AI Proving Ground Podcast: Exploring Artificial Intelligence & Enterprise AI with World Wide Technology
AI deployment and adoption is complex — this podcast makes it actionable. Join top experts, IT leaders and innovators as we explore AI’s toughest challenges, uncover real-world case studies, and reveal practical insights that drive AI ROI. From strategy to execution, we break down what works (and what doesn’t) in enterprise AI. New episodes every week.
AI Proving Ground Podcast: Exploring Artificial Intelligence & Enterprise AI with World Wide Technology
HPE’s AI Factory Is the New Baseline
HPE's Craig Dillman and WWT's Earl Dodd explain why AI Factories rooted in decades of supercomputing expertise are becoming essential infrastructure for enterprises seeking to operationalize AI at scale, ensure sovereignty and brace for next year's data bottleneck.
The AI Proving Ground Podcast leverages the deep AI technical and business expertise from within World Wide Technology's one-of-a-kind AI Proving Ground, which provides unrivaled access to the world's leading AI technologies. This unique lab environment accelerates your ability to learn about, test, train and implement AI solutions.
Learn more about WWT's AI Proving Ground.
The AI Proving Ground is a composable lab environment that features the latest high-performance infrastructure and reference architectures from the world's leading AI companies, such as NVIDIA, Cisco, Dell, F5, AMD, Intel and others.
Developed within our Advanced Technology Center (ATC), this one-of-a-kind lab environment empowers IT teams to evaluate and test AI infrastructure, software and solutions for efficacy, scalability and flexibility — all under one roof. The AI Proving Ground provides visibility into data flows across the entire development pipeline, enabling more informed decision-making while safeguarding production environments.
From Worldwide Technology, this is the AI Proving Ground Podcast. Across the enterprise landscape, leaders are confronting a familiar tension. Everyone is talking about AI factories, but very few can tell you what one actually is. The term has become a kind of industry Raw Shack test, a buzzword loaded with promise, urgency, and confusion. At its core, the idea is deceptively simple. Take raw data in, move it through a series of production lines, models, workflows, and automation, and produce something of value on the other side. Insights, decisions, new applications, even entire agentic systems. But as today's guests will tell you, the gulf between that concept and executing it inside of a real organization is wide. We'll be talking with Craig Dillman from HPE, who comes from a world where factories weren't metaphors, they were supercomputers. And Earl Dodd from WWT, who has spent years helping organizations modernize, operationalize, and when necessary, rebuild the very foundations beneath their data and AI ambitions. These are two experts who have lived the AI factory story longer than most of the industry has had a name for it. Because for enterprise leaders, the question isn't whether you need an AI factory, it's whether you're ready to build one that will make sense a year from now. So let's jump in. Craig, welcome to the AI Proving Ground Podcast. How are you today? I'm great. Thanks for having me. Excellent. And Earl, it's always good to see you. How are you?
SPEAKER_01:Great, glad to be back.
SPEAKER_04:Awesome. Awesome. We're talking about AI Factory, which is a growing buzzword right now in the industry. Uh everybody seems to be latching onto it, but I don't really know if everybody has clarity and definition around what it is. Craig, from HPE's perspective, what is an AI factory and what should it be doing for organizations?
SPEAKER_03:Yeah. I'd I'll think about this a a couple of there's kind of two ways I think about it. Yeah. And AI factory, let's call it the lowercase F. We're gonna think about it conceptually. Okay. Uh and then and then I'm sure we'll get into a conversation about what's HPE's AI factory, and we'll we'll capitalize the F for proper noun brand marketing purposes there. But the the conceptual idea, I I think it is a good analogy, right? It's we're trying to to draw the distinction or help our customers see that there's a there's a way to build and deploy and operationalize specialized uh infrastructure, and that's software as well, to take raw inputs. And if our we think about our raw inputs are are really data here, move it through the production floor, and that's really AI models. And and then there's gonna be an output, there's gonna be a product that you produce. And that's actually probably a broad category of things that you produce. If you wanted to try to encapsulate it, you might just say actionable insights. And there's a lot, you know, it's a trained model, or there's actual decisions, there's automation, there's agents, there's a new application. Any of those things might be what you produce, but you could all kind of grab them all under the idea of the heading of some kind of actionable insights. So raw data in, push it through the factory, yeah, through those production lines and output insights.
SPEAKER_04:Yeah. And Earl, sticking with the lowercase F on in this instance, it's easy to look at a factory and say, oh, there it is. It's producing whatever the end uh goal might be. But what are the piece parts, what's making up the factory as we as at least as we have it to be.
SPEAKER_01:Yeah, yeah, so Craig, you I felt a little bit of PTSD coming on of my macroeconomics class, you know, thinking about factors or you know, of production and and everything else. So so so you're right, and we have the same concept and we put that into our wheel, you know, Studio Foundry Factory and everything else. But and and you mentioned, you know, the the components of software. What what we need is uh on a factory, on the factory floor, so to speak, there are the bits and pieces, acceleration as a service, et cetera. But there don't don't forget about the people in the factory, the talent. And we find that is one of the big skills gap right now out there, and a competitive advantage with our partners, especially like with HPE, to have the people with that partnership to help drive and build the factory. That's right. And and so it's it's more than just the componentry, it's more than just how you put these people, it's the process and the people together. So I we if you're familiar with the something called an iron triangle, access quality cost, what we're doing is is working on all legs of that iron triangle. And that's what the factory is is is about giving you that access, you know, that the quality, the people, the capabilities at a price performance per watt cost.
SPEAKER_04:Yeah.
SPEAKER_01:That's the iron triangle.
SPEAKER_04:Okay. Well, let's put the capital F on it here, and that stands for I don't want to get into explicit rating on the uh on the podcast this week. Um Craig, let's talk about the differentiation here. A lot of companies have their own flavors of AI Factory. What is HPE's vision for AI Factory and what where do you think you differentiate in the market?
SPEAKER_03:Yeah, there's uh there's a couple of things to think about here, too. First and foremost, because we're here at SC, I think it's probably interesting to take a step back. And in fact, one of my colleagues and I were just talking about this this morning. That if you if you kind of peel back and think about what HPC, which is traditionally, you know, not really an necessarily an AI workload, but has been around for decades, building a workload-focused, specialized set of infrastructure and people and processes uh for this for this one workload. What the AI factory looks a whole lot like is an HPC cluster. Well, it's a bar.
SPEAKER_01:So telling people that all the time.
SPEAKER_03:That's right. So we've been doing we've been doing the AI factory, or we've been building a factory that is very similar to what we now instantiate as an AI factory actually for decades. Yeah. And and so that's that's uh maybe a good backdrop. Now, what is HPE's vision for it? And uh what Earl mentioned earlier, kind of around that triangle, uh, is an important piece, especially as we we help customers operationalize it. So what what I mean thinking about you know AI as a relatively new uh workload. Maybe the building this large specialized cluster is not a new thing, but the workload is. That that is the new thing. That's setting we all know it's setting the world on fire. But moving from experimentation or even just proof of value or starting out looks a lot different uh in production. And uh and so being able to take this blueprint for an AI factory and make it repeatable, make it uh predictable, both on performance and governance, how do you take this workload and make it operationalized? And so maybe a little double entendre, we think about the factory metaphor here that you've got a production line. Well, it's producing something, but in the IT world, we know that we're talking about how do you deploy something in production. That's different than test dev.
SPEAKER_02:Right.
SPEAKER_03:And so our HPE AI factory is really about how do we help our customers move AI into production at scale.
SPEAKER_04:Yeah. Well, Earl, I mean, I I I like that we're talking like Craig's mentioning this is just a big HP C cluster. It's a supercomputer. And before we start rolling here, you're talking about how that's right in HPE's wheelhouse. So that's from your perspective. Tell me about why this is you know the kind of the right time, right place where HPE and the AI factory are coming together and really making a big difference.
SPEAKER_01:Absolutely. Just you know, for the audience, SC stands for the Supercomputing Conference, it's the largest conference in in the world. It happens to be here. Your supercomputing hometown. Uh, you know, it it's I've I've been doing it for almost 30 years now for for SC. I've been I started on my first Cray, you know, which is now part of HPE in in 1978, right? And so what goes around comes around, it's it's back to the future here. And and we didn't call it AI back then, we called it analytics, or then there was this big data and everything else. But you know, it only takes you what 20, 30 years to be an overnight success. Right. No, so so bringing it uh it all back is most people don't realize the value of HPC and supercomputing as an underpin, a a literally a foundation piece of technology and computing and capability to drive new and novel capabilities. And one of those capabilities is this thing called AI. Now, what what's a little bit different, and and and this is where uh we we have this convergence coming on between AIHPC. Same coin, different sides. Okay. There's still an the integration, the co-integration that we do with HPE and in building it, whether it's in a an a cray supercomputing doing uh CFD computational fluid dynamics work versus AI and generative AI work, it's still the same foundation. Now, where it started is uh a little bit different. The the HPC guys have been doing scientific and technical computing, the AI guys has been doing this, you know, things around the computer vision and modeling and simulation, and started out this, you know, on the other side. Now they've come together, and and and the most one of the things I think is the most important part here as we work with HPE is how do we co-integrate these systems because it's not a one size fits all, and there's just not one application. There is multi-tenancy, which is a big thing that I look at for HPE, is bringing that all together. So software and the processes to build that to give me the accordion effect of more over here in classic 64-bit and more over here now in this AI and generative AI and and the and the and the explosion that's coming with inference. I mean, if fund foundational models are are good and we're still selling and doing a lot of integration there, but the but now I got to use and consume, I gotta produce.
SPEAKER_04:I got that factor on it, I mean Craig, is that all part of the oper or oper operationalizing go-to-market here? I mean, that that sounds like that's really how we're helping organizations operationalize the concept of an AI factory. Is it is it bringing it to the data? Like how do you help that?
SPEAKER_03:Yeah, well, let me I I'll try to circle back to that, but let me I want to respond to something that uh Earl brought up. And this is what I I personally and I know our HPE teams will run into a little bit, especially in kind of this new world of of AI. It's a new workload. And so there's a there's a reasonable response that we have from enterprise customers, not not from research universities, not from you know, Department of Energy, not the supercomputer, but the ones that have been building large-scale HPC clusters, they they understand everything that Earl was just talking about. But a lot of our enterprise customers are trying to kind of find their way of where do I start with this? And so it's easy to get a little bit skittish when you hear a vendor come in or a partner come in and start talking about supercomputers, these kind of row scale data center scale clusters. And I'm an enterprise customer, and I go, slow your role, Craig. That is way that's way too big. And so part of HPE's AI factory portfolio solution approach actually serves both of those customers. So we have an AI factory. We were kind of building this before AI factory became a buzzword. Yeah. It's called private cloud AI. Yeah. And this is really built for those enterprise customers that are trying to plus you're adding that business model to that green lake.
SPEAKER_01:That's right. That's very important to dip your toe into the water, so to speak, to build up into the eventually the bigger shirt shirt sizes.
SPEAKER_03:That's right. So we've got private cloud AI, which is an AI factory. It it fits those definitions that we've talked about, but it's really aimed for inference, rag, fine-tuning at a relatively smaller scale, but still something that is very useful to operationalize IT for our enterprise customers. And then there's the bigger brother, there's the AI factory at scale that HPE offers, and there's uh AI factory for sovereign, what we call for sovereign customers, uh, which introduces an uh an additional layer of security, of multi-tenancy, hard multi-tenancy. Uh so HPE really has three key pillars of our AI factory portfolio: private cloud AI for the enterprise, AI factory at scale, and sovereign AI factory. And WWT has been a a clear early adopter of this vision and this technology. In fact, we showcased it that uh private cloud AI, you guys bought the maybe the very first thing.
SPEAKER_01:We were the first, and it's it's we we have two, we have small and medium and grown to large at the advanced technology center. So it w we're we're growing.
SPEAKER_03:Yeah. Can I just say too, I'm here at Supercompute with a lot of these customers and and channel partners as well, and and vendors and you know, consulting everyone is here, all of the all of the ecosystem is here. And you can't go very far without someone mentioning or knowing the ATC. But that's right. It's just the ATC.
SPEAKER_01:They may not know WWT, they know the ATC. Absolutely.
SPEAKER_03:We'd love to hear that.
SPEAKER_01:Yeah, no, I mean it it it it is the biggest country club of technology on the on the planet. You know, over over a billion dollars worth of investment in there. Uh HP has been a very sound partner from the beginning, coming up. I'm not saying I've been trying to get the Cray in there, you know, the Ed GX, but uh for some reason they w, you know, I don't have a I don't have the GX in there. But we have all the PCAI stuff and the capability. Uh to tie this all together, the customer, the industry, and it doesn't matter which domain, you know, whether we're we're we're talking about governments or sovereign clouds or or or whoever, you know, the automotive company, aerospace, et cetera, et cetera. Everybody's at a certain place along the journey. So that's why we start where they're at and go through a capability maturity curve, do the inventory and everything else. And that's why we have Studio Foundry uh factory capability. That's why we have our wheel, so to speak, and we meet them where they're at. And then we kind of step back a little bit and talk about the use cases, what they want to produce, the five whys, everything else, you know, your kids. Why, why, why? Anyway, we we we do that and we bring them on a journey. The thing that's really, really important, and it was important in HPC, and it people forget about this in this AI. It's a journey. It's a journey. It's just that we don't drop off, that you procure it, drop off the kit, and voila, it magically is there, and ROI is coming out of it. Right. Return on investment and stuff. So we spend a lot of time, you know, in the beginning to make sure we know the metrics of success to drive this, de-risk it through our best partners, through the Advanced Technology Center to deliver it, to give you the best opportunity to drive ROI, TCO. We we have a new term when we talk to academia and institutions and things like that, return on the research, R-O-R. You know, when who's getting the next Nobel Prize? You know, that kind of thing. So we just people there there are different groups out there in the world that says, oh yeah, I can get you the kit, I can throw it into the rack, I can and install it. What do you do next? Right. There there is there's people involved.
SPEAKER_04:Yeah. A slight shift, but I I've seen Antonio Neary, you know, HPE CEO uh Antonio Neary talk a lot about the idea and need for control when it comes to to AI and AI factory. You mentioned sovereignty. Why is that control increasingly important and what type of value does it provide kind of on the back end?
SPEAKER_03:Yeah. I to me, I you you you start to have to go kind of back to the beginning of where is where is value being created. Yeah. Okay, so we talked about this AI factory and something's being produced, and increasingly AI is helping our our customers, and that could be uh, you know, an enterprise, that could be a small and medium business enterprise all the way up to a a state, to a national, you know, uh interest. And so if the the value that you're creating, those insights, those applications, those become intellectual property, those become strategic differentiators, or they become, in some cases, instrumental to national health or security. Sure. Clearly you can see why there's an interest on having as much control over that as possible. There's also regulation that comes with that. We've we've seen it, we're we're living it right now.
SPEAKER_01:Well, right, especially in Europe. In Europe right now. We're living the the amount of you know reduction in water, cap in power, decibels in in in in the data center itself.
SPEAKER_03:So there's there's absolutely a a control piece that is about how do I make sure that I've I've got my IP, my value creation stack, and others don't have access to it either accidentally or or deliberately. Yeah, through our or whatever. But then there's also a there's also a cost piece to this. Sure. And and the public cloud has been transformative to all of our all of our customers, right, over the last several years. What we see with specialized workloads, especially ones that are in high demand, like AI, it brings a cost element to it. And the volumes of data that are that are associated with training, fine tuning, and inference. Inference as well, especially when you get into agentic multi agent workloads.
SPEAKER_01:Tokens per tokenomics, tokens per cent. That's right.
SPEAKER_03:The data volumes are massive. And so it changes the economics of public cloud. So you kind of combine these two things of what's my tokenomics in a public environment and what are my data sovereignty or intellectual property concerns. We're seeing, and I'm sure you are too, a massive repatriation on some workloads. That's not to say others don't get it.
SPEAKER_01:Absolutely. You know, there was this uh mentality that, you know, it started out with, you know, my granddaddy said it had to be, you know, on premises, you know, here. And then my daddy said that it it had to be cloud first. What I'm saying right now, it's to my kids, you're analogy, sorry, I mean it's a terrible analogy, but but is what is appropriate, what is practical, what is cost effective. And and it's based on many, the business model is based on many things. The world is heterogeneous, the world is is hybrid. Now going back to the control, there's the there's not only the control of the intellectual property, uh everything, I agree. But there's the control plane. There is that plane that needs to come together and optimize across workloads that make up workflows to drive this. And and and so it and uh are we you know mature in some of this? We're not mature that. I I I I'm sure we're adolescent and and beyond, but we're we're starting to get into more of how did this work, especially with the sovereign clouds and requirements. Now, there is regulation, you know, it's coming to North America. You're gonna see these swings and what I tell everybody at the end of the day, math always wins. Math always wins. Remember, remember when Deep Seek hit, you know, a year ago or so, the market went, oh my gosh, you know, the sky's falling and everything else. And I go, just wait. The the problem and opportunity is still there. We're still, we still have the need to solve something. We got to produce something. They may have produced it better. When would uh you know, anytime I install a supercomputer or a machine or superpod, it doesn't it doesn't matter, right? A PCAI system, it it I can optimize it and more work comes. There's always a backlog. Right. And so we're gonna see more math, we're gonna see better models, we're gonna see truncations and and pruning of things of nature, and that is good for everybody, and that means I can start solving more and more difficult problems.
SPEAKER_00:This episode is supported by HPE Juniper Networking. HPE Juniper Networking delivers high-performance networking solutions to simplify IT operations. Build a secure and scalable network with HPE Juniper Networking's innovative technologies.
SPEAKER_04:As as we think about kind of feeding the AI factory, capital F or lowercase F, whichever one you want to choose, with data, recognizing that these AI workloads are coming from everywhere, increasingly at the edge, how does HPE kind of configure that into here? At least as I've read into it, that seems to also be a differentiator here for us.
SPEAKER_03:Now I'm regretting my my my case. Well, they'll add it out, don't worry. No, that's right. But the the portfolio that we would kind of brand under AI factory is clearly going to be data center focused, as as it should be. But that doesn't mean there's not significant capabilities or opportunities for this distributed edge AI. And so whether it's uh data ingest from multiple sources, having an AI factory wherever wherever it's landed, being able to collect from, manage, pre-provision, do all of the what we've learned over the big data history of what do I, how do I what are all these ETLs that I have to do? Consolidate and put governance rings around data, who can access it and when, to make it ready for the workload. That's just that's actually a very significant piece of an AI pipeline is ingest to training or fine training.
SPEAKER_01:Ingest to insight. Exactly. I'm gonna answer your cyber question. I wanna I wanna get that, but my number one net new growth area is the intelligent edge. Okay. And and so when we think of yes, there's a factory. Well you little F factory here, but not everything is built in that factory. There might be another factory to the left, the right. Remember, there's different tiers of the supply chain feeding that. So and everybody has a different definition of what the edge is, but there are different factories or tiers starting to feed into the big beast. Okay. And and so what I I tell people is we are just getting into this. Um, you know, there there's a circle here or a spring that goes from specialization to commoditization, democratization, and going around. We're we're kind of going around this. So now this AI factory people are took and putting it into the big systems, but you've got to look at all the other tiered providers of that supply chain coming in. Those are on the edge, maybe even far edge and farther edge out there. And so not everything has to be the big supercomputer like in my world, but we're seeing a lot of this. So getting back to orchestration and control, that's very important to know the feeder system in that. And from a cybersecurity perspective, that becomes paramount because those are new attack surfaces in there. The thing I love about HPE is that cybersecurity is not an afterthought.
SPEAKER_02:Yeah. Right.
SPEAKER_01:It is part of the zero trust model. It is in there in the beginning from the hardware architecture, from securing it in their entire supply chain. I know whether I have to go to a government customer or an enterprise customer that still requires FIFS 140-3, stuff like that. You've got to make sure that you think I I think of the yin and yang kind of thing, I think data is there, and I think cyber, cybersecurity is there. They're both together. That is the center or hub of my universe. And then I put the spokes of this accelerator or that storage or that piece.
SPEAKER_03:And so to that to that point, Earl, let's think about the HPE AI factory. And I know I we talked about that, you know, at the far right of the spectrum, we're selling to, we're working with sovereign states, or you know, in some cases it's it's governments, in some cases, it's federally funded or mandated universities. But if we are meeting the security requirements at a national sovereign level, you're absolutely right that we have to bake in from the beginning. It's not a bolt-on, it's a bake-in on how do we ensure that we meet uh those regulations and exceed them in some cases.
SPEAKER_01:Observ observation, you know, observability, repeatability, and and and and you know, uh, as my Marine Corps guys say, you know, agility is the new capability. And and and so, but that agility starts with having a a true uh distinguished uh foundation and sit there. And that foundation is based on trust, and trust is made on based on cyber.
SPEAKER_04:You're talking about how agility is kind of that superpower. Craig, how does how does HPE think about agility, resilience, knowing that things change so quickly all the time? Does does the concept of the AI factory just continuously evolve with those evolving standards, or how do you how do you think about it?
SPEAKER_03:Yeah, I'm glad you brought that up because we've kind of been talking about the factory a little bit without I don't know that we need to go into every you know piece of the layer cake of what that solution provides, but a key piece of what we're doing in our AI factory strategy is I think how Earl said it earlier, there's a the you said an accordion, right? You're talking about flexibility. So for us, what that means is the the capability to bring in, to integrate tools, whether it's on the data side, whether it's on the machine learning operation side, whether it's in the Kubernetes orchestration and automation side, or even in the HPE scheduling. Like tools that are highly utilized and leveraged. What we have today is a I would I've it's fair to call it a curated stack, but it's not locked.
SPEAKER_01:That's that's really important. It's still open architectures that allow you to meet the customer where they're at.
SPEAKER_03:That's right. And we don't know what tomorrow's tool is going to be. And so we're designing our AI factory solutions in such a way that when that next wave comes or when a tool chain or new model or new capability hits the market, and they're in they're happening faster and faster, right? We've got to be able to incorporate those in, but still maintain the governance observability, scalability parameters, right? We're about productionizing. We're not we're not in science project land. This is grown-up production grade utilities.
SPEAKER_01:You've got to do return on research, return on the investment. At the end of the day, these are not toys. They keep us safe, uh, they keep uh they find the cure for cancer, you know. They they I'm not sure if we can get the airline's uh scheduling systems working yet, but we can certainly try, right? For me, uh you know, I started, you know, like I say, back in the 70s, not not 1870s, 1970s, okay. And this is deja vu to me. We we kind of go through these cycles every 15-ish years and thing. So this is the new third cycle. What I think is important, especially for for HPE, is you you've acquired you know, SGI and and cray, you know, from my world and and a lot of other people, but you you have this entire zeitgeist, this knowledge base to draw from. Okay. And and and the people that succeed in the world is learn from history. Right. And and to do that. Do we all make mistakes? Of course. But we lead, we fail forward, move forward, fix, and and and continue to go on and build these tools and capability and have the people. Have the people. The number one threat to the United States if you go into the intelligence compute community is not these countries, it's the lack of talent, the lack of skills.
SPEAKER_03:That's interesting to bring up. You're right to keep coming back to the people side of it. It is interesting as we have been deploying both the enterprise AI factory and and also these at-scale AI factories. We've learned we've learned some things. Oh, yeah, absolutely. And I think one of those key learnings has been it's got uh something that provides true value to our customers has got to be more than just a reference architecture. And that is that's a that's a starting place, right? Certainly. But then where does that reference architecture enter into the real world? Just maybe a simple example of that is is one of our AI factory deployments. You know, on paper, we we draw out the schematic the schematics, we've got the rack elevations, we have calculated where the cobalt, where the cables go, how many we need, and then we move into deployment phases with the with the customer and their facilities team. And it turns out they don't have room for 16 racks in a row. We're gonna have to span across multiple racks. It also turns out this customer uh had a compartmentalization, kind of a security policy years and years ago in their data center that said each rack in my data center must be self-contained. In fact, they welded it so that it would be, and the cable, the the cable uh conduit passage between racks was about that. Yeah, well. Okay, for an AI factory. No, no, when I got that, yeah. Immediately he knows like that's not gonna work. Exactly. So in the real world, now we're including in part of the statement of work and part of what we're doing with our customer, includes a blowtorch. Yeah. Okay, that you don't find that in the reference architecture documentation that you may need to.
SPEAKER_01:Well, that that's that's again, that comes back to that agility or flexibility. Yeah. Uh so I just came from the ATC. Yeah. Yeah, we were doing a customer demo, uh very large organization to put in a very large uh HPC AI supercluster and everything else. And they ask, you know, what's the biggest risk? And the the the biggest risk, you know, there's there's many risks, but you know, going into these customers that have older data centers that have been doing it has a culture of this is the only way it can be done or is done here. We find it it's not necessarily a technology issue. It it's it's the smaller things, the blow, you know, bring your own blowchorch, you know, things like that. We we have problems with now testing ceilings to hold, you know, for ceiling loading. We in the old days it was floor loading, now it's ceiling loading, yeah, because these cable trays, you know, the bundles are are huge. And and and those are the learning curves, but the the key is what we learn and what they learn, HPE learns, we put it together. That's right. You know, yeah, it's a synergistic effect. I'm I'm positive I have not been uh to the data. I'm positive that I'm gonna go to a data center that's gonna have a new challenge. Okay. I don't know what it is. I'm I'm afraid that there's like an eight-inch water main above me or something like that. But we just have to be agile. We have to ask these questions up front. We have to have a reference architecture, it's good. Again, it's like my one one meth people. Our battle plans are great until I hit the beach. I have to have that agility.
SPEAKER_04:And I think we're or or is what you're getting at saying, you know, if you think about the AI factory, it's not just about the gizmos and gadgets inside that you're it's easy to think of.
SPEAKER_01:You're talking about the bricks, and I'm talking about a holistic thing. And and and this is the beauty that WWT brings, okay, and and and a big bad business partner here, like H T B HPE brings, is that we've done this before, and we can put our uh cut sheets at the NAIC or or whatever integration center we're at, and and put this all all together, and then we can problem solve. Not everybody has the skill sets that WWT has to do what DOE, I'm trying to get DOE, the Department of Energy, to rethink their word. They call it co-design, you know, which means they're extorting money from you, but to do co-integration. It is all about creating a balanced, capable system that has runs from day zero to day two. Yeah, yeah.
SPEAKER_03:Clear, that's great that you brought up day zero to day two. I let me I want to come back to that in a second. I want to pull back for a second too to think about the enterprise side of it. Like we, if you live in the HPC world again, oh I got that gets that sometimes gets a little scary, right? That I'm like, I'm not the DOE. And we're talking about some of the challenges throughout the whole stack, whether it's facilities or data center or technology or people. That sounds like a big, hairy problem I've got to solve. And again, what private cloud AI for our enterprise customers who maybe aren't staffed or don't have the history of building high performance compute, GPU-enabled clusters and all that comes with that, private cloud AI is a way to say, let's make that friction part easier for you. And we're gonna we're gonna build it in our factory, and that's not just racket, stack it, power it, cable it, it's also install it so that when it arrives on day one, you're ready to start.
SPEAKER_01:And you can do the the manage piece of that through like model. So you you we have to look at things through the entire life cycle, the entire value chain. And and the challenge that I see with some vendors and and maybe some of our competition is they're good at one thing, but the value chain is long. Yeah. And it's you've got to have this hysteresis loop. Sorry, that was a technical, you've got to have this self-learning, learning and reinforcing loop in there to learn. Not not just, hey, I'm building a model to do an LLM or an SLM. To do it, how do we integrate it? How do we operate it? How do we secure it? You know, how how do I just get all the different stakeholders in the same room? Oh, that's what WWT provides. It's just getting everybody in the same room to get to a common vocabulary.
SPEAKER_04:Yeah. We're coming up on the bottom of the episode here. So short on time.
SPEAKER_01:Oh, I got I got 20, I got 20 hours left. 20 hours left.
SPEAKER_04:Yeah, there you go. But what we're talking about here is building an AI factory that's resilient to whatever the future holds. So we'll we'll end as we typically do, kind of on a what do we expect coming forward? It's a good time of year because you know it's where we're approaching the end of 2025. We're about to head into 2026. Craig, what are we? going to see as it relates to the future of the idea of the AI factory. What is it going to have to account for in the future? It could be, you know, agentic, more of a shift towards inference. It could could even be something like quantum or or who knows what it might be. What's on the horizon that we need to be paying attention to next year?
SPEAKER_03:Yeah, I think you hit two two really key ones and agentic and inference that go to me hand in hand.
SPEAKER_02:Yeah.
SPEAKER_03:What we've no maybe a year ago or so and and beyond beyond that. LLM was everyone was starting to learn how do I think about how many billion parameters do I need to be able to handle? And we're thinking almost in monolithic what's the biggest baddest model and will it solve everything? And I think this year in in 2025, what we've seen is a rapid innovation cycle to say maybe that's not the most efficient way. And this is always the way, right? If I can subdivide the work and make it more task specific, it becomes more high I can utilize more. It's more efficient. And so this is where the agentic AI has landed this. And and Earl mentioned he said SLM, LLMs and SLM small language models in a a task specific agentic workflow that's operating together has become kind of the the next flywheel I think for how do enterprises, how do our customers actually deploy this and make it more and more usable. I think there's a lot of talk you see like I think it's the New York Times and Wall Street Journal both had some articles. If you're on LinkedIn at all you're seeing it probably every week you'll see these AI is not providing any ROI. What's the promise? But the customers aren't actually seeing it and and to me being able to get to these more task specific, interactive and integrated workflows are the way forward to start see some real customer value that is you know narrowed in on your industry or your specific account. Now that's kind of the background because what happens is even though it's a smaller language model the data needs explode. Like the data that is produced the compute resources that are required for inference are going to far outstrip what we've seen already that have been massive for model training. And so having the capabilities to scale quickly in a secure governable repeatable you know IT ops driven kind of flow I think is going to be crucial for our customers to catch and ride that wave.
SPEAKER_04:Yeah.
SPEAKER_01:Yeah Earl any closing thoughts based on I'm in total agreement. I'm going to add a little bit what I think 2026 is going to be the year of the data bottleneck by the way there'll be an article coming out on on that what I I've seen in the through this whole process the last couple years is by accelerator. By the way I want to use the word accelerator not GPU okay why because there are other kinds of accelerators GPU is a very good accelerator we still sell a ton of FPGAs. We're seeing quantum in there there is a new class of inference engine engines that are being being built and those are going out to these small language models at the edge out into the smaller factories the AI factories out there. But the thing that runs the AI that runs the world is data. Data's the new oil data's the new gold etc so we're going to see in 2026 the move fr definitely to inference an agente but the world of being truly data centric at this point. Well just real quick I know that was the last question but what happens downstream when those when that data bottleneck leaks or or whatever it might be like what are the impacts downstream there's significant impacts the the what most people don't realize for every single byte of data that that is created or whatever there is 10 to a thousand metabytes of metadata. Data itself is not necessarily the intellectual property it's the meta information associated with that data. And that's where the protections are starting to come. Yes it's a little bit different with PII and PHI and things like that. But you know they would use that information to find out that I have a genomic characteristic that I might be susceptible to this or or or that that's the meta information associated. So we've with HPE and others you know in our portfolio is working to make sure that the the conveyor belt this factory of production is secure the new attack vector is the parameterization of your workload. If I mess that up I can make I can do nefarious things.
SPEAKER_04:Yeah absolutely well I mentioned we're short on time here so we'll we'll wrap on that but uh to the two of you thank you so much for taking the time out of your busy schedule this week I know a lot going on at SC25 so uh hopefully you're enjoying that show and we're learning a lot that we could take into 2026 and really drive some some meaningful customer outcomes and uh Craig thank you again to you and HPE for the great partnership that we have here today. Oh yeah yeah thank you okay today we heard something subtle but important an AI factory is a living system people processes infrastructure and governance all working together to turn data into durable defensible value. The AI factory you build today has to be resilient to whatever comes next agentic workloads, data bottlenecks, new accelerators, new regulations. It's not a one-time project it's an operating model. This episode of the AI Proven Ground podcast was co-produced by Nas Baker, Kara Kuhn and Diane Devry. Our audio and video engineer is John Knoblock. My name is Brian Phelps thanks for listening and we'll see you next time, it's
Podcasts we love
Check out these other fine podcasts recommended by us, not an algorithm.
WWT Research & Insights
World Wide Technology
WWT Partner Spotlight
World Wide Technology
WWT Experts
World Wide Technology