HPE Discover 2024 | theCUBE Research Analysis

Clips
News
More from HPE Discover 2024

David Linthicum

Principal Analyst

theCUBE Research

Rise of heterogeneous architecture and AI integration highlight evolving cloud strategies

Cloud repatriation is becoming a key focus as businesses reevaluate their cloud strategies in light of advancements in artificial intelligence.This shift is not just about moving to the cloud, but about integrating a mix of hardware platforms within the cloud. Enterprises may have oversold cloud utilization over the past decade, according to David Linthicum (pictured), principal analyst at theCUBE Research. Now, as companies seek better cost-efficiency and scalability, they are considering a more heterogeneous architecture to achieve the promised benefits of the cloud.David Linthicum, Dave Vellante and John Furrier of theCUBE discuss cloud repatriation at HPE Discover

play_circle_outline Cloud migration failures and oversold promises

play_circle_outline Maximizing Cost-Effectiveness: The Importance of Cloud Application Optimization and Workload Repatriation

play_circle_outline Comparison of on-prem growth rates to cloud providers

play_circle_outline Importance of interoperability and heterogeneous networks in modern enterprises

play_circle_outline Simplifying Complexity: HPE's Approach to Prepackaged AI Solutions for Targeted Use Cases

Info

theCUBE Research Analysis

David Linthicum

Principal Analyst theCUBE Research

search

>> Welcome back, everyone, to theCUBE's live coverage here in Las Vegas. We're on the show floor at HPE Discover 2024. I'm John Furrier, host of theCUBE, with Dave Vellante, theCUBE Research and David Linthicum with theCUBE Research. Also, luminary speaker here at HPE Discover '24 on the big stage for Tech Talk right next to theCUBE. David, great to see you. It was nice to see you holding court on stage here at the Tech Talk arena next to us.

David Linthicum

>> Yeah, it's great to be here. It was the perfect infrastructure. I could look at you guys, who were doing reporting, and you could look back at me. So we had the whole corner.

>> You got my attention. I love those slides. Okay. Cloud migration failures. Really talking about the on-prem movement back... or, say, never left.

David Linthicum

>> Right. Well, there is some back-

>> Distributed computing paradigm. Give an overview.

David Linthicum

>> Yeah. Ultimately, as we've been covering in theCUBE, there's really a movement into heterogeneous kinds of architecture. So people aren't moving just to the cloud anymore, and certainly with AI, but they're moving to a mix of hardware platforms in the cloud. And there's reasons to use both. And I think what happened over the last 10 years is that utilization of the cloud was probably oversold within the enterprises. Who would've thunk that? Well, it was the fact that we were looking for scalability, cost efficiency, all the thing that cloud is able to do. And cloud is able to deliver on some of that, but where they weren't able to deliver on, cost-effectiveness. So what happened over the last 15 years that we've been migrating stuff into the cloud, we're not getting the value that's coming back into the enterprise from utilization of these cloud resources. So people are starting to understand that. I view the reckoning moment being 2022. Everybody moved to the cloud in 2020 with the pandemic. Suddenly these big, huge cloud bills came in once these things are operationalized. And I would get call after call after call that said, "What the heck's going on? This thing's a lot more expensive than I thought it was going to be." So when you look at why-

>> You can't just turn it off. You can't just turn a switch.

David Linthicum

>> You can't just turn it off.

>> Yeah.

David Linthicum

>> Yeah, you got all your applications in there. And yeah, here we go. But what happened was is that no one modernized the systems that were moved. They just lifted and shifted them. So it was very easy. And Cloud actually is very good at this, at providing good platform analogs for LAM stacks and even AI-based platforms. We're able to pick it up and push it over to the cloud and have it run in that environment. And the problem is that operational inefficiency is going to cost you money. In other words, it's like plugging a very inefficient electronic device into the grid. It's going to cost you money. And so organizations saw this. And this was kind of like a dirty little secret for a few years, where I watched all this repatriation going on. No one was talking about it because repatriation is kind of code for, "We screwed up." Oopsie daisy, hit the reset button. But it got enough momentum. We're putting some emphasis behind it, some intelligence, some methodologies, frameworks, how to make this stuff happen. So it's really right-sizing or normalizing of the application workloads.

>> And the cloud people called it cloud optimization.

>> Well, so we over rotated, you're saying, during the pandemic?

David Linthicum

>> Yes.

>> And then Andy Jesse says that cloud optimization has attenuated. Is that because people are spending more and not optimizing, or because they're moving back on prem?

David Linthicum

>> I think they're moving back on prem. And when he's looking why they're doing it, he would argue that you did not optimize your applications for the cloud, which means leveraging cloud-native features, things like that. That's tremendously expensive, to do that. You have to hire consulting companies and service firms and hire very expensive talent to modernize these applications, moving them to container-based systems, serverless-based systems, things like that. Most organizations didn't do that. Very few applications were modernized for the cloud. And so the way that the cloud providers fought back against it is they say, "Well, you guys haven't optimized in the cloud." So we heard the O word at the last re:Invent. And the reality is the O word is just them admitting to the fact that you're going to have to change a significant amount of your application in order for the thing to run on the cloud at a rate and an operational spend rate that would be reasonable.

>> Well, they'll also mean storage tiering, which, that's like a little sliver of cost reduction. Okay, not a huge deal, but what if you do optimize? What's your experience? If somebody moves to the cloud, they bring in your former firm or others, and they optimize. What is the TCO comparison? Because you talk to the on-prem guys. They say, "Well, we did this study with IDC and it shows that we're more cost-effective." And then you go to Amazon and they say, "We did this study with IDC and it shows that we're more cost-effective." And the customer's left to say, "Well, wait a minute." So what's the truth?

David Linthicum

>> It's the it depends answer everybody hates. So ultimately it's how the application was architected, how tightly it's coupled to the database, how it's bound to the processor. All these very geeky things that come into play. And so you can take this application that was built 20 years ago that didn't consider the optimization on the particular cloud provider. And so what we're doing is just basically putting it out on the platform in the cloud. We didn't have to optimize in those days. We had the entire server. It was at our disposal. So now when we put it on the cloud, they're charging us for all of that under-optimization within the system. So optimization means rebuilding the application, in many instances from the ground up. So it's going to leverage much fewer resources in a more optimized weight. And actually, if you do that right, you can move an application into the cloud and get it to burn 1/3 of the costs. And so the opportunity's there, but in many instances it was $1 million dollars an application and the enterprise is like, "Yeah-"

>> So big upfront cost. You can eventually get there. Then you don't have as much headroom and flexibility with that. The leftover capacity is really what I'm hearing here.

David Linthicum

>> That's absolutely right. Absolutely.

>> Maybe virtualization could help, Dave. No, only kidding. Ajay Patel, when he was at VMware, said on theCUBE. I'll never forget this, Dave. He said, "Oh yeah, I'm a middleware guy. The cloud is just hardware." And I'm like, "Actually, you know what? You're actually not wrong." The benefit of AWS was storage and servers, EC2, and queuing. Those are the building blocks. If you wanted to lift and shift, you can get hosting on demand, and variable pricing. Great for data processing. Okay, check. But what we're seeing here at HPE, David, is that you can cut hardware. You can just put another rack in for AI so, for the folks that want to play around, that seems to be a good pitch.

David Linthicum

>> It is a good pitch.

>> Yeah, because they know data center. There's a power envelope issue. They're addressing it. That's a sellable value prop.

David Linthicum

>> What T-shirt size do you want?

>> It's like, "Okay, I know how to deal with racks. Put a rack in, and that's my AI rack." Now the question is, will that be scoped for the workload? So is it the right hardware? So we're in this hardware scoping stage of generative AI. How do you see that playing out? And how does that factor into the cloud, which is essentially hybrid distributed computing?

David Linthicum

>> The cloud is always going to be a more convenient place to run AI because the ecosystem comes along with it. In other words, I get to vector databases, I get to odd middleware layers and streaming systems, things like that all on my portal. I'm able to assemble it together. It's just going to be more expensive to run. So that's going to be the trade-off. So what we're seeing here at HPE Discovery is that we're in essence bundling this stuff into an AI framework to compete with the cloud. You look at what they announced today in the keynote I was in. HPE Private AI, Private Cloud AI, is basically trying to replicate the convenience of the cloud, but doing so on premise within a data center-based system. So we have the advantage there. Number one, we own the hardware. Some people like to do that. But the price of hardware has just come down like crazy in the last 10 years, and so that blew a huge hole into cloud computing. So they're offering alternatives and not a DIY kind of situation where people have to assemble it themselves. I think that's where people get nervous. And if they're offered bundled solutions that are racks, that looks very cool to them.

>> Yeah. And Dave, what's interesting about Nvidia is that, and we talked about this earlier, they could have an Amazon-like situation here where they can get the benefit of HPE's enterprise because their NIMs are proprietary. That's the dirty little secret for the model stack. So they say, "Hey, you want to build your own RAG and some inference? Use NIMs, our Nvidia Inference Microservice, which is essentially proprietary." It's theirs, so if they get them locked in on Nvidia-

David Linthicum

>> But it's yours. But it's your-

>> All the enterprise get to have their own models. So the question is, how does that fit in? Because now you have, it feels like a network operating system, in a way. It's like they're giving a NOS to the enterprise. And they'll probably use it. My point is I think that might be the biggest land grab opportunity for Nvidia to jump in on the model stack and saying, "Hey, we'll own the model stack."

David Linthicum

>> Well, because people-

>> That's what I'm seeing.

David Linthicum

>> Because people are concerned about open source licensing. And some of them, read the fine print, Meta says, "Look, we reserve the right to pull the rug out from under you."

>> Well, no. No, they got the data to worry about. So if I'm an enterprise, I might say, "Hey, you know what? I'll go with Nvidia and HPE because I want to protect my data. Why do I want to put my model into anything else that I can't control?"

>> Well, exactly.

>> So okay, NIMS, Nvidia, spine, leaf, horsepower. They closed.

David Linthicum

>> iPhone?

>> Exactly.

David Linthicum

>> Oracle?

David Linthicum

>> Nvidia's going to make more money on selling hardware with partners like HPE and other folks there as well than they will in the cloud. And I think you absolutely hit the nail on the head. If we're going to leverage the core Nvidia features, we're going to be locked in. There's that word, into that particular platform. And if we're building these net new generative AI systems, then that's going to be an HPE movement for the next 10 to 15 years.

>> Yeah.

David Linthicum

>> And so we're building these things for the long term. They're going to take a tremendous amount of hardware resources in doing that. I think Nvidia sees the land grab as well, and that's why they're working with partners like HPE.

>> Well, and also AMD and Intel aren't sitting on their hands right now. Nvidia's got a great run right now. Their market cap's at all time high. Why not just nail down everything you can right now? And one of them is the enterprise AI market.

David Linthicum

>> I would be doing the same thing. That's why we're seeing the Nvidia CEO at every conference out there. And he's in front of everybody now because he knows he has to grab the land before eventually some of the innovation catches up and his competitors catch up. And he's not going to have the monolithic share that he has now.

>> I want to come back to repatriation, if I can, for a moment because we know anecdotally, when you talk to customers, you got the phone calls. And we talk to people. They go, "Yeah, we're going to move some workloads back." Yet when you look at the numbers, help me square this circle. So Dell last quarter, let's use Dell and HPE as a proxy. We'll throw in Supermicro for kicks, but they're not really an enterprise.

>> Yeah. They're definitely doing well right now.

>> But they're selling to hyperscalers and big MSPs. But taking Dell and HPE. Dell last quarter grew 6%. They're promising 3% to 4% growth. They're a $95 billion company trading at 1X, they're a $95 billion value. HPE grew at 4%. They're promising only 1% to 3%. I think that's conservative. They're a $30 billion company trading at 1X revenue. And then you got AWS growing at... this is just hardware, by the way. Stripped out, tried to anyway, all the 365 and other stuff that Microsoft throws in. AWS grew at 17%. Azure, almost 30%. GCP, 28%. So help me understand, you got the cloud guy's growing much, much faster.

David Linthicum

>> Right.

>> If there wasn't repatriation, would Dell and HPE be growing at 0%?

David Linthicum

>> No, I think they would be growing at a more rapid rate because of the utilization of AI. You wrote in your analyst brief on Friday that really the cloud is leading the way, because people see that as the path of most convenience for building these AI systems.

>> Yeah.

David Linthicum

>> But I think that the on-premise stuff is going to catch up fairly quickly. And they're going to see some huge growth as well, which is also going to be driven by AI. So everybody gets the AI tide that rises all ships, brings everybody forward. The cloud folks have a lot of momentum. Huge marketing momentum right now. They have the eyes and ears of the CIOs out there. So they're going to hold that position for as long as they can.

>> I think you're right, by the way, which suggests... of course Dell has run up a lot in this past year. HPE has not run up as much. If you actually look at the stock prices of these companies over the past year, Dell is up 173% as of last Friday, Supermicro's up 250%. But forget those guys for a second. HPE's only up 30%, but they're up 22%, in the last 30 days.

David Linthicum

>> Right.

>> So HPE is catching up. And so my point is, I think these growth rates that Dell and HPE are promising, basically 1% to 4%, is really conservative.

David Linthicum

>> yea-

>> And I think they're going to probably beat these expectations.

David Linthicum

>> Because there's a massive demand for more hardware right now.

>> And they have visibility, and big backlogs.

David Linthicum

>> The on-prem growth is happening. The AI bug is all about, if I'm going to have a neural network, I need different hardware. You're not going to get more out of an x86. You need the systems around a ethernet. And you heard liquid coolant's going to be in the home of the system architecture. So I think the AI wave is interesting, David.

David Linthicum

>> NVLink, CUDA, NIMS, NEMO. Forget it.

>> Well, David, I want to ask you because you got the new episode out, the AI Insights and Innovations with David Linthicum-

>> Which is killing it, by the way. Congratulations on-

>> Great job on your podcast on SiliconANGLE. Of course you've got your own. You also wrote for InfoWorld. The Whiplash article was hot. Just posted. But as you look to chronicalize the AI innovations and insights, there is a new architecture model emerging. What's your strategy with the podcast, the show? What are you going to be covering? Actually, you wrote the Insider's Guide to Cloud Computing, which is a nice backdrop, by the way, in your camera angle. It's in mind in my podcast, too, by the way. So check that book out, Insider's Guide to Cloud Computing. But as AI starts, we start documenting the progress. Where are we on that progress bar?

David Linthicum

>> We're not very far. We're at the caveman stage in terms of where this stuff is moving into. We're just figuring out... by the way, AI's been around since the '50s. We're still trying to figure out how to optimize it and make it work. And here we're in 2024. So in terms of generative AI, which is really the restart of AI, we're at the stage where we're trying to figure out what platforms to run them on, the use cases we need to use, how much this is going to cost us. Is this is going to burn up the planet? All this stuff that really haven't thought of. Everybody gets real excited about generative AI, hasn't looked at the hard planning it needs to occur. So the use cases are the biggest thing right now. In other words, businesses need to figure out how to use this stuff, and for what purpose. And it's funny. Even the vendor shows that we go to and things like that, they haven't understand the use cases that are there. They don't figure out how to sell this stuff. They know it's hot, but that needs to be the hardest part. That's going to be the hardest part to solve. The other thing is understanding we're moving to a heterogeneous platform. We're moving to AI on our iPhones with Apple Intelligence and all these other things that are coming forward. What does that mean for the enterprises out there? How are you going to leverage this effectively? And so it's looking at the innovations where things are coming to, but not just looking at them and reporting what the innovations are, but looking at the CIO and looking at the CTO and then these enterprises out there and telling them how it's going to affect them and how they have to change behaviors to make themselves a winner.

>> I'm glad you brought up use cases because this is a strong show. The Sphere, the keynotes, the Nvidia partnership, the OpsRamp stuff, the ecosystem growth. All that, very high marks. The one missing piece that I see is HPE is not talking... They're a $30 billion company. They're not talking about how their dog-fooding AI inside the company and what use cases they have. How they're prioritizing, where they're potentially seeing ROI. They haven't talked about that at all, the dog-fooding or drinking your own champagne. I think that's one area where they could really help customers. What are your thoughts on that?

David Linthicum

>> I think they could. I think ultimately I would've loved to see... in the keynote I sat in today I heard the guy from Mercedes who was talking about a storage solution. Okay, I get it. That's cool. But I'm talking about a killer AI application that runs on their stuff that they can point to that they're even doing internally.

>> Yeah.

David Linthicum

>> In other words, we built an engineering system, we built a deployment system, we built a BCDR system using this AI platform. I don't see those use cases from them. You do see it from Amazon, you do see it from Google, and you do see it from Microsoft and the other AI players that are out there. But I would love them, and I love the point. Eat their own dog foods is exactly right. So in other words, how are you using it? How are you successful with it? Therefore, how are you recommending that your customers be successful?

David Linthicum

>> Yeah, we brought the AI to the data-

>> Well, I think also, maybe a bad choice of words, but spoon-feeding their customer base what AI is with Nvidia. That's a big thing to grok. One of the things that I think they're hiding the ball on is the hottest companies right now are Databricks. That came out of the failure of Hadoop, and Spark made that happen. Remember, they bought MapR was the Hadoop for the enterprise, which was different than... so they have a DNA in this data that became Ezmeral. They're not telling that story.

David Linthicum

>> Well, but they're integrating it. They're not really telling the story. You're right. But they are integrating it.

>> Well, it comes down to use cases.

David Linthicum

>> It's their data fabric. And it does come down to use cases. You're right.

>> If you're going to do a use case and show compute, you got to show some meat on the bone. That's where the data piece comes in. To me, they don't have that yet. And I'm not sure if that's just, they didn't want to confuse the market, or they don't have it.

David Linthicum

>> They did represent a data lake that's attached to their private cloud system.

>> Ezmeral.

David Linthicum

>> Yeah, absolutely. And I think that's going to be fine. Sometimes that can be problematic because everybody doesn't want to use that particular-

>> Bring your own data lake. Right.

David Linthicum

>> People want to bring their own data. They want to bring their own licenses. They want you to run the core AI systems. And so if you're going to show me a framework, show something that's going to be open and extensible, and we're able to plug this stuff in. And I think that's what's missing from a lot of the vendors. Even if you go to AWS, and I guarantee you re:Invent's going to be like this, they're going to show you a stack that's only their stuff, nothing else, again, and don't show you how to make all your various systems work and play well in the environment. Now, they are open here and they're showing you how the integration occurs, but that's where the war's going to get fought. The complexity, the integration of it, the ability to have this information coexist.. We're not going to start moving stuff to proprietary databases just because you tell us that's attached to your AI system.

>> Yeah. Remember back in the old days of proprietary NOSs? I'll bring it again. The big buzzword was interoperability. Heterogeneous was happening and those networks were happening. That's happening now. I totally agree with you on that. I think that's the key. What would you advise HPE if they asked you? We're the board. We're Antonio's management team. David, advise me, what should we do?

David Linthicum

>> I would tell them to figure out the complexity angle and also figure out how to prepackage AI in such a way that it's going to be a bundle that people are going to consume. I saw a pretty good start today. They're going to have to build on that. Lots of missing pieces, lots on how you deal with inference engines, all these sorts of things. What are the 10 based architectures that I would base your system around? And how do I build those architectures in your particular environment? And I would publish those and I would push those out, and I would have, next year at this show, have one, two, three, four, five, all the way to 10, where people can walk around and figure out if these architectures are going to be right for their tech. And the ability to bundle it and come with a prepackaged solution has a huge amount of value because people are moving to cloud because it's more convenient to build AI in cloud because they have pre-configured ecosystems in the cloud. Make that happen within HPE. Equipment. Go after the architectures. Go after the 10 best use cases. That's what I would be telling HPE.

>> You're right.

>> Final question. I know we got to wrap because our next guests are here, but since you're here, you had the great talk. We were watching you at the Tech Talk. What was the reaction after the event? Because this always happens after you're done talking. Big crowd. I saw a big crowd there. So I'm assuming people came up after and said, "You did a great job," or, "You did a bad job." What did they say? What was the feedback? Not good job, bad, but would they say, "You're on point?" "You missed something?" What was the tone? What was the sentiment?

David Linthicum

>> I'm so glad you said that because no one has said that. And the ability to look at repatriation and the value of cloud and the ability to normalize the value of cloud within the modern enterprises. And there were CIOs over there, CTOs over there, high level individuals who make those core decisions. And quite frankly, they're confused.

>> Yeah. So you spoke reality to them. To them it was, "Thank you for giving a dose of reality?"

David Linthicum

>> Right. I spoke truth to confusion.

>> Truth to power who was confused, so confused power. Truth to confused power.

David Linthicum

>> That's exactly right.

>> Hey, that's what we do. We try to make people simplify the message. Of course we're bringing you theCUBE data here. CUBE Research is on the ground in force. SiliconANGLE team is here as well. theCUBE and SiliconANGLE team coverage again, getting the most of data in real time for you. We'll be back with more. I'm John Furrier. Dave vellante and David Linthicum and Rebecca Knight, Rob Strecce, Bob Laliberte. The whole team's here. Mark Albertson, Rob Hoaf. We'll be right back.

theCUBE Research Analysis

search

>> Welcome back, everyone, to theCUBE's live coverage here in Las Vegas. We're on the show floor at HPE Discover 2024. I'm John Furrier, host of theCUBE, with Dave Vellante, theCUBE Research and David Linthicum with theCUBE Research. Also, luminary speaker here at HPE Discover '24 on the big stage for Tech Talk right next to theCUBE. David, great to see you. It was nice to see you holding court on stage here at the Tech Talk arena next to us.

David Linthicum

>> Yeah, it's great to be here. It was the perfect infrastructure. I could look at you guys, who were doing reporting, and you could look back at me. So we had the whole corner.

>> You got my attention. I love those slides. Okay. Cloud migration failures. Really talking about the on-prem movement back... or, say, never left.

David Linthicum

>> Right. Well, there is some back-

>> Distributed computing paradigm. Give an overview.

David Linthicum

>> Yeah. Ultimately, as we've been covering in theCUBE, there's really a movement into heterogeneous kinds of architecture. So people aren't moving just to the cloud anymore, and certainly with AI, but they're moving to a mix of hardware platforms in the cloud. And there's reasons to use both. And I think what happened over the last 10 years is that utilization of the cloud was probably oversold within the enterprises. Who would've thunk that? Well, it was the fact that we were looking for scalability, cost efficiency, all the thing that cloud is able to do. And cloud is able to deliver on some of that, but where they weren't able to deliver on, cost-effectiveness. So what happened over the last 15 years that we've been migrating stuff into the cloud, we're not getting the value that's coming back into the enterprise from utilization of these cloud resources. So people are starting to understand that. I view the reckoning moment being 2022. Everybody moved to the cloud in 2020 with the pandemic. Suddenly these big, huge cloud bills came in once these things are operationalized. And I would get call after call after call that said, "What the heck's going on? This thing's a lot more expensive than I thought it was going to be." So when you look at why-

>> You can't just turn it off. You can't just turn a switch.

David Linthicum

>> You can't just turn it off.

>> Yeah.

David Linthicum

>> Yeah, you got all your applications in there. And yeah, here we go. But what happened was is that no one modernized the systems that were moved. They just lifted and shifted them. So it was very easy. And Cloud actually is very good at this, at providing good platform analogs for LAM stacks and even AI-based platforms. We're able to pick it up and push it over to the cloud and have it run in that environment. And the problem is that operational inefficiency is going to cost you money. In other words, it's like plugging a very inefficient electronic device into the grid. It's going to cost you money. And so organizations saw this. And this was kind of like a dirty little secret for a few years, where I watched all this repatriation going on. No one was talking about it because repatriation is kind of code for, "We screwed up." Oopsie daisy, hit the reset button. But it got enough momentum. We're putting some emphasis behind it, some intelligence, some methodologies, frameworks, how to make this stuff happen. So it's really right-sizing or normalizing of the application workloads.

>> And the cloud people called it cloud optimization.

>> Well, so we over rotated, you're saying, during the pandemic?

David Linthicum

>> Yes.

>> And then Andy Jesse says that cloud optimization has attenuated. Is that because people are spending more and not optimizing, or because they're moving back on prem?

David Linthicum

>> I think they're moving back on prem. And when he's looking why they're doing it, he would argue that you did not optimize your applications for the cloud, which means leveraging cloud-native features, things like that. That's tremendously expensive, to do that. You have to hire consulting companies and service firms and hire very expensive talent to modernize these applications, moving them to container-based systems, serverless-based systems, things like that. Most organizations didn't do that. Very few applications were modernized for the cloud. And so the way that the cloud providers fought back against it is they say, "Well, you guys haven't optimized in the cloud." So we heard the O word at the last re:Invent. And the reality is the O word is just them admitting to the fact that you're going to have to change a significant amount of your application in order for the thing to run on the cloud at a rate and an operational spend rate that would be reasonable.

>> Well, they'll also mean storage tiering, which, that's like a little sliver of cost reduction. Okay, not a huge deal, but what if you do optimize? What's your experience? If somebody moves to the cloud, they bring in your former firm or others, and they optimize. What is the TCO comparison? Because you talk to the on-prem guys. They say, "Well, we did this study with IDC and it shows that we're more cost-effective." And then you go to Amazon and they say, "We did this study with IDC and it shows that we're more cost-effective." And the customer's left to say, "Well, wait a minute." So what's the truth?

David Linthicum

>> It's the it depends answer everybody hates. So ultimately it's how the application was architected, how tightly it's coupled to the database, how it's bound to the processor. All these very geeky things that come into play. And so you can take this application that was built 20 years ago that didn't consider the optimization on the particular cloud provider. And so what we're doing is just basically putting it out on the platform in the cloud. We didn't have to optimize in those days. We had the entire server. It was at our disposal. So now when we put it on the cloud, they're charging us for all of that under-optimization within the system. So optimization means rebuilding the application, in many instances from the ground up. So it's going to leverage much fewer resources in a more optimized weight. And actually, if you do that right, you can move an application into the cloud and get it to burn 1/3 of the costs. And so the opportunity's there, but in many instances it was $1 million dollars an application and the enterprise is like, "Yeah-"

>> So big upfront cost. You can eventually get there. Then you don't have as much headroom and flexibility with that. The leftover capacity is really what I'm hearing here.

David Linthicum

>> That's absolutely right. Absolutely.

>> Maybe virtualization could help, Dave. No, only kidding. Ajay Patel, when he was at VMware, said on theCUBE. I'll never forget this, Dave. He said, "Oh yeah, I'm a middleware guy. The cloud is just hardware." And I'm like, "Actually, you know what? You're actually not wrong." The benefit of AWS was storage and servers, EC2, and queuing. Those are the building blocks. If you wanted to lift and shift, you can get hosting on demand, and variable pricing. Great for data processing. Okay, check. But what we're seeing here at HPE, David, is that you can cut hardware. You can just put another rack in for AI so, for the folks that want to play around, that seems to be a good pitch.

David Linthicum

>> It is a good pitch.

>> Yeah, because they know data center. There's a power envelope issue. They're addressing it. That's a sellable value prop.

David Linthicum

>> What T-shirt size do you want?

>> It's like, "Okay, I know how to deal with racks. Put a rack in, and that's my AI rack." Now the question is, will that be scoped for the workload? So is it the right hardware? So we're in this hardware scoping stage of generative AI. How do you see that playing out? And how does that factor into the cloud, which is essentially hybrid distributed computing?

David Linthicum

>> The cloud is always going to be a more convenient place to run AI because the ecosystem comes along with it. In other words, I get to vector databases, I get to odd middleware layers and streaming systems, things like that all on my portal. I'm able to assemble it together. It's just going to be more expensive to run. So that's going to be the trade-off. So what we're seeing here at HPE Discovery is that we're in essence bundling this stuff into an AI framework to compete with the cloud. You look at what they announced today in the keynote I was in. HPE Private AI, Private Cloud AI, is basically trying to replicate the convenience of the cloud, but doing so on premise within a data center-based system. So we have the advantage there. Number one, we own the hardware. Some people like to do that. But the price of hardware has just come down like crazy in the last 10 years, and so that blew a huge hole into cloud computing. So they're offering alternatives and not a DIY kind of situation where people have to assemble it themselves. I think that's where people get nervous. And if they're offered bundled solutions that are racks, that looks very cool to them.

>> Yeah. And Dave, what's interesting about Nvidia is that, and we talked about this earlier, they could have an Amazon-like situation here where they can get the benefit of HPE's enterprise because their NIMs are proprietary. That's the dirty little secret for the model stack. So they say, "Hey, you want to build your own RAG and some inference? Use NIMs, our Nvidia Inference Microservice, which is essentially proprietary." It's theirs, so if they get them locked in on Nvidia-

David Linthicum

>> But it's yours. But it's your-

>> All the enterprise get to have their own models. So the question is, how does that fit in? Because now you have, it feels like a network operating system, in a way. It's like they're giving a NOS to the enterprise. And they'll probably use it. My point is I think that might be the biggest land grab opportunity for Nvidia to jump in on the model stack and saying, "Hey, we'll own the model stack."

David Linthicum

>> Well, because people-

>> That's what I'm seeing.

David Linthicum

>> Because people are concerned about open source licensing. And some of them, read the fine print, Meta says, "Look, we reserve the right to pull the rug out from under you."

>> Well, no. No, they got the data to worry about. So if I'm an enterprise, I might say, "Hey, you know what? I'll go with Nvidia and HPE because I want to protect my data. Why do I want to put my model into anything else that I can't control?"

>> Well, exactly.

>> So okay, NIMS, Nvidia, spine, leaf, horsepower. They closed.

David Linthicum

>> iPhone?

>> Exactly.

David Linthicum

>> Oracle?

David Linthicum

>> Nvidia's going to make more money on selling hardware with partners like HPE and other folks there as well than they will in the cloud. And I think you absolutely hit the nail on the head. If we're going to leverage the core Nvidia features, we're going to be locked in. There's that word, into that particular platform. And if we're building these net new generative AI systems, then that's going to be an HPE movement for the next 10 to 15 years.

>> Yeah.

David Linthicum

>> And so we're building these things for the long term. They're going to take a tremendous amount of hardware resources in doing that. I think Nvidia sees the land grab as well, and that's why they're working with partners like HPE.

>> Well, and also AMD and Intel aren't sitting on their hands right now. Nvidia's got a great run right now. Their market cap's at all time high. Why not just nail down everything you can right now? And one of them is the enterprise AI market.

David Linthicum

>> I would be doing the same thing. That's why we're seeing the Nvidia CEO at every conference out there. And he's in front of everybody now because he knows he has to grab the land before eventually some of the innovation catches up and his competitors catch up. And he's not going to have the monolithic share that he has now.

>> I want to come back to repatriation, if I can, for a moment because we know anecdotally, when you talk to customers, you got the phone calls. And we talk to people. They go, "Yeah, we're going to move some workloads back." Yet when you look at the numbers, help me square this circle. So Dell last quarter, let's use Dell and HPE as a proxy. We'll throw in Supermicro for kicks, but they're not really an enterprise.

>> Yeah. They're definitely doing well right now.

>> But they're selling to hyperscalers and big MSPs. But taking Dell and HPE. Dell last quarter grew 6%. They're promising 3% to 4% growth. They're a $95 billion company trading at 1X, they're a $95 billion value. HPE grew at 4%. They're promising only 1% to 3%. I think that's conservative. They're a $30 billion company trading at 1X revenue. And then you got AWS growing at... this is just hardware, by the way. Stripped out, tried to anyway, all the 365 and other stuff that Microsoft throws in. AWS grew at 17%. Azure, almost 30%. GCP, 28%. So help me understand, you got the cloud guy's growing much, much faster.

David Linthicum

>> Right.

>> If there wasn't repatriation, would Dell and HPE be growing at 0%?

David Linthicum

>> No, I think they would be growing at a more rapid rate because of the utilization of AI. You wrote in your analyst brief on Friday that really the cloud is leading the way, because people see that as the path of most convenience for building these AI systems.

>> Yeah.

David Linthicum

>> But I think that the on-premise stuff is going to catch up fairly quickly. And they're going to see some huge growth as well, which is also going to be driven by AI. So everybody gets the AI tide that rises all ships, brings everybody forward. The cloud folks have a lot of momentum. Huge marketing momentum right now. They have the eyes and ears of the CIOs out there. So they're going to hold that position for as long as they can.

>> I think you're right, by the way, which suggests... of course Dell has run up a lot in this past year. HPE has not run up as much. If you actually look at the stock prices of these companies over the past year, Dell is up 173% as of last Friday, Supermicro's up 250%. But forget those guys for a second. HPE's only up 30%, but they're up 22%, in the last 30 days.

David Linthicum

>> Right.

>> So HPE is catching up. And so my point is, I think these growth rates that Dell and HPE are promising, basically 1% to 4%, is really conservative.

David Linthicum

>> yea-

>> And I think they're going to probably beat these expectations.

David Linthicum

>> Because there's a massive demand for more hardware right now.

>> And they have visibility, and big backlogs.

David Linthicum

>> The on-prem growth is happening. The AI bug is all about, if I'm going to have a neural network, I need different hardware. You're not going to get more out of an x86. You need the systems around a ethernet. And you heard liquid coolant's going to be in the home of the system architecture. So I think the AI wave is interesting, David.

David Linthicum

>> NVLink, CUDA, NIMS, NEMO. Forget it.

>> Well, David, I want to ask you because you got the new episode out, the AI Insights and Innovations with David Linthicum-

>> Which is killing it, by the way. Congratulations on-

>> Great job on your podcast on SiliconANGLE. Of course you've got your own. You also wrote for InfoWorld. The Whiplash article was hot. Just posted. But as you look to chronicalize the AI innovations and insights, there is a new architecture model emerging. What's your strategy with the podcast, the show? What are you going to be covering? Actually, you wrote the Insider's Guide to Cloud Computing, which is a nice backdrop, by the way, in your camera angle. It's in mind in my podcast, too, by the way. So check that book out, Insider's Guide to Cloud Computing. But as AI starts, we start documenting the progress. Where are we on that progress bar?

David Linthicum

>> We're not very far. We're at the caveman stage in terms of where this stuff is moving into. We're just figuring out... by the way, AI's been around since the '50s. We're still trying to figure out how to optimize it and make it work. And here we're in 2024. So in terms of generative AI, which is really the restart of AI, we're at the stage where we're trying to figure out what platforms to run them on, the use cases we need to use, how much this is going to cost us. Is this is going to burn up the planet? All this stuff that really haven't thought of. Everybody gets real excited about generative AI, hasn't looked at the hard planning it needs to occur. So the use cases are the biggest thing right now. In other words, businesses need to figure out how to use this stuff, and for what purpose. And it's funny. Even the vendor shows that we go to and things like that, they haven't understand the use cases that are there. They don't figure out how to sell this stuff. They know it's hot, but that needs to be the hardest part. That's going to be the hardest part to solve. The other thing is understanding we're moving to a heterogeneous platform. We're moving to AI on our iPhones with Apple Intelligence and all these other things that are coming forward. What does that mean for the enterprises out there? How are you going to leverage this effectively? And so it's looking at the innovations where things are coming to, but not just looking at them and reporting what the innovations are, but looking at the CIO and looking at the CTO and then these enterprises out there and telling them how it's going to affect them and how they have to change behaviors to make themselves a winner.

>> I'm glad you brought up use cases because this is a strong show. The Sphere, the keynotes, the Nvidia partnership, the OpsRamp stuff, the ecosystem growth. All that, very high marks. The one missing piece that I see is HPE is not talking... They're a $30 billion company. They're not talking about how their dog-fooding AI inside the company and what use cases they have. How they're prioritizing, where they're potentially seeing ROI. They haven't talked about that at all, the dog-fooding or drinking your own champagne. I think that's one area where they could really help customers. What are your thoughts on that?

David Linthicum

>> I think they could. I think ultimately I would've loved to see... in the keynote I sat in today I heard the guy from Mercedes who was talking about a storage solution. Okay, I get it. That's cool. But I'm talking about a killer AI application that runs on their stuff that they can point to that they're even doing internally.

>> Yeah.

David Linthicum

>> In other words, we built an engineering system, we built a deployment system, we built a BCDR system using this AI platform. I don't see those use cases from them. You do see it from Amazon, you do see it from Google, and you do see it from Microsoft and the other AI players that are out there. But I would love them, and I love the point. Eat their own dog foods is exactly right. So in other words, how are you using it? How are you successful with it? Therefore, how are you recommending that your customers be successful?

David Linthicum

>> Yeah, we brought the AI to the data-

>> Well, I think also, maybe a bad choice of words, but spoon-feeding their customer base what AI is with Nvidia. That's a big thing to grok. One of the things that I think they're hiding the ball on is the hottest companies right now are Databricks. That came out of the failure of Hadoop, and Spark made that happen. Remember, they bought MapR was the Hadoop for the enterprise, which was different than... so they have a DNA in this data that became Ezmeral. They're not telling that story.

David Linthicum

>> Well, but they're integrating it. They're not really telling the story. You're right. But they are integrating it.

>> Well, it comes down to use cases.

David Linthicum

>> It's their data fabric. And it does come down to use cases. You're right.

>> If you're going to do a use case and show compute, you got to show some meat on the bone. That's where the data piece comes in. To me, they don't have that yet. And I'm not sure if that's just, they didn't want to confuse the market, or they don't have it.

David Linthicum

>> They did represent a data lake that's attached to their private cloud system.

>> Ezmeral.

David Linthicum

>> Yeah, absolutely. And I think that's going to be fine. Sometimes that can be problematic because everybody doesn't want to use that particular-

>> Bring your own data lake. Right.

David Linthicum

>> People want to bring their own data. They want to bring their own licenses. They want you to run the core AI systems. And so if you're going to show me a framework, show something that's going to be open and extensible, and we're able to plug this stuff in. And I think that's what's missing from a lot of the vendors. Even if you go to AWS, and I guarantee you re:Invent's going to be like this, they're going to show you a stack that's only their stuff, nothing else, again, and don't show you how to make all your various systems work and play well in the environment. Now, they are open here and they're showing you how the integration occurs, but that's where the war's going to get fought. The complexity, the integration of it, the ability to have this information coexist.. We're not going to start moving stuff to proprietary databases just because you tell us that's attached to your AI system.

>> Yeah. Remember back in the old days of proprietary NOSs? I'll bring it again. The big buzzword was interoperability. Heterogeneous was happening and those networks were happening. That's happening now. I totally agree with you on that. I think that's the key. What would you advise HPE if they asked you? We're the board. We're Antonio's management team. David, advise me, what should we do?

David Linthicum

>> I would tell them to figure out the complexity angle and also figure out how to prepackage AI in such a way that it's going to be a bundle that people are going to consume. I saw a pretty good start today. They're going to have to build on that. Lots of missing pieces, lots on how you deal with inference engines, all these sorts of things. What are the 10 based architectures that I would base your system around? And how do I build those architectures in your particular environment? And I would publish those and I would push those out, and I would have, next year at this show, have one, two, three, four, five, all the way to 10, where people can walk around and figure out if these architectures are going to be right for their tech. And the ability to bundle it and come with a prepackaged solution has a huge amount of value because people are moving to cloud because it's more convenient to build AI in cloud because they have pre-configured ecosystems in the cloud. Make that happen within HPE. Equipment. Go after the architectures. Go after the 10 best use cases. That's what I would be telling HPE.

>> You're right.

>> Final question. I know we got to wrap because our next guests are here, but since you're here, you had the great talk. We were watching you at the Tech Talk. What was the reaction after the event? Because this always happens after you're done talking. Big crowd. I saw a big crowd there. So I'm assuming people came up after and said, "You did a great job," or, "You did a bad job." What did they say? What was the feedback? Not good job, bad, but would they say, "You're on point?" "You missed something?" What was the tone? What was the sentiment?

David Linthicum

>> I'm so glad you said that because no one has said that. And the ability to look at repatriation and the value of cloud and the ability to normalize the value of cloud within the modern enterprises. And there were CIOs over there, CTOs over there, high level individuals who make those core decisions. And quite frankly, they're confused.

>> Yeah. So you spoke reality to them. To them it was, "Thank you for giving a dose of reality?"

David Linthicum

>> Right. I spoke truth to confusion.

>> Truth to power who was confused, so confused power. Truth to confused power.

David Linthicum

>> That's exactly right.

>> Hey, that's what we do. We try to make people simplify the message. Of course we're bringing you theCUBE data here. CUBE Research is on the ground in force. SiliconANGLE team is here as well. theCUBE and SiliconANGLE team coverage again, getting the most of data in real time for you. We'll be back with more. I'm John Furrier. Dave vellante and David Linthicum and Rebecca Knight, Rob Strecce, Bob Laliberte. The whole team's here. Mark Albertson, Rob Hoaf. We'll be right back.