 Five, four, three, two, one. Welcome back everyone to Cube's coverage here at the NVIDIA GTC. This is a conference for the error of AI. I'm John Furrier with Dave Vellante who's here with me. This is the Cube on the ground mobile run gun kit. And we've got Cube alumni Scott Baker who's the CMO and the storage marketing among other titles he has at IBM. Scott, great to have you on. Always great to get your perspective. You know, IBM has really been moving the needle on the infrastructure as cloud. And now AI set the table for software defined, large scale, scale up, scale out. For these new AI infrastructures that are emerging that are going to power these next applications. So you're seeing a whole new systems revolution going on, but it's still the same game. It's compute and networking. Maybe more emphasis on the storage and networking as usual. So real, real interesting market dynamic. We've talked about this before. Welcome back. Hey, thanks very much. It's a pleasure to be here and to be with you again, John. Always good to see you. And you're absolutely right. When we think about AI, you know, we like to look at it as just another workload. It just happens to be both compute intensive as well as capacity sensitive. The ability to push that data back and forth between whatever that cluster of compute layer is requires heavy reliance on the networking. And it certainly requires a storage infrastructure that's capable of delivering that performance. I want to get your thoughts on storage because, you know, Dave and I have been covering storage forever. I mean, you got the pure storages of the world, NetApps, they've been around doing file, blog, now you got Flash arrays. They seem to be flat footed a little bit with AI, it seems, because AI is a whole mother requirement. I'm not them specifically. I'm just talking about in general old school storage. And maybe I don't want to get in the conscious about the different vendors approach because they have reference architectures. Okay, I see that. That means they don't have anything. But there is a difference. There's a distinct architectural shift that's impacting storage. So you're either on the wrong side of history or the right side of history. I want to ask you, what is that right side of history? If you had to say, hey, I got to do storage like this in this paradigm, what does it look like? That's such a trick question, right? It's always about having the right tool for the job, right? There are times where, you know, a traditional Flash primary storage array would work well for an AI workload where the size of the language model that you're working with is very well-defined and very consolidated, if you will, can be consolidated. But by and large, when we think about large language models, the emphasis is on the word large, where they're looking for distributed types of storage where capacity becomes maybe even a bit of a bias when you're thinking about the kind of storage that you want to deploy. The other thing that we look at is mobility of information, right? Large language models need to stretch from the edge of the business into the core and out to the cloud. And that's a natural reliance on protocols like S3 or cloud-based architectures that you can deploy on-prem, allowing that data in the application stack to move as it needs to, you know, from the edge of the business to the core to the cloud. A lot of times what we're seeing at IBM is that there is a bias toward object-based storage or distributed object-based storage. I think there's some emphasis on the use of files, but what we're talking about here is the ability to store large amounts of data across clusters that may exist in the same location or stretched out around the organization itself. So one of the big, obviously, we've been covering this for years, but compute and storage has been separated. That's opened up a lot of innovation. If you look at the AI right now, you're seeing a couple of different trends. One is the software-defined, I've heard Jensen even use the word software-defined as he talks about what's going on in the GPUs with the spine. He's got all these, the systems. It's basically a system, it's not even, it's a bunch of GPUs connected together with Switch. So you have a software-defined paradigm that's not new and physical. So in the storage world, this is a challenge. Can you explain the difference between the software-defined approach and the physical approach? You mentioned a little bit there, but it seems to be, you got to do both, you still have physical storage, but how is that evolving? What's the right side, what's the right story there? Sure, no, it's a really great question. When we look at developing technology at IBM, we take a software-first approach, whether that's on the IBM Cloud side or the IBM infrastructure side, how do we deliver the data services to the organization if they want to repurpose their own investments? In the case of software-defined, one of the nice things about that is that it allows organizations to repurpose investments that they've already made, in effect taking advantage of maybe an old Hadoop cluster or some servers that they've got laying around, layering on top of that those data services that are provided as software-defined storage. Software-defined storage also makes for a really great consumption model for organizations. Being able to go to the cloud and receiving that software layer and working with that directly as a piece of software means that if you move the capabilities from the cloud to on-prem, then you're moving with that data and application stack, those data services that I had mentioned. That's one of the things that we really looked at at IBM as we thought about products like our own IBM storage scale, being able to deliver that as software or on very purposely designed hardware that's absolutely architected in a way to give you the highest degree of throughput as well as the highest degree of read-write access to service out those GPUs. I think the other thing too to mention, whether it's software-defined or it's physical hardware, the integration that each vendor makes or chooses to make with companies like NVIDIA, i.e. GPU Direct, to give the absolute fastest performance available to the compute cluster that you're working with for the data in question that you're using. Well, you know, more compute, more clusters, better performance, means bits move around faster, means storage needs to map that. I see your IBM shirt says NVIDIA partner on the side, nice little logo there. I got to ask you about the relationship with NVIDIA because you guys have been doing a lot of work with infrastructure on the storage side, so that's changing a lot. You're seeing these AI systems here, certainly they're pretty impressive and I think it's a high bar on the AI side that NVIDIA's set on that side. And you get the cloud, you guys are both in both. You guys have AI with Watson X. Jensen said there's two paths to AI in the enterprise. Get apps built on with an AI API, like a NIMS, which is basically RAG and augmentation and then other ones is getting the IT data that's exhaust today and turn that into gold. And he says, quote, they're sitting on a gold mine. So I have to answer the question. What is IBM doing with NVIDIA one and two? How are you guys solving that problem where you're enabling apps to be built and you're giving the IT platforms capabilities to leverage their tools that they have that's thrown off all this data exhaust and turning that into value, whether it's RAGs or some systems that way. So what are you guys doing with NVIDIA and then how are you doing to enable the enterprise? Yeah, so IBM has had a really healthy NVIDIA partnership for quite some time. I use the reference GPU direct. It's something that we've integrated into our storage systems. We've also looked at delivering GPU as a service from an IBM cloud perspective so that organizations who maybe can't afford to purchase the size and clusters that they need would have access to those and a consumption model that would make the most sense to them. So that partnership that we've developed has cultivated into having reference architectures for different super pod deployments whether you're talking about maybe H100s as your basic compute structure or not, all of that connected to that storage scale system that I'd mentioned before. But that's an infrastructure story. For me whenever we think about AI, it's about how quickly can I get to delivering value, not about how long does it take me to get to day two operations about connecting the different parts of the infrastructure together. To that end, one of the things that we've done is we've really taken full advantage of the WatsonX product line. WatsonX.ai.data.governance so that we've got a full platform approach to delivering that kind of infrastructure backed by all of the technology that I'd mentioned so that organizations can actually step in and start working with defining what their AI model looks like and how they integrate that into the business. That WatsonX technology that's containerized by nature runs right on top of Red Hat OpenShift which sits right on top of what we affectionately like to say, you know, maybe WatsonX in a box. I'm probably not supposed to say that, but that's to me a turnkey appliance. Turnkey basically, basically making a turnkey. Absolutely, and when we roll that into the data center, that's WatsonX sitting on top of Red Hat OpenShift running on IBM Fusion that has something like storage scale on the back end. That is essentially a cloud architecture sitting on prem for rapid AI deployment and development that you could replicate as software in IBM Cloud. Yeah, well I think, I wouldn't, first of all, good plug there for WatsonX, but I would not downplay the infrastructure changes on the system side because it's pretty amazing. It's not too shabby, that's going on there. In fact, if you look at the major AI momentum right now, look at NVIDIA, Broadcom, all these big infrastructure providers building these AI clusters, the access on the consumer side. So the Facebooks of the world, Apple, ByteDance, you name all those big companies that have massive hyperscale, they're kind of indicating what's coming and that is these AI systems. Now it's heavily optimized for compute memory because it's actually in the weeds, but you can almost extrapolate and connect the dots and say, hey, there's going to be storage systems coming that are going to look different. So you have to look at this and say, in the enterprise AI, not consumer, maybe the workloads aren't as customized or purpose built given these billions of dollars of infrastructure on the table for the consumer, big players, but when you look at enterprise, there's a lot of money involved. So workloads need to run almost with custom silicon and or custom clusters. So there might be the rise of these super clusters, these super storage environments where you have to rethink that, what's your vision on, and it's kind of out there, it's kind of an abstract question, but if you want to go to connect the dots on enterprise AI and storage, you got to have storage clusters that match the speed and performance of these new AI compute and memory clusters that are optimized for those AI workloads. I mean, they're different. Are they different? Kind of a concept question, what's your reaction to that? Yeah, so what's interesting to me is you have to also sort of balance that against the organizational requirements around sustainability, right? So whatever solution you decide to put into place has to deliver on the performance requirements you're looking for, but at the same time meet those ESG goals that the organization has. So it's really going to think about the super clusters. It's not so much about how big of the storage environment we can sort of create in a single offering. It's how easily it is that we make it so that you can connect multiple storage environments together. We often refer to that as the global data platform, this idea that we can create an aggregate view of your entire digital estate without moving data around. In fact, when we talk about the storage scale system just to give you some context, 4.5 petabytes and 4U effective is what we can deliver. 4.5 petabytes, that's a lot of data that you can store directly on that disk, right? Or that array. I've also mentioned the throughput, the read-write speeds, et cetera, but we're doing that at a 70% reduction in the total amount of power that's being consumed. So as we begin to think about the future of building out these super clusters of storage, interconnectivity becomes key, self-optimization becomes key, and I think the storage itself has to be integrated or has to have AI integrated into it so that it can move and work as efficiently as the models it's responsible for supporting. That's a great point, actually. We mentioned software defined earlier versus physical. You just brought up the whole sustainability. The constraint that everyone's designing around is power and cooling. Power and cooling. And so if you look at the software opportunity, it's massive. It is. So, okay, we'll have physical drives that'll get better, faster, cheaper, but you got to run software on and that's the opportunity because you have to manage the power envelope within the rack. Right. Power, energy, watts. I mean, how do you squeeze more out of the hardware? Again, software becomes key. All right, so let's get back into the tactical. So what are you guys offering today? So when a customer comes in and says, hey, look it, I want to position myself. We're not repatriating that much. We're basically net new investment spend on premises and edge. I got to have an architecture and a storage solution. What do you guys say to that question? Yeah, so a lot of times it's, what's the intent that you plan to do with that data, right? Obviously we're at NVIDIA, so let's just assume that what you're referring to is AI. When we look at the storage scale system to address the distributed file and object needs that most organizations will have for large language models, we have multiple offerings there. So we have a 3,500 that would be more entry-like in nature, the larger 6,000 to deal with the larger capacity of storage that you need to be able to use to hold that data. Those are two offerings that we could bring to market very quickly. But then again, it depends on what the infrastructure looks like that we're moving into. So you had mentioned on premise, we do have organizations out there that rely on IBM consulting and ask them how do we in effect create a hybrid by design AI architecture that gives us the best of both worlds, takes advantage of the compute capabilities that are provided as a service, maybe through IBM Cloud, connected to something like an IBM storage scale system that's holding onto the data at the edge or the core and then blending those together with other parts of the business. It just depends on the organization we're working with. That's awesome. Obviously the AI is the hot trend. Quick plug, what are some cool things you've been working on lately since last time we saw each other a bit of few months? What cool things you've been seeing emerge, observations in the industry, things you're working on. Give us a quick highlight reel. Just a quick plug. One of the things that IBM prides itself on from an innovation perspective is what we do with computational storage. The same drives that we refer to as the FlashCore module actually work in our flash system for primary storage as well as the scale system for that distributed file and object offering that I was mentioning. But because it's computational storage we can actually push AI and machine learning down to the drive and actually trend the entropy with respect to how data changes from the host that's accessing it or writing to it. And we begin to look for patterns in that data. So one of the things that we announced recently was this FlashCore module four. Again, since the last time you and I talked where we're doing hardware accelerated inline threat detection to look for patterns that would be indicative of a cyber threat and being able to respond much quicker. And as we begin to standardize the use of FlashCore modules across all of our storage portfolio we're going to take that same AI and hardware acceleration and we're going to embed that throughout the on-prem infrastructure. And use of data is critical in that piece there. Absolutely. All right, so what is the IBM storage strategy? Since you're the CMO this is a quick test for you on messaging. What is the strategy of the storage IBM storage? You know, when I first joined the company I was asked to develop what that strategy would sound like in a news bite and it hasn't really changed. I think when you craft a strategy and you talk about it you build against a vision that you're after and that vision is pretty straightforward. It's literally to make every bit of the organization's data available to them in the most insightful and secure way possible so that they can take some kind of informed action on it. And then from that vision, how do we deliver that? We deliver that across three different categories. It's IBM storage for data and AI, IBM storage for hybrid cloud, and IBM storage for data resilience. How do we make the data that you rely on assured from a veracity and an accuracy perspective available to you without worrying about threats? And then how do we do that in a hybrid environment where some data will run on-prem, some data will run in the cloud and maybe some will actually run in a mix between the two and then build for an infrastructure platform that's capable of supporting AI and future workloads that are just as compute and data-intensive. Well Scott, great to see you, great to have you on for the quick update and thanks for the industry commentary, really appreciate you. We'll be seeing you in the studio for the super studio session in Palo Alto. You got it. And in about a month or so. We'll see, actually this month. Coming up, that's right. That's right. We're going to unpack it all. Thanks for coming on theCUBE. We are here at NVIDIA's GTC conference for the air of AI. A new inflection point is here. This is attracting not only just the hardcore developers in AI but mainstream investors, investment bankers, startups, a whole nother level we're going to here. It's the real deal, it's totally legit. AI is here, AI systems, AI storage, AI system solutions and applications will all be running on with AI. It's going to be great. We're in a systems revolution. Of course, we're covering on theCUBE. Stay with us for more coverage here in San Jose after this short break.