Clips
News
More from Databricks Data + AI Summit 2024

Tony Baer

Principal

dBInsight

Sanjeev Mohan

Principal

SanjMo

Databricks leads the way in open-source and no-code technologies: key analysis

The data and artificial intelligence landscape is undergoing a transformative shift with the integration of no-code technologies to enhance efficiency and innovation.One of the companies leading this charge is Databricks Inc., whose recent developments promise to streamline data management and processing through open-source initiatives and advanced automation.DbInsight’s Tony Baer talks to theCUBE about no-code technologies.“We’ve seen the waves of innovation,” said John Furrier, executive analyst at the CUBE Research

play_circle_outline Integration of data and AI stack

play_circle_outline Standardization and unification of data engineering practices

play_circle_outline Progress and adoption of Databricks products by customers

play_circle_outline Catering to different user personas in both Databricks and Snowflake

Info

Tony Baer | dBInsight & Sanjeev Mohan | SanjMo

Tony Baer

Principal dBInsight

Sanjeev Mohan

Principal SanjMo

>> Welcome back, everyone, to theCUBE's live coverage. Day two of theCUBE, day three of Databricks's Data AI Summit. I'm John Furrier, host of theCUBE, with Savannah Peterson, Rob Strechay, George Gilbert, the whole team here. We've got the two CUBE collective, VIPs, Sanjeev Mohan with Sanjo, and then Tony Baer, Principal of dbInsights. Guys, thanks for coming on as part of theCUBE Collective, contributing to our open source analyst angles.

>> Great to see you again, John.

>> You guys are the pros. Both legends in the data community. We've seen the waves of innovation. The ball moves down the field, inch by inch, sometimes goes far. Now we're in a whole nother era. You're seeing a lot of, maybe potential big moves coming. And Databricks is laying down a lot of strong game here, by going for the long game, by saying, "Hey, we're going to open source things. We're go for formats." Does their conjecture match the reality of the customers? Or is this Databricks just doing their thing, next level? What's your take?

>> Okay, let's see if we can take it for 100,000 feet. Part of this I see is like a whole, there's perception and then there's some differences. And obviously, there was another event here in the last week, what I call the Snowbricks month. And the thing is, on one hand we have Snowflake, basically is in a more mature market, which is data warehousing. Databricks is in a market which is a lot less mature, which is the data and AI market. Now they're both going into the same places. I think there's a lot of interest on the open source side. I think the challenge, and I'll just make this short on this one, we can go back into more detail, but I think the challenge for Databricks is that their history of open source, aside from Apache Spark, has been projects that Databricks has been largely at the helm of. Whereas, basically, when you look at, let's say like Iceberg, it was a more of a community project, more contributors. So I think the open source, especially open sourcing of Unity, is an excellent development. On the other hand, I think there's some perceptions here that Databricks has to overcome, to overcome the fact or the perception that they're actually controlling this.

>> Got it. Sanjeev, what's your take on the big picture?

>> So on big picture, what we are hearing from customers, and what we saw last week at Snowflake Summit and here, is that there is a massive integration of the data and AI stack that is happening. And it's all coming together. Last week we were all pretty much taken aback by the Databricks tabular merging. And we were thinking that, "Oh my God, this is such a big news," but I am happy to report that table formats are not the topic, they're not the center of topic.

>> No.

>> They're a much bigger picture what Databricks is doing with Lake House Federation, or the new ETL that they're doing with LakeFlow, where you can connect to not just Snowflake, but SAP, Salesforce. You can do ETL, you can do sharing of data, you can do clean rooms. I really feel that both Snowflake and Databricks, to be fair, are really innovating. And they're helping their customers really focus on data product, AI assistants, agents, co-pilots, and think less about table format and ETL. And so there's a lot of automation. Like LakeFlow, you can do so much of transformation without writing a single line of code.

>> Right.

>> So you're thinking, what jumped out at you is the federation was good, but don't you need the data formats to be compatible? It's in the weeds, for sure.

>> Yeah.

>> I'm not a big-

>> Correct.

>> But if you get that right-

>> Yes....

>> the enablement's strong.

>> Right.

>> Okay.

>> So you do need formats, but not for everything. You need a table format if you're going to have structured data and you want any compute engine on top. But when you do AI, then the table formats won't come in the picture. When you do data sharing, again, you need delta sharing as a standard, but not the table format. So there are many-

>> Sure.

>> There's a lot of standardization going on, of access points.

>> Yeah. The other thing too, also, and I think you're just working up to that in what you were saying, is that at the end of the day, vendors are not differentiating on table format. We just want to get that issue out of the way.

>> Yeah, get it done.

>> Yeah, exactly.

>> Every major inflection point in my career, I've seen, things like TCPIP as a standard, defacto standard, Kubernetes, Cloud Native. Is the unification of the governance piece, the catalog, is that the area that this world needs to standardize, or defacto agree, or unify around, to make data engineering truly scale? Because with all the fragmentation, there could be a food fight all day long. Every year will be a food fight. Table formats today, my federation versus yours. At what point is there some sort of consensus? And what would that look like?

>> Well put it this way, I think federation will be one area that still will be highly contested, just as data sharing is. And of course, obviously, Databricks' strategy is to open source it. And they do have a pretty impressive ecosystem of supporters coming out of the gate. I do think in the data federation, and in the catalog part, that's where you get to the point of who really owns this data? Who controls this data? Who governs it? What's the point of governance? What's the point of discovery? And that's really where the value add starts.

>> Yeah. So when I look at this, this question comes up almost every day, what is the mode that I should protect?

>> Yes. Yes.

>> People thought all these large language models are mode, but they're not. Because the only thing with them, there's no technology mode. It's like, who has the money to spend? And how can you distribute? If you are Googles of the world, then even with a few missteps with Gemini, you still, you have a distribution, and you have lower money, OpenAI, Azure. So then the question is what is a mode? And that mode is in metadata governance, federation, ETL. And I think whoever owns that simplified integrated stack has the benefit. They can do... Like for example, just to complete, you can use any BI tool, you can have a semantic layer. So last year, John, you and I actually went to bat. The database is missing semantic layer. Guess what, it's front and center. So you got semantic layer, you've got lineage, you've got ingestion.

>> Yeah, the data intelligence platform, DIP, DIP.

>> Dip on NIM.

>> DIP. Love Jensen's comment yesterday. It's almost as good as Nim. He's so funny on stage. He's winging that, by the way.

>> He really is.

>> He wings that too.

>> Oh yeah. He got close to 20 minutes on stage. That was like, whoa, we're writing The Jensen Minutes.

>> Ollie was laughing. We could see him smiling. He was loving every minute of it.

>> Yes.

>> Who wouldn't? So okay, you're seeing the data platforms grow.

>> Correct.

>> You're basically saying. So there only can be a couple of really intelligent data platforms out there. The big get bigger. Okay? And so in every inflection point, not only is there standards, there's also times where we as analysts or industry participants will look at a company, whether we're a VC or an analyst. That's not a company, that's a feature. So as you start to see LakeFlow and these other things come out, there are companies that do these things.

>> That's right.

>> And they're funded. And so the question is, do they go away and see old classic, you're going into the storm where they're going to dominate, where Databricks will have a solution? Or are they going to do the AWS model, where there's, oh, we have a product, you're going to be differentiated? I'm not sure I see it that way, but what do you guys see?

>> I see it that way. I think they're in trouble.

>> Which way. Which way?

>> I think a lot of, I don't want to mention these companies by their names, but data transformation and all, I think they're in trouble.

>> Yeah, yeah.

>> Which categories get sucked into the big platform. And then the next question is, obviously Ollie's saying, "I love the ecosystem. It's super important to our business." Where does the ecosystem flower out? Where's the soil at the plant?

>> It's where the value comes. And I think is a really good example, is I think what Databricks has shown from the Mosaic... I keep coming to... Was it Mosaic ML? Or Mosaic AI? I think last year was Mosaic ML, and this year-

>> They changed their names from ML .

>> They changed they're names. Exactly.

>> It's AI washing.

>> Exactly, exactly. But I think that's going to prove to be a real sleeper. Actually, from that, and some of this come from Mosaic, some was from internal development, but they put this all under a common platform, of being almost, they're putting together. We're at the beginning of this-

>> It's all the tooling.

>> Exactly, of a complete life cycle for basically developing and building these model's, the .

>> It's an app. To me, it's an app suite.

>> Exactly.

>> So think, the fact that they started the keynote today with the small language models...

>> Yes....

>> I think was a North Star saying, "Hey, models are important. Your data's important. And don't buy the hype from Open AI."

>> Right.

>> Models will talk to each other. They'll be protected, but they're going to be in, obviously Mosaic AI. So they're teeing that up. But I think that was a smart move, because you got to give the headroom.

>> Yeah, oh, gosh, yeah.

>> The bridge to the future is going to be a reset of the entire data engineering fabric. So the question is, what does that look like? And again, which companies survive and thrive, and which companies, what do they pivot to?

>> Yeah.

>> Where are the opportunities to add value? And then, because you got, Snowflake is doing the same thing.

>> Right, right.

>> Amazon and Azure, they got, they're big company.

>> Sure, sure.

>> So at some point the platform will subsume the largest enterprises.

>> Right.

>> Yeah, mid market, people can have some little patch of grass there. But...

>> So you know yesterday we were in a session with Ollie?

>> Yes, yes.

>> And a lot of stuff we are not at liberty to talk about, but I asked him this question, because CEO of a very prominent company behind us was complaining about the competition now they're getting from Databricks. And so there's this ecosystem versus competition. And Ollie's answer was brilliant. He basically said that we are not at his level. He's not going to play favorites. But we live in a capitalist society, best products should win. So he's told his product team to build the best product, and compete at merit. But then at his level, he's going to maintain the ecosystem.

>> Yeah, but there's some categories as we were just talking about, that just absolutely fall into a core competency of Databricks.

>> Right, right. Right, right.

>> For their stuff to happen, you can't just co-op petition those.

>> Right, right.

>> You got to own them. Microsoft went through this in the '80s and '90s when they were building out the Windows ecosystem.

>> Yes, same story. Yeah.

>> Remember that same thing. Amazon didn't have that problem, because they weren't in the app business. So they were in the compute business. Everyone-

>> Exactly....

>> they were an arms dealer.

>> Well, the other thing too is also, I think this is sort of an extension of the backlash against what we used to call the modern data stack. And that the modern data stack was supposed to be modern, but it was really complex. It was very complex tool chains. You had to put all the pieces together. It was kind of like the best of breed all over again. The fact is, in a more complex world, we want simplicity. So I think there is something to be said for the fact that we are seeing more of a drift towards platform plays. There will be some specialized providers of, I think domain-specific models. I really like the whole small language model thing, the meek shall inherit the hearth.

>> Yeah, I agree. So how would you compare, I did a post on this before the event, that the Hadoop ecosystem could look a little bit like this, complex stack, not a lot of ecosystem? But I think the serverless announcement got my attention. One, because that's going to take away a lot of the knobs from a deployment standpoint, and cost. I'm not sure if that was one of the convenient things. But also, Amazon's all a hundred percent serverless too. So serverless is a good architecture. So to me, that might make it easier.

>> Yes.

>> And some of the Unity catalog features I saw today, and with Delta, DeltaLake, they're automating a lot of stuff that's getting better.

>> Yes.

>> Got metrics in there now.

>> Correct.

>> So starting to see some progress from Databricks on the product side. The question is, what customers are adopting it? What are you guys seeing from the customer side? Are customers buying the Databricks concepts? Are they leaning in? Not always, conjecture might not always translate, but for the most part it does.

>> Look, at a conference like this, obviously, it's going to skew towards those that are going to be more leading adopters. I'm still seeing a lot of discussion with a lot of clients, basically between, we have Snowflake and we have Databricks. And they are going to continue to coexist, and we have to figure out how do we basically rationalize this? So I think that's really still a very dominant thing. The fact is, within any organization, you're going to be having these dueling platforms. The thing is, it's better to have dueling platforms than have dueling, basically, very complex tool chains of hundreds of tools.

>> Yeah. I love that comment you made about earlier, about value creation. That's where the action will be. What I like about the keynote today, from Ollie, words like the Databricks kernel. They're nerds. They're talking about building almost like an operating system. And what we haven't talked about yet is, if data's going to be the lifeblood of a neural network infrastructure, with knowledge grass and new kinds of data structures, and you get the governance right, the data can just do its thing. And so that's the question. What's after Databricks Snowflake? Whether they collide and kill each other, or they collide and coexist, there's going to be a-

>> You take it first.

>> Okay, sounds good....

>> there's going to be a post conversation or layer of operating environment.

>> Yeah.

>> Yeah, obviously, they're going after the same pot of gold, so they will have collision. But I just think that the pot of gold is too big, for they don't need to have that aura of the colliding, and the political moves. They can both thrive. I still feel that Snowflake has some distinct advantages. Even you can see with the keynote, how did the keynote start? We were trying to diagnose a neurodiagnostic theorem with formulas.

>> Yeah, yeah, that's nerdy. And today, it's even more nerdy.

>> Exactly. Exactly. I mean, I liked it.

>> So you don't see that at Snowflake.

>> No, no.

>> I think Snowflake still has a lot of good things happening for them, as does Databricks. The use cases are very wide and broad, and I love my clients. Actually, 45% overlap. That's the ETR data. Between Snowflake and Databricks, you know?

>> And that has not changed in-

>> Snowflake ran the table on the business analytics and the cloud.

>> Right, right.

>> So they have all that business content.

>> Here's a good example of how I think they're going to differentiate. Because the fact is, even though they're both coming towards to the same place, they both have, where they're coming from, they've served different constituencies. And I think one of the things that's been overlooked, is Snowflake Cortex. If you think about it, take a look at what they're doing with universal search. They're putting all the complexities of RAG under the hood. And that really typifies, or basically the type of end user, the constituent that Snowflake addresses, that's the type of thing which is falling into Databricks. So they both are going to have, come at it from different directions, and have different users, different categories of users, for which it's going to be home turf.

>> So Tony, one thing that I see as little bit of a risk, with both the company, is that they both had their core personas. Like data analysts for Snowflake, data engineer and data scientist for Databricks.

>> Sure.

>> And now they want to cater and be everything to all personas. And I think that-

>> That's the nature of competition in our business, frankly. The fact is, I did a piece years ago. This is during the Hadoop era, and it was also during the era of the flowering of all these NoSQL databases. And I basically said, "Look, what we found out was that there was a Cambridge explosion, we had all these different needs." But in the long run there was going to be a kind of a convergence. Not that all databases are going to be the same. In other words, yes, you can do JSON on Oracle, but if you're just doing pure document database for application, you're going to go to Mongo. The fact is, there is going to be a convergence, but each are going to have their unique areas of strength.

>> I think there'll be a lot of diversity of data storage, data sources, for sure.

>> Right, right.

>> Diversity of models that you pointed out, I think she's right on, and right on to get behind small language models. We've been reporting on that. That's going to be a big part of the software data integration of apps. You got to have small models.

>> Exactly.

>> You know we keep forgetting how we've gone through many girations of trying to bring everything together in one place, right?

>> Yeah, decades.

>> I mean, corporate data factory, enterprise data , how did we actually... So it seemed like the pendulum has shifted again. It's not going to be like that. It's going to be a very heterogeneous environment, whether we like it or not. And others have a good chance of being strong players. Oracle, Cloudera, IBM, Watson Extra Data. I think they're all...

>> Yeah, well that's a good point. Let's bring that out. I want to just... Because I think we talked about this before, those were major initiatives, but they didn't bring a lot of structural change to the overall fabric of an enterprise. Like IT and how people organized. If you look at intranets, and internet, and email, and then that changed IT. Top of rack switches, versus installing office on PCs and running around with disks. So IT was born. That was a structural change that was a generational shift in how money was spent, and how IT was built, databases integrated more. But I think gen AI brings that structural shift. It's a whole other thing. So the question is, does the tail wag the dog, or is it now, whoever controls the infrastructure? And so the question is, if I got the business users-

>> I have a favorite tagline.

>> Okay, so go ahead.

>> A favorite tagline from yesterday, what we heard. Gen AI gets us the introductions, and then the discussion shifts to machine learning and ETL.

>> Right.

>> That's where the work is.

>> There was this popular perception, and I actually wrote a piece for SiliconANGLE some months back, to say that there's more to AI than Gen AI. But the popular perception, because ChatGBT has, it was kind of like consumer tail wagging the dog was, "Oh, it's all Gen AI. Well, underneath it, we have a lot of predictive analytics. And in fact, if you want something that's deterministic, you're not going to go to a generative approach.

>> So do you think there'll be structural change that'll be different, that'll be a forcing function? Or it's a minor tweak? And-

>> No, it's going to be a major generational change, but it's going to take some time for all this to kind of sift itself out. And so, another thing that, a post I had written recently, was that there's a lot of excitement about this. The vendors are actually under the gun right now to have to put an enormous investment. But things, enterprises are just starting to digest this. They're basically kicking the tires. And also, it's like we're seeing rapid innovation, in terms of how we're doing models. They were transformers, now we're seeing a certain mixture of experts. We're starting to see this more this ensemble approach. And it's amazing how, and the fact is, it's going to take time for all the debris to settle, to figure out, okay, what are going to be the sustainable approaches? It's going to be several years before this market really materializes.

>> Yeah, it's debris, and also fog's got a lift too. It's like, where are we?

>> Yes.

>> What's your take on that, his point there?

>> Yeah, Gen AI may be attracting all the attention, but to get it right is a very, very tough job. We are now finally facing it, after almost two years of trying to do it. The idea was, in 2023, we'll do experimentation. In 2024, we've got production. But they're nowhere close to being production. So now we've got RAG, we've got types of RAG, like graph RAG. We've got fine-tuning has now become a very important topic at this conference. And then we even talk about training small language models. So we basically, at a stage where NLMs are amazing for basic language task, like summarization, recommendations. But when it comes to doing reasoning, and more complex tasks, which is what the customers want, at the end of the day, we still have a lot of work to do.

>> It's just the beginning.

>> Yeah, we all right at the very, I would say, we are the first inning.

>> Generative does not do reasoning. Basically, it's a huge probability model on steroids.

>> Yeah, reinforced learning to come next.

>> There's no context.

>> All right, Tony, Sanjeev, great to have you on. As always, you guys have been following the industry for decades, and more recently, this big data movement's now happening, I'm starting to see some action. The last minute we have left, put a plug into some of the work you guys are working on. I know you talked to a lot of customers, you certainly talked to all the vendors. They want to talk to you, actually. But you got a lot of customer stuff going on. What are you guys digging into? What are you unpacking in your research these days? Tony, we'll start with you.

>> Data is still important. Don't forget data. I'll just put it at that.

>> Yeah.

>> My main interest is on what is going to be the next platform. So Gen AI becomes another compute, rather than a standalone piece. So that's where I've been .

>> And you buy compute separating from data as a legit thing?

>> Oh yes.

>> Hundred percent? Yeah.

>> A hundred percent. Yeah.

>> I think we got to that like five years ago.

>> Genie's out of the bottle there. Now we're going to see some meat on the bone. Guys, thanks so much. You're watching theCUBE. We're day two live. John Furrier, your host, with Savannah Peterson, George Gilbert, Rob Strechay, and of course, theCUBE Collective analysts here. We'll be right back after this short break.

Tony Baer | dBInsight & Sanjeev Mohan | SanjMo

>> Great to see you again, John.

>> Got it. Sanjeev, what's your take on the big picture?

>> No.

>> Right.

>> So you're thinking, what jumped out at you is the federation was good, but don't you need the data formats to be compatible? It's in the weeds, for sure.

>> Yeah.

>> I'm not a big-

>> Correct.

>> But if you get that right-

>> Yes....

>> the enablement's strong.

>> Right.

>> Okay.

>> Sure.

>> There's a lot of standardization going on, of access points.

>> Yeah, get it done.

>> Yeah, exactly.

>> Yeah. So when I look at this, this question comes up almost every day, what is the mode that I should protect?

>> Yes. Yes.

>> Yeah, the data intelligence platform, DIP, DIP.

>> Dip on NIM.

>> DIP. Love Jensen's comment yesterday. It's almost as good as Nim. He's so funny on stage. He's winging that, by the way.

>> He really is.

>> He wings that too.

>> Oh yeah. He got close to 20 minutes on stage. That was like, whoa, we're writing The Jensen Minutes.

>> Ollie was laughing. We could see him smiling. He was loving every minute of it.

>> Yes.

>> Who wouldn't? So okay, you're seeing the data platforms grow.

>> Correct.

>> That's right.

>> I see it that way. I think they're in trouble.

>> Which way. Which way?

>> I think a lot of, I don't want to mention these companies by their names, but data transformation and all, I think they're in trouble.

>> Yeah, yeah.

>> They changed their names from ML .

>> They changed they're names. Exactly.

>> It's AI washing.

>> It's all the tooling.

>> Exactly, of a complete life cycle for basically developing and building these model's, the .

>> It's an app. To me, it's an app suite.

>> Exactly.

>> So think, the fact that they started the keynote today with the small language models...

>> Yes....

>> I think was a North Star saying, "Hey, models are important. Your data's important. And don't buy the hype from Open AI."

>> Right.

>> Yeah, oh, gosh, yeah.

>> Yeah.

>> Where are the opportunities to add value? And then, because you got, Snowflake is doing the same thing.

>> Right, right.

>> Amazon and Azure, they got, they're big company.

>> Sure, sure.

>> So at some point the platform will subsume the largest enterprises.

>> Right.

>> Yeah, mid market, people can have some little patch of grass there. But...

>> So you know yesterday we were in a session with Ollie?

>> Yes, yes.

>> Yeah, but there's some categories as we were just talking about, that just absolutely fall into a core competency of Databricks.

>> Right, right. Right, right.

>> For their stuff to happen, you can't just co-op petition those.

>> Right, right.

>> You got to own them. Microsoft went through this in the '80s and '90s when they were building out the Windows ecosystem.

>> Yes, same story. Yeah.

>> Remember that same thing. Amazon didn't have that problem, because they weren't in the app business. So they were in the compute business. Everyone-

>> Exactly....

>> they were an arms dealer.

>> Yes.

>> And some of the Unity catalog features I saw today, and with Delta, DeltaLake, they're automating a lot of stuff that's getting better.

>> Yes.

>> Got metrics in there now.

>> Correct.

>> You take it first.

>> Okay, sounds good....

>> there's going to be a post conversation or layer of operating environment.

>> Yeah.

>> Yeah, yeah, that's nerdy. And today, it's even more nerdy.

>> Exactly. Exactly. I mean, I liked it.

>> So you don't see that at Snowflake.

>> No, no.

>> And that has not changed in-

>> Snowflake ran the table on the business analytics and the cloud.

>> Right, right.

>> So they have all that business content.

>> Sure.

>> And now they want to cater and be everything to all personas. And I think that-

>> I think there'll be a lot of diversity of data storage, data sources, for sure.

>> Right, right.

>> Exactly.

>> You know we keep forgetting how we've gone through many girations of trying to bring everything together in one place, right?

>> Yeah, decades.

>> I have a favorite tagline.

>> Okay, so go ahead.

>> A favorite tagline from yesterday, what we heard. Gen AI gets us the introductions, and then the discussion shifts to machine learning and ETL.

>> Right.

>> That's where the work is.

>> So do you think there'll be structural change that'll be different, that'll be a forcing function? Or it's a minor tweak? And-

>> Yeah, it's debris, and also fog's got a lift too. It's like, where are we?

>> Yes.

>> What's your take on that, his point there?

>> It's just the beginning.

>> Yeah, we all right at the very, I would say, we are the first inning.

>> Generative does not do reasoning. Basically, it's a huge probability model on steroids.

>> Yeah, reinforced learning to come next.

>> There's no context.

>> Data is still important. Don't forget data. I'll just put it at that.

>> Yeah.

>> My main interest is on what is going to be the next platform. So Gen AI becomes another compute, rather than a standalone piece. So that's where I've been .

>> And you buy compute separating from data as a legit thing?

>> Oh yes.

>> Hundred percent? Yeah.

>> A hundred percent. Yeah.

>> I think we got to that like five years ago.