, Cortex, the NVIDIA stuff with NIMS, kind of supercharges Cortex. Then of course the applications piece is really exciting, the vision of iPhone for data apps. There's a new set of applications coming. Data is the new development kit. John Furrier said that 10 years ago, but it's actually now happening. And it brings Snowflake into a whole new realm with new competitors. It's going to be really interesting to see how this plays out. I'm super excited for you guys.

>> It's fun to be on, and I'm excited to share whatever in the next few months of how that journey's been going. So yeah.

>> Great. Thank you for hanging out this week.

Rebecca Knight

>> No, thank you. It's been a great show. As you said, a great show. I've really enjoyed myself here at Snowflake. And thank you for watching alongside of us here at theCUBE. This has been the Snowflake Data Cloud Summit 2024 at the Moscone Center. I'm Rebecca Knight for Dave Vellante and all of us here at theCUBE thank you for tuning in. You've been watching theCUBE, the leader in enterprise tech news and analysis.