Snowflake Summit 2025 | Christian Kleinerman, Snowflake

Clips
More from Snowflake Summit 2025

Christian Kleinerman

EVP of Product

Snowflake

play_circle_outline Snowflake Summit 2025 day two highlights with Christian Kleinerman

play_circle_outline Snowflake's Advancements in Data and AI: Introducing Cortex Analyst and Search for Structured and Unstructured Data Utilization

play_circle_outline Openflow deployment models for data life cycle management

play_circle_outline Snowpark and dynamic tables for data transformation and optimization

play_circle_outline Snowflake: Transforming Business Processes Through AI and Data Collaboration for Improved Productivity and Value Creation

Info
Transcript

Christian Kleinerman, Snowflake

Christian Kleinerman

EVP of Product Snowflake

In this Snowflake Summit 2025 segment, Christian Kleinerman, SVP of Product at Snowflake, joins theCUBE’s Dave Vellante and John Furrier for a fast-paced deep dive into how the AI Data Cloud is evolving from data warehouse to full-stack intelligence platform. Kleinerman reflects on Snowflake’s first-ever $1 billion revenue quarter and explains how the company’s latest product wave accelerates that momentum by empowering every enterprise to build, scale and govern agentic AI.

Kleinerman unpacks headline announcements, from Snowflake Openflow for seaml... Read more

explore Keep Exploring

What recent developments and announcements have been made by the company being discussed? add

What are some of the key features and tools that Snowflake is offering to its customers to enable them to leverage their data for AI-driven productivity? add

What deployment models does Openflow have and how are they utilized in helping organizations with the entire data lifecycle process? add

What is the explanation for how Snowflake's new workload opportunity with Snowpark and Openflow is both economical and performant? add

What is the importance of partners and customers in creating value through collaboration and common data governance in the context of the AI data cloud vision? add

bolt Powered by CUBE AI

Christian Kleinerman, Snowflake

search

Dave Vellante

>> Hi, everybody. Welcome back to San Francisco. You're watching theCube's wall-to-wall coverage of Snowflake Summit 2025. This is day two. My co-host, George Gilbert, and I are excited to have Christian Kleinerman back on theCube. He likes it rowdy. We do too.

Christian Kleinerman

>> Very much.

Dave Vellante

>> All right. Good to see you. Big day yesterday. Big day today. Huge week. Congratulations on the keynote yesterday. All the announcements. We're going to go through them. What a fire hose. How are you feeling?

Christian Kleinerman

>> Super excited. I'm very happy to be here with you, Dave and George. I love always chatting with you.

Dave Vellante

>> Yeah, so it was interesting. We saw your last earnings quarter, billion dollar quarter on track for over $4 billion this fiscal year, so looking good. It's kind of spring in the step, but what's most notable is the number of innovations and product announcements that you guys have come up with. We kind of joke on theCube, but really not, going at the speed of Nvidia these days, you have to, and it looks like you guys are keeping up.

Christian Kleinerman

>> Oh, definitely we've seen a visible increase on product velocity and how much innovation we're cranking out, in many parts because of our own engineering practices, our own ability to leverage AI itself, but also AI as a category is innovating so fast that you just have to be on that pace and we're very excited about where we are and where we're headed.

Dave Vellante

>> So many announcements here. I mean, you guys laid it out that the data life cycle's accelerating. You've got to meet the data where it is, not just when it comes into Snowflake. You've got a number of sort of tooling to get it in there, but so many things, adaptive compute, the Gen 2 warehouses. Let's do the rundown. We don't have time to go through all of them, but what are the ones that are most salient to customers in your view?

Christian Kleinerman

>> Okay, maybe let's start with pure AI. The ability to leverage organizations' data to effectively drive productivity through AI, through AI agents and through agentic applications. That, I think, is in many ways the core of what we're enabling for organizations here at Snowflake Summit, and it starts with we are helping organizations make their data be AI ready, whether it's structured data or unstructured data, something that I would emphasize as many times as I can. It is not just traditional what Snowflake is known for, but unstructured is on a tear, enabled and lit up by AI. Cortex Analyst is our ability to do text to structured data. There's a SQL step in between, but as you saw in some of the demos at Summit, you don't have to stare at a piece of SQL, you can just get results or get visualizations. Cortex Search is our hybrid search capability that does semantic search, keyword search to do retrieval on unstructured data and to orchestrate it amongst those two multiple data sets of those two Cortex agents. That we're very excited. That enables applications, some of them that someone can build inside their organization, but also for partners for commercial opportunities that want to build an app, publish it. And then last but for sure not least in this world of AI is many of our customers said, "I don't want to have to go and build an app. That was what we did last year and it turned out to be harder than people thought. Can you deliver the end-to-end experience that we can put in front of business users?" And that is what Snowflake Intelligence does. It's a beautiful, very usable experience on top of Cortex agents and the feedback that we're getting and here at the show and with our early adopters is phenomenal.

George Gilbert

>> I heard a use case with Baris yesterday with S&P Global where they were analyzing tens of thousands of earnings transcripts but financial statements and it sounded like a really interesting case where you're taking the structured data, the P&L, the balance sheet, but you're also looking through the transcripts and then you're comparing things that executives said on the earnings calls with follow-on financial performance. So you can predict it's almost how good a management team is by how on point they are on a call. Have you seen other use cases like that where you're leveraging both types of data and where the agents have come on top?

Christian Kleinerman

>> There are many use cases and then there are two dimensions to different types of data. One is first-party data augmented with third-party data. Some of the use cases that S&P is doing is super cool. They have an ability to publish something on the Snowflake marketplace where it enables the consumer of that application to provide their data, but it gets augmented and put in context with data that belongs to S&P. That's one dimension. The other piece is the augmentation between structured and unstructured data. We're showing some examples where you can evaluate the performance of an ad campaign, saying, "Show me campaigns that have better than some quantitative result, but also where the creative looked like this image." That is a very difficult question to ask until what we announce here.

Dave Vellante

>> Very cool. I wanted to ask you about a couple of other announcements. Openflow, the acquisition of Datavolo, that's key, and I want to get into Gen 2 as well and some of the personas, but let's start with Openflow.

Christian Kleinerman

>> Openflow, I think in your opening comment, Dave, you got to the crux of something that matters to us. We want to help organizations with the entire life cycle of data and oftentimes that starts in their premises or in their VPC, and instead of waiting for some other process to move it to the cloud and then we say, "Hey, we're Snowflake. We can help." No, how about we help from the very beginning, the creation of that data set, that's where Openflow has two deployment models. One is typical Snowflake. It's Snowflake-managed resources, it's in the cloud, but there's also a BYOC, bring your own cloud, which can be deployed in the customer's VPC. Later on in the year, it will be deployed fully on on-prem or air-gapped environments, so now as data is created, call it an application, call it an OLTP database, it can start flowing, potentially get some initial information, initial data quality checks or pre-aggregation and make it all the way to Snowflake, and if I can say for the 10th time, applicable to structured and unstructured data.

Dave Vellante

>> Nice. Okay. Let's talk about Gen 2 because that's a big deal, big platform enhancement. My understanding is customers have to pay a little bit more, but they get 2x the performance. So I mean, I don't know what the pricing is, but let's say it's 1.3, 1.25, whatever it is, you got to get better than 1.3x performance. Tell us about Gen 2. How confident are you that customers will see that hit so they'll have better TCO?

Christian Kleinerman

>> Yeah, so you already alluded to the most important part, it's at a higher price point, but our expectation is that the performance benefit far exceeds that increase in price point. A lot of the difference in price comes from the underlying hardware, which is higher than what we've seen traditionally on Snowflake instances. All performance statements that I made have to be caveated, which is mileage may vary, but we shared some stats. 1.9 better than managed Spark, 2.1 times better than Snowflake roughly a year ago, and even though those are representative workloads, I would say customers should easily get more than the 1.35, 1.25 that you see in the price difference. By implication, not only the economics are better, but the price performance is dramatically better. One other thing that I would highlight is there's a big difference on the composition of workloads. If it's a read-only or read-mostly workload versus a higher write mix, the writes have even better performance than the reads. So for customers that have data transformation-intensive workloads, they're going to see lots of amazing results from Gen 2.

Dave Vellante

>> Interesting. I mean, what's typical? The typical majority of workloads are reading when you'll see... Is that not the case?

Christian Kleinerman

>> If you think about it from an analytics and data warehousing perspective, that's right, but lots of organizations leverage Snowflake for data transformation. You landed raw data and now you want to cleanse it, you want to improve quality, that is write-intensive and we have an amazing performance enhancement.

Dave Vellante

>> So always the bottleneck when you're storing something.

George Gilbert

>> So let me ask you, just drill into this, just because traditionally you guys were the strongest in the analytic workloads, but now it sounds like not only is the price performance accelerating for the data engineering workloads, but you have the fully Pandas compliant data frame for the Python guys as well as I understand more the Spark compatible version of the Pandas DataFrame. In other words, you can appeal to both the guys who start on the laptop and scale out and then the guys who want to work in Spark compatible.

Christian Kleinerman

>> I feel extremely, extremely happy of where we are for data transformation, because it is both dimensions, what you said, one, I start in Pandas in my laptop, I can scale it to Snowflake Pandas, which is scalable as large a data set as you want. Also, I can start with Spark, move it to Snowpark. That one has a little bit of a migration step which we are addressing. We'll have a future conversation with you guys on that topic, but also applies to people that want to do data transformation in SQL. I've gotten no in the comments on, "You did not say anything about dynamic tables in the keynote," and the reality is like, "Yeah, maybe I should have. We were just cutting... There was too much innovation," but dynamic tables is becoming an amazing tool for people that want to do transformation in SQL. We're seeing workloads migrate from Spark to SQL-based dynamic tables.

George Gilbert

>> Is that because you're using your world-class optimizer to incrementally move the data through a set of declarative tables where your optimizer knows how to do the best job, whereas a programmer doing it procedurally just could not match that?

Christian Kleinerman

>> That is true and I would broaden it to not just the optimizer but our overall data processing engine because just how you consume records, how you write records. We've been tuning this thing for 12 years now. So now when you apply it to a Spark workload or a SQL workload, the results are unmatched and we're very happy about it.

Dave Vellante

>> Carrying through on that thread, you're going after very strongly the sort of data science persona. You've got a data science agent. That data engineering workload, which we heard a couple of years ago, some customers were taking out of Snowflake, doing it elsewhere. That now becomes really a new workload opportunity for you that's extremely economical and very-

Christian Kleinerman

>> Performant.

Dave Vellante

>> Performance, yeah.

Christian Kleinerman

>> Very much.

Dave Vellante

>> Explain how you got there.

Christian Kleinerman

>> Well, so we started on the far right of the data life cycle. If you think of ingestion, processing, consumption, and usually the consumption was primarily BI tools, but increasingly, it became AI-based experiences or others downstream data transformations. In these days you talk about agent-to-agent communication. But we started on the far right, but many customers said, "Hey, your engine is quite good. How about we do data transformation?" That's what led to Snowpark. Then the laptop people showed up and said, "How about Pandas?" And now what are we're doing? Openflow is moving to the left and Snowflake Intelligence is moving to the right. So our general statement that we want to help organizations with the life cycle is spot on.

George Gilbert

>> So elaborate on what we were talking about before, why using all those together is more than the sum of the parts?

Christian Kleinerman

>> Yeah, it is a very important part of the conversation, which is we don't want organizations to think of each of these parts of the life cycle, which usually correspond to different personas, we don't want them to think of them as independent isolated problems to solve because they're very interrelated. One of the original pieces of feedback that we got when we were hosting the Python runtime was data science teams all of a sudden stopped fighting with the data governance teams because they could do data science without copying data out, which means you're honoring permissions. You want to do AI, you want to do text to SQL with Cortex Analyst, Cortex Analyst honors all the permissions that you have, so again, no tension is more synergy. Things like feedback loops. How does Cortex Analyst understand what queries are more popular, what queries are more likely, which ones are used by a specific user? Maybe George versus Dave have different definitions or answers that they prefer to a given question. We have all of that inquiry history so they can feed to our semantic models, they can feed to Cortex Analyst. There's a lot of synergy, would be my word, between the different personas.

George Gilbert

>> And then just let me follow up on that for a sec, because one of the things, just talking to Jeff Hollan because he was on just before you, and the semantic models, or the metric and dimension models that the BI tools have, these are the top-down specified and they're really just for the visualization data, but yours is learned and bottom-up and the key thing is as each persona uses them, you get a richer model of your data that all personas can use.

Christian Kleinerman

>> Correct. So traditionally, semantic models have lived embedded in BI tools, which is useful for the BI tool, but if the data governor or data steward wants a different type of modeling, they do it all over again. We went through the process of do we build it in Cortex Analyst, do we build it in the database, and we ended up, it's a separate layer that both AI and BI can leverage. Another question is where does the industry go, and how do we standardize this in an open and interoperable way?

Dave Vellante

>> And what does that mean for partners? I think of you guys sometimes, Christian, as you're like the AWS of data and that's a compliment, because moving fast, if you see a customer need, you go to fill it and then the partners, they have their specialty as well. So I'm curious, when you think about partners, you mentioned BI, where the industry goes, where you go, we talk like text to SQL. If a partner has another text to SQL capability, where does that fit? Same thing with search. So how do you think about the partner value add and how that evolves?

Christian Kleinerman

>> It's a great question and we're in the middle of the partner pavilion and we see that this place is vibrant and healthy and lots of excitement. Customers at the end of the day want to make sure that they have control of their data and increasingly their metadata. Our goal is how do we enable that to be customer controlled, nobody wants lock-in, everyone wants openness, and then have as broad a partner ecosystem as you can that operates within those constraints? That means they honor security and permission, they leverage the single copy of the metadata because I don't want to have 10 different semantic models, and I think that's the opportunity you see here. We shared yesterday the remaining independent BI players, Omni, Sigma, X, all of them are integrating with the semantic view because they see that the opportunity is not to compete with it but add value to what customers already have.

George Gilbert

>> So the key thing here is as the semantic model migrates out of the BI tools, it becomes broader and reinforces... There's a learning loop, a feedback loop so that more tools then can get better value out of all the data, because it's not just a narrow semantic view tied to one tool.

Christian Kleinerman

>> 100%. The role of a semantic model or semantic view is how do I translate business terminology with physical realities of data? That's a very generic capability that is applicable to everything. I can imagine natural language-based pipelines that use the semantic model. I can do it for analytics, I can do it for AI.

Dave Vellante

>> So thinking about the future, we often have this conversation about the vision of the future of people, places, things and processes, digital representations of a business, digital twin of a business in real-time and we often ask ourselves, "Okay, what's the role of Snowflake in that world?" Is Snowflake the feeding that fourth-dimensional map? Is Snowflake that actually fourth-dimensional place where you build those apps? Maybe it's uncertain, but how do you think about that?

Christian Kleinerman

>> I'll refer back to our broader vision. When we talk about the AI data cloud, it is not just, hey, some pretty words to say database. No, it's truly the insight of a technology platform. We talked about the life cycle and we talked about the pace of innovation, but it's the ecosystem. That's why I cannot emphasize enough partners, but also customers in the role of partners. When you see folks like Capital One that took what they had built for Snowflake and decided to turn it into a business, that's the blurring of roles, and we do believe that our role is to enable that collaboration, that enable that facilitation, which is common data, common governance, common semantic models and metadata, and then let's all create more value from leveraging each other's data and data products.

Dave Vellante

>> Okay, we're out of time almost. Go ahead.

George Gilbert

>> So considering now, this year was really interesting because so many products came together with the agents on the foundations that you've been building out for years, but things are moving, as you said, so fast at the AI layer and now the semantic modeling being built out so that the agents can navigate it better. 12 months out, predict for us some of the new persona experiences that we should anticipate.

Christian Kleinerman

>> Well, 12 months in this day and age is such an eternity, but I think we will see business users be able to configure workflows or agentic workflows so that business processes are now supported by AI. I'm not going to predict a wholesale replacement of the machines took over, but I would say that we will see massive augmentation of virtually any business process with minimally context and potentially lots of action-taking decision making.

Dave Vellante

>> That's user-driven?

Christian Kleinerman

>> Correct.

Dave Vellante

>> That's business user-driven. I mean, look, a lot of business users say, "Hey, my job is to put myself out of a job and move up the ladder."

Christian Kleinerman

>> That's what productivity has been about and I have very high conviction that we'll see a massive improve in productivity almost in any business process across industries.

Dave Vellante

>> Well, it's a great vision. Now what's next for you? You take this around the world, the world tour?

Christian Kleinerman

>> Yes. So we love the Snowflake World Tour. We do it in, I don't know the official number of cities, 20 or 25. I think I'm signed up for four or five of them. We love going and sharing this in local settings. Turnout is amazing and we're excited about what we're doing.

Dave Vellante

>> Well, I think you're going to be in New York in October I think. I don't know. We'll see. We're going to try to be down there to go to bring you over to the New York Stock Exchange, our studio there. Would love to have you. It'd be unbelievable.

Christian Kleinerman

>> That'd be great, and yes, I think New York is in my dance card.

Dave Vellante

>> Good. Well, let's meet in New York, if not before. Christian Kleinerman, thank you so much for coming back.

Christian Kleinerman

>> Dave and George, thank you.

Dave Vellante

>> All right, appreciate it. And keep it right there everybody. We'll be back with more action from Moscone, theCUBE's coverage of Snowflake Summit 2025. We'll be right back after this short break.