 We're back live in our Palo Alto studio, the IBM Storage Summit. The panel is here, the analyst panel is next, Sarbjit Johal, friend of theCUBE. Great to see you again, and Rob Streche, a CUBE analyst. So guys, let's unpack what we heard today. But I want to start with the industry. You know, Rob, we've been in the storage industry for a long time. Obviously it's evolved into the data business, right? With data, it's much more interesting than what my wife used to call snorage. But storage is not boring anymore, because it is so much tied to data and AI. But Sarbjit, you've been an observer as well. You've worked for a lot of different companies, both in the buy side and the sell side. How have you seen the evolution of, you know, what used to be generally known as storage, storage box industry, to where we are today? Yeah, storage has evolved a lot. Actually, I used to work at EMC, the storage company. You know, I've heard of that. Yeah, I've heard of that. Yeah, the storage has evolved all along. But since the virtualization, storage needed to evolve even faster. And it's always a dance of computer storage and networks. Like, you know, sometimes one of these three players runs ahead, it's that delicate dance of these three entities, if you will, in computing. Storage is becoming smarter over the time. And we have storage policies now, like what data needs to move, where, how fast we need to get back to that data, and do we need it real-time access? Do we need to cache some stuff? So it's all over the place, but with the invent of AI, which we'll talk about a little bit in a few minutes, I think the intelligence needs to come more to the storage layer to provide security, as well as performance. I think those are the two key factors. And economics matters all along, and actually I'm an economics major, we always talk about the consumption economics, and personas matter, economics or the practitioners matter, like how much training they need, what they can, what is in it for them to change, if you will, developers, architects, they all play a role in this. But storage is one of those constructs in computing, which is a little abstracted from the developers, to be honest with you, like a developer actually thinks last thing about the storage, right? They still think of the network traffic bandwidth and all that, but from the security part of you, I think storage is key. Our RTO, our PO's depend upon how quickly we can get back to data and how- And not lose it. And not lose it, yeah. How about you, Rob? I mean, storage has become programmable. I mean, that's obviously a big change. What else have you seen? Yeah, I think that they're moving up into data platforms. And I think what we've been talking about is really how there's always the accessibility versus the security aspect of it, where there's challenges and trade-offs that need to be made. And I think that a big piece of it is gonna be that as it evolves and as protocols become less of a thing, and it's more about access methodologies, such as rest or going through and using file or POSIX, NFS, what have you to get at the data, I think that transparency of that, meaning that the systems can do everything now. And they're not just one specific system you go buy, here, I'm gonna go buy this for file, I'm gonna buy this for object, I'm gonna buy this for block. The coming together of that has really, I think we're at that point where they are together. And that's really a key. I think the platform angle as well, I mean, we love them and hate them, the Gartner Magic Quadrants, but you've seen how the storage Magic Quadrants evolved. I'm not that up to speed on it, but I mean, the last time I looked at it, it was like every company was up and to the right, because, and my point of all that is that the industry has just really morphed into a data platform, and that means different things to different people. It could mean data management, right? It could mean the data science piece of it. It could mean managing storage, but they're all sort of coming together in a van. Yeah, I think that the technology is one part of it. I think people in processes matter a lot, which we, most of the vendors tend to talk less about, they always talk about products, you know? This is our product. It's self, people in processes, well, the services companies do, right? The consultancies do. I like Pete's session today a lot, because you guys also touch upon the personas, you know, who are different personas. You guys do appeal to. I think it matters a lot, like how do we change the consumption patterns? What are the new protocols? What are the new needs? Like streaming was not that much needed earlier on, but now that's the thing, right? So what does that do to the data access and caching, and I think the performance matters a lot, performance means different things to different people. It means different thing to developer, then architect, then a platform engineer, performance means different thing, and of course the CFO looks at the performance in numbers, right? How economically viable these systems are. So I think speed is the new scale we say, we heard that term, that phrase many times. I think that matters in storage sort of scenario a lot. And speed also is of different types, the speed to deploy systems, the speed to detect, attack, speed to respond to the attack, right? And also the performance of the system itself is a speed construct, if you will. So I think getting to performance systems from all fronts, including economic performance, if you will, is a trick. Quite a few things have to converge to get there, and it's the technology processes, people storytelling, partnerships and alliances, openness of your system. We heard about the ecosystem earlier, and an open data standard, I think Vincent was talking about that. I think that's a great development, and a few other vendors are hopping onto that too. Let's talk about AI a little bit. So we know from some of this ETR survey data that people are going to buy it embedded, or they may be going to buy it direct. My question to you guys is, let's talk about IBM and others in the storage business. How should they be leveraging AI? Obviously they can do it for AI ops within the system and automation in the system, but I feel like Rob, there's more than that in terms of an opportunity for the storage industry. I think in building off Sargeep's comments on that, I think there's the aspect of locality, data locality, and where does the data live, and how do you get at that data, especially when you're talking about AI and you're building models, and I think there's going to be much more specific models or small language models versus these ultra large ones that are billions and billions of different parameters. I think you're going to start to see that same thing is going to get applied to the actual systems themselves. How do you bring the actual compute to the data and the data to the compute? There's going to be some of that motion, and I think some of that was hit on as well during the different sessions we had today is how do you have that transparency of data both on-prem and in the cloud? The standard line these days is I'm going to bring the AI to the data. So what does that actually mean? You just said bring compute to the data, and then the other way around, if Einstein move, variation of Einstein move, move as much data as you need to, but no more. But what does that mean to the data platform? I mean, the data's sitting in storage, so if I'm going to bring the AI to the data, it's in databases too, of course, but the databases are sitting on storage. So what should the storage vendors be doing to take advantage of that? Is it just making sure it's resilient? Get it there fast? I mean, that plumbing piece of it is critical? Are there other things that customers should be thinking about that they should be asking their vendors to do? I think number one is security. Number two is performance. Performance from all different aspects, which we just talked about. The economic performance, system performance, and the productivity of the practitioners who are operating these systems, that is part of that whole thing, right? So I think the use of AI, we have to talk in three different buckets. One is very short-term, like what we can do in the next six to 12 months, and then what we can do from one to two to one to three years, and then what we will do in long-term. We have to strive for that long-term as a few IBM fellows were talking and he was talking about that, I think he explained it really well, like where we are headed. So ideally, let's talk about the ideal situation. Ideally, we should get the packets coming to us, which needs to be stored, right? And we can analyze those for their authenticity, where who's sending them, is it good data, should we persist it, or is it attack coming in our way, right? So that kind of mechanism, we need that. And right now, Andy was talking, they're doing sampling of the packets and take a look at the compressibility of the data, how much they can compress, that's one measure. The other measure is how the data is looking from the pattern of the data is any anomaly in it. But there'll be more things like that. There was an example in the very early session, I just, which I really loved, that how many times you pay attention to the car alarm. But if the car alarm is at home, then you pay less attention, because you know your surroundings and you live in decent area, hopefully, right? Or if the car alarm is happening in Tenderline in San Francisco, you run to your car because you know that something is wrong. So the context injection on the fly is very important. And for that, you need a lot more compute. We need compute to help storage. And then we talked about that today, that in the controllers, some of the compute is being utilized from the controller to put the intelligence into the storage layer. And I think the compute is the key. Storage needs compute to be intelligent. Very interesting, because cloud, I've said number of times, cloud is code, code is now natural language, that requires compute. Yeah, and I think to both your points, I think bringing it to the flash modules, which they're talking about, and to the individual, and having it be this large distributed system, because data is distributed. I think that's, there's no company where all their data is in one platform or in one place. I think some companies would love that to be the case to make the money off it. But I think what you need to look at is, how can you look at and approach all of the data to bring it to where you need it? And I think it goes back to what you were saying around iceberg tables and going and looking at data formats and how you bring that kind of transparency and openness to it. And I think that's gonna be a huge key for them is how do you really get to that next level of data usability with the security, with the performance, so that it's the right place, transformed the right way? Yeah, I think the standards and openness brings the, we need to bring the new, sort of, we need to neutralize the different protocols for access, different types of storage. There are tons. One session, in one of the sessions, where I think Winston was talking about, we are applying virtualization to the storage layer. And once you apply the virtualization, you can do a lot with it. So you can neutralize the formats and the economics of the data, which is very close to storage, will come from that layer, I think, from the, by virtualizing it. By virtualizing it, you can change the formats and you can slice and dice the data. You can compress it. You can say, oh, this is only incremental data. Like, we need to only send this data further down where it needs to be sent. Or we need to store this data here versus their tape versus, you know, higher performance storage and so forth. So I think virtualization in the storage layer, it will progress. Again, coming back to the time horizon, in the very long term, I think we will apply a lot more compute to the storage layer, just like TPU we use today for security reasons and for network performance. We have offloaded the CPU into different sort of cards, if you will. I think we will do the same thing with storage. Oh yeah, there's, you know, historically a lot of waste in terms of managing storage with, you know, traditional, whatever, x86 systems or spent a lot of time not doing things that you could have a specialized, you know, process or do. But your point about virtualization, data virtualization, I think there's widespread acceptance now that data is distributed, you can't just, you know, it's the Jamak Tagani, you got to rethink your data architecture. You're not just going to shove it into one centralized container with one centralized group that manages it. Yes, they can manage that virtually. Even, you take even a snowflake, which is they put everything into our system, but they recognize that it's a single global instance that spans regions across the cloud. Certainly Databricks is doing the same thing. IBM is sort of, I mean, go back to mainframe global cysplex, has always had that sort of philosophy of having, you know, strictly consistent capabilities, but at remote distances. So that's happening. And now, you know, you bring, we haven't talked really much about edge today, but people talk about AI inferencing at the edge. They talk about real time. And that is a whole lot of data. Now, whether or not that data gets persisted, you will see. But I thought that's, I think, a really interesting thing that it wasn't talked about, but they're more or less doing it. If you think about how they're using AI in the flash modules to be able to do this intelligence, and that is actually those models are being trained in the cloud and pushed down in new firmware. So technically, it depends on what you consider edge. Also, you can bring it that way. And it's not, I don't think it's such a huge leap to say this is the direction they're going in. This is how we're going to build these systems to be more agile and more edge-aware or edge-like in that. Yeah, another great point, sir. I think another thing is that it's maybe in the storage company's interest to sell more boxes to us, right? But I think that shouldn't be the case. They should sell us more smarter storage, but maybe smaller boxes, but smarter, which where we sort of don't store most of the data, but some data, whatever data we need to be stored. Compression is a huge concept in storage, like how much you can compress and where we are storing it. That also matters, like you talked about earlier, tape is still relevant, right? So I think bringing again, coming back to the intelligence and storage, we still have a long way to go to bring intelligence into storage because we are storing a lot more data. We are storing exhaust. We're storing logs, like we just, by the way, one very important part, like you must have seen that interview from the open AI's chief scientist talking to Jensen. Jensen, yeah, yeah. Jensen of Avidia. Iliya. Iliya from. Yeah, Iliya from. Great interview, if you haven't seen it. It's a great interview. If you haven't watched viewers, you should watch that. So one of the key concepts behind Generator AI is compression. It compresses the data, a lot of the data. So one of the models with, which is not from Avidia, I forgot the name of that, but it just applies to the pictures, like you can paint the picture when you say, okay, just draw this picture for me or it will be used for movies. That whole data is compressed to two gigabyte file, two gigabyte. So world's knowledge about all the user interface or all the video rendering is compressed to two gigabyte. Can you imagine that? It's crazy, you can run that on your cell phone. So the compression is important. So from that compressed data, because after training, the models are very small, right? In size. So we are using that for inference. We will use that at the edge, right? So that means we need less storage for more intelligence. That's something we need to think through a little more when we talk about storage and new access. So I know we're tight on time, Ken, I wonder if we can extend this session a little bit longer because I want to unpack some of the things that Sarbjit said and maybe we'll just, we'll just go into the wrap from here if we can. You know, you talked about, first of all, you talked about smaller boxes. I mean, I think people want to consume, like in the cloud model, in the consumption model, right? I mean, that's, I think pretty clear and pretty much everybody will allow you to do that. The other thing that Ilya talked about in that interview was scale, that until they realized the importance of scale, it was like a breakthrough. And then we were talking about scale and GPFS today. And then that allowed them to, how did he say it, Sarbjit? He said basically to understand the human condition, right? And that's why it hallucinates because it's got all this data in there, right? What he was saying was that, like the question was why we were always failing in this AI, like all of a sudden, what happened all of a sudden? We succeeded big time, where was the magic? It was the amount of data thrown at that. We were doing smaller experimentations in academia. And academia doesn't have much data, right? So they then, the likes of Jensen was because they want to sell chips, right? They said, okay, let me help you. We have billions of dollars. Like let me give you tons of data. And he partnered with the University of Toronto scientists, data scientists, and here you go. Like it's a big bang in computing. Now the other thing I want to pick up on is you and I were having a little Twitter, I don't think it was threads. I think we were, yeah, we were on Twitter. And we were talking about moats. And I was saying, well, the moats are going to be the quality of the data, the propriety of the data. And then all the other things that matter, go to market, brand, reputation, and execution, dot, dot, dot. And then Crawford chimed in, Crawford Del Pret, CEO of IDC, I thought he had an interesting comment, which is like, it's also the currency of the data. This is David Floyer's big thing. It's like data value declines significantly over time. Now, I'm not saying you don't want to go back and reach back and look at history, you do. But the real time nature, the more current the data is, generally speaking, it's a power law, the more valuable it is. Right, and I think that's why some things will be done in certain places and the models will be continually trained, right? And that are going to be continually refreshed because they become out of date and they're even seeing that with chatGPT now, where it's declining in accuracy and speed in certain aspects of what it tries to calculate as they make them smaller and more compartmentalized. I think that's- Entropy, it's like Andy Walls was saying. It is, and I think that when you start to look at all of these, how people are gonna use AI and how companies and storage companies build it in, they need to be able to understand that. And this is, I think the storage companies in particular have been doing this because if you look at it, they had to understand how often a drive was going to fail and they wanted to know that that drive was gonna fail and they would ship you one before it failed because they knew the parameters that they were looking for. Back then it was called analytics and things of that nature. Now it would be called modeling and AI and all of that fun. But I think that's the key is that the data where it lives, how it's applied and how it's refreshed is gonna continually put stress on the storage that companies have. Yeah, during the SuperCloud 3 session, which we had, we talked about the specialized model for different verticals, but also horizontal models for all verticals. For example, some part of the security for companies is very horizontal, like lead-os attack is horizontal, like anybody can get it, right? So we need a solution for that, for example, right? But some part... Identity. Identity is horizontal, but some parts of those security are very vertical specific as well. So I believe in the long-term, we will have the models which are horizontal by the capability in IT stack, if you will, in security scenario and the story scenario as well. But some will be very specific to healthcare, form, auto, so they will be very specific to that. So we'll have a... Data, data, data, data, all specific to those industries. And the models will stick to it, like mRNA vaccine that works. It's sticking to us right now. Yeah, it's sticking to the protein. So it will be like a mRNA kind of a mechanism where we stick the AI to our moving blood flow, which is data. So it's been a good day. Again, we love this format live in our Palo Alto studio. We inject prerecords. We have one more prerecord. I interviewed Chris Maestis. He had this concept of this information supply chain, which of course you've heard that before, but how a global data platform turns storage for AI into an information supply chain. That's a really good session that we had there. And Sarbjee, appreciate you sort of watching today. I know you're going to be socializing it in your massive network like you do Rob, Rob Streche, co-host. And of course I want to thank IBM who made this possible, brought in its ecosystem, its experts, and we're going to continue this conversation. Of course all this data is available as replays on demand at thecube.net, it's on YouTube as well. So we want to thank you, reach out, check out siliconangle.com. That's where all the news is going to be. And hit us up on social, LinkedIn, Twitter, sometimes Facebook, sometimes threads, Instagram, we're out there, you know where to find us. X now, X now. X now, right, X, that's right. He paid nothing for the logo, I understand, but is willing to pay somebody who has the X handle. Handle there, right? Somebody just took it. Thanks, it's great to see you in person. Yes, absolutely. Great to have you. Thanks again for watching everybody. Keep it right there for the next segment on AI into an information supply chain. Thanks for watching.