Advancing AI 2026 | Derek Dicker, AMD

Clips
More from Advancing AI 2026

Derek Dicker

CVP, Enterprise Business Group

AMD

play_circle_outline Balancing TCO, Token Costs, and Optimization: AMD, Silicon Diversity, Openness, and Hybrid Edge-to-Cloud AI Strategies

play_circle_outline Agentic AI shifting compute toward CPU plus GPU workloads

play_circle_outline Vertical market opportunities: healthcare, telco, retail, manufacturing, FSI

play_circle_outline Deployment velocity: bottoms-up adoption versus top-down mandates

play_circle_outline Sovereignty concerns driving demand for open, on-prem AI options

Info

Derek Dicker, AMD

Derek Dicker

CVP, Enterprise Business Group AMD

search

(INTRO)

John Furrier

>> Welcome back to theCUBE's live stream here in San Francisco for AMD's Advancing AI event. It's a global event. It's open, free for practitioners, packed house, sold out. I'm John Furrier, hosting with Dave Vellante, my co-host, breaking down all the action as we've been looking at all the advances of AI being injected in companies up and down the stack from the semiconductors to the application layer, agents and physical AI, all happening, the concrete's being poured, the roads are being paved for a new generation. Derek Dicker here, theCUBE alumni, corporate vice president, Enterprise Business Group with AMD. Derek, great to see you again. Thanks for coming back on theCUBE.

Derek Dicker

>> Thank you for having me.

John Furrier

>> It's a pleasure to see you guys again.

Derek Dicker

>> Hasn't been that long.

John Furrier

>> Yeah, right. So I wrote a post on LinkedIn, Dave wrote a post as well. We're kind of in a historical view right now, and one of the big takeaways I was preparing for the event is looking at all the different waves of the computer industry.

Derek Dicker

>> Yes.

John Furrier

>> the PC wasn't a mainframe, cloud wasn't a data center, mobile wasn't a desktop. Everyone had that kind of thing, but we're in a whole other world in the enterprise. You're seeing a wake up call, tokenomics, instant time to value, the value equations changing in the enterprise. What's your take? How do you see the historical relevance of this shift?

Derek Dicker

>> I think you caught it really well. We're in a very, very interesting period of time for the industry. the demand that we see being placed and coming from the enterprise right now is at a monumental level, and actually more broadly across the entire industry. There are a few areas that we see fairly regularly in discussions with customers. One key trend is definitely an interest in silicon choice, silicon diversity, and openness. So a desire to see an ecosystem get built out. The second is a lot of the data centers that currently exist right now, they're all being tasked with taking in AI, but they're bursting at the seams. And so modernization requirements become a real issue. And then lastly, you mentioned it in your intro, the whole total cost of tokens and people figuring out how they're going to build optimized solutions is an open question.

John Furrier

>> The role of compute, storage, networking, data, databases are all changing. And what's interesting is if you look at the modern era, and I want to get your reaction to this, it's not hard for anyone to have a cloud now. Because hybrid computing has kind of won. You see the role of cloud native, Kubernetes. that's fueling massive data center build outs, net new. Pre-existing data centers are being modernized to connect to the cloud, and you got the edge emerging with intelligence being injected in. How do you guys at AMD view that? Because one, you have a couple nice things happening. The bet on hybrid and cloud worked. The modernization of the on -premise customer facility, the edge where the action is for autonomous and physical, all kinds of boiling up.

Derek Dicker

>> Absolutely true.

John Furrier

>> How do you guys view this from a company standpoint?

Derek Dicker

>> Well, the company really specializes in building silicon products that optimize around some of the toughest problems the world sees. That's how Lisa describes the mission of the company, addressing the world's toughest challenges. And one of the starting points is, if you're sitting inside a data center today, how do you end up creating room for artificial intelligence? and that comes with building great TCO focused solutions. That can come from how you architect a device, a silicon device, to how you work with folks to build a server to make sure that you're building the most power efficient solution, as one example. And that's sitting just in the core of the data center infrastructure. The other side of things, if you just look, is enterprises are rolling out artificial intelligence. Exactly as you said, there's a need for edge and edge computing, building GPU centric solutions that sit inside edge servers is one where you can see our MI350P solution there. And then lastly, the very notebooks that we all use today, being able to run models locally inside an actual device with an AI PC is something that we continue to see people deploying and building out, gaining momentum now. Because developers are now using their PC to run Ollama, all these new tools, unbelievable. Well, I'd say, and by the way, the other piece is just the deployment of the GenAI and what's that doing to architectures inside the data center.

Dave Vellante

>> I want to get your take on how wrong the conventional wisdom has been in the last three or four years, and how enterprises are dealing with that. just writing down here, the assumptions that the market made is going to be one dominant player, token costs are irrelevant, RAG-based chatbots are going to deliver a ton of value, US frontier models, they're essential, they're going to dominate everything, cloud, same, they're going to dominate deployment, on and on and on. All of that has been upended.

Derek Dicker

>> That's right.

Dave Vellante

>> It's totally the opposite now. You're seeing Chinese open models, you're seeing other open models, new startups coming out. How are enterprises dealing with that chaos and what role does AMD play in helping them navigate that?

Derek Dicker

>> Well, first off, our objective is to be a trusted advisor to the enterprises with whom we engage, but we can't do it alone. We have to do it through an ecosystem. So a lot of the work that we do is with ISVs as part of a broader ecosystem, channel partners, global system integrators. There's a whole community. It takes a village to go pull this off. That's the first thing. The second thing is understanding what the use cases are of these enterprises. Everybody's on a very different journey. Some folks are far down the path. They have a hybrid deployment that actually takes into consideration how do they run local open models internally. And some of them are all fully frontier model and they're looking at the bill and they're going, oh my goodness, this is probably not going to hold on for the long term. We're going to have to find a way to adjust. So, the way that we approach this is very much through a consultative partnership with a customer base and with partners, and then addressing the enterprise markets. The big thing we find that they're looking for is solutions, and stitching together solutions across a multitude of those parties.

John Furrier

>> You guys have such an installed base from the historical perspective of AMD, got the puzzle pieces, Xilinx, Pensando, AMD IP, ZT Systems. I'm on Wall Street now with the NYSE studio with theCUBE, and I hear a lot of Wall Street, they're kind of short, medium term thinking, not long game like the Valley. They think it's a GPU story, and my theory is it's an infrastructure story. We are in an AI infrastructure build out, so AI really is about infrastructure. If you look at all the things that are happening, now the application layer could be anything.

Derek Dicker

>> That's right.

John Furrier

>> Coding, agents, work, physical AI, we see that clearly here. As an infrastructure supplier of silicon and software, How are you guys packaging that up to the enterprise? Because like Dave said, they got a real wake up call, which is their budget cycles, how they procure, and even how they do TCO is changing.

Derek Dicker

>> That's right.

John Furrier

>> They got to bring that to the front lines on a quarterly basis, not a one to five year payback and all kinds of analysis to justify what an IT spend was. Now they're in value creation mode, not a support role. What's your take on, what's your angle on that?

Derek Dicker

>> Well, first off, there's a massive infrastructure build out that's occurring right now, exactly as you described. We'd say it's of epic proportions, being the CPU guy pun intended.

John Furrier

>> Pun intended, pun intended. Well done.

Derek Dicker

>> A big part of it, interestingly for me, and it goes a little bit back, Dave, to your comment earlier, the world's changed a ton in the last year. If you looked at GPU being the center of a lot of these workloads, what's happened over time as Agentic has unfolded is a realization that it's actually very much as much a CPU workload as it is a GPU workload. And so we're now in the process of working with a ton of the ecosystem players to go build out fairly dense GPU racks where the value is on core count, for instance, the number of threads that are being run at a level deeper than maybe we want to go to, but a lot of the things that AMD's been building in its roadmap for the CPU side of the house is super well -tuned to where this is heading. And the only way that we deliver these products? Partnering with the OEMs. You'll see a ton of announcements, a ton of rack infrastructure here that's designed to exactly go after that segment of the market.

John Furrier

>> I love the rack scale because Dave and I were saying, back in the day, servers, PCs were categories. The rack is basically a server with a bunch of servers in it.

Derek Dicker

>> Absolutely true.

John Furrier

>> The complexity to make that happen is pretty high. So how would you answer this if someone's like, well, the rack and stack days of IT are over, because they kind of are. Top of rack switch, oh, we need more compute, put another box in there, all good, old way. The new way's a lot different.

Derek Dicker

>> Yeah.

John Furrier

>> AMD is not doing that. You guys are clearly here saying, we're not your rack and stack set of servers. It's more of a system play. How do you address that? How would you communicate that you're not just a set of servers and you're the system? What's the key value points there?

Derek Dicker

>> I think if you just look at Helios as an example, and I know you'll have several others on the show over the course of the next day or two, what the company decided to do is instead of building systems themselves and competing with our customers, we took the approach of open and building a rack scale architecture, publishing a spec, engaging the ecosystem, bringing them all together so that the industry can actually realize the best of breed on all the different technologies that are required. But in order to do that, you got to go build an actual full system and infrastructure. infrastructure, and if you have the opportunity, if your viewers have the opportunity to come see it, to check out what Helios represents. It's a coming together of the entire industry to go develop a solution, it's a perfect example of exactly what you described. We're not operating just the same way as before, here's a reference design, go build it.

Dave Vellante

>> So Derek, you would agree then, you have to build that whole system. You don't necessarily have to sell it as a whole system, you have to build it, and I want to follow up on John's question because the critics would say, oh, it's just a server farm, they'll use that term. What makes it not a server farm, and what is it? What makes it a system, and what are the salient aspects of that system?

Derek Dicker

>> If you look at the system, you're essentially integrating 72 GPUs into a high performance rack, which means that you need high performance networking, you're going to need compute to go orchestrate across all of it, you have to integrate all of the GPUs into a system, be able to deliver it in a way that's serviceable also. And all of those different pieces, it really, if you want to do it as a single company building it, absolutely possible to go do so, but then you don't get the advantage of all of the different best of breed technologies that can be built to deliver an entire solution.

John Furrier

>> We're a decade and a half into Open Compute Project, which was kind of founded with Facebook and others. We were there at the beginning, John. I remember when the 19 - What a journey it's been. The rack spec, Cole Crawford actually did the rack spec on that. If you fast forward to today, that open standard has helped a lot. But also now with AI, vertical markets are now wide open. You're seeing the intelligence bringing on-prem back. So you're seeing healthcare, I'm going to put a monster server rack scale maybe in a hospital if it's big enough. Or it might be a small telco, CO or tower, whatever the use case is, you're not always going to have the same form factor footprint. Absolutely. This has become a huge design concern. Talk about the vertical market opportunities that you guys see for AI.

Derek Dicker

>> Well, it's really interesting. There is not a vertical that exists today that isn't trying to figure out how they apply AI to their infrastructure. It could be an FSI, it could be telco, healthcare, retail, manufacturing, pick any one of them. The challenge is to figure out what are the unique use cases that they have, and then partner with both the software vendors that play in that space, the hardware vendors that play in that space, and then put together solutions. At the end of the day, no enterprise has really come to me and said, hey, I want to buy a CPU. They're looking for a business outcome. And so a big focus of ours is identifying what are those business challenges that they're facing and how do we build it with the best of breed technology we have here or with partners in the industry. I think you can see that in spades.

John Furrier

>> Talk about the velocity of turnaround time on solutions. Let's take a telco, FSI, they're always fast, they have big budgets. Healthcare, obviously you're seeing a lot of action with AI in healthcare. Healthcare and telco I would put as two nice verticals because they represent what I call slow moving verticals from an adoption standpoint. They all want rapid deployment. How are some of the turnaround times on the velocity of identifying a reference use case, turning that around? Have you seen an uptick in the velocity or is it the cadence of the silicon? How would you describe it?

Derek Dicker

>> What I would say is that what's really, truly interesting is we're watching the industry use more and more advanced technologies that are rooted in AI to help go build solutions for AI. And as a result, some of the development times that are required are compressing. I wouldn't put my finger on any one particular vertical that I could see and say, hey this one's moving extremely quickly. But I would say in each one of them, there's a pretty amazing set of use cases that are being developed. Whether it's, as you suggested, in the healthcare space, or even in retail, looking at how people are combining video analytics into an existing store and delivering insights to ownership.

John Furrier

>> The top question that comes up in the past six months has been the SaaS-pocalypse and legacy vendors, in some cases partners of AMD, getting caught in the noise of, oh my God, stock's down, they missed a quarter, maybe they don't have AI right, or AI can replicate that. As you look at the AI success formula, the enterprise spigot doesn't seem to be fully open. The floodgates haven't opened, at least from our data. They're encoding, they're seeing the costs. So we're seeing a mid -tide, early shift. We think 2027 will be a pretty monster year for enterprise. What are some of the issues that you're seeing in the enterprise right now? Is it change management? Is it personnel? Is it budgets being sucked in from the tokenomics? Is it infrastructure expenditures? Why aren't the businesses moving faster?

Derek Dicker

>> I think that the unfortunate reality in enterprises is they tend not to move as quick as something like a hyperscale infrastructure where they have full control over their entire stack. In the enterprise space, what I find is different enterprises are in different stages of awareness of what they want to do. So in some cases they're caught up in trying to figure out what the solution is. In other cases, it's actually building out a plan to deploy solutions in their environment. I agree with you that the current momentum, if you were to just look at it at a point in time, you'd say, hey, this doesn't feel like it's as far along as you might expect to see given some of the size and scale of the numbers that are being put forth. But if I use our own infrastructure as an example at AMD, and how quickly we've actually seen token consumption increase, how quickly we've come up with solutions to go deploy to run on-prem versions of models on open source so that you can delineate between the two like John was describing. It's happening rapidly.

John Furrier

>> Well, it's almost reversed.The user adoption is far outpacing the speed of deploying solutions.

Dave Vellante

>> Yeah, well, it started top down. The CEO said thou shalt do AI, and then everybody started saying, This is great. I could download this stuff and now it's bottom-up, kind of like the PC was in the productivity boom. I want to come back to something you said, Derek. You said agentic is basically a CPU and a GPU story. This is something you and I talked about last month. And John, you mentioned the SaaSpocalypse. People talk about the deterministic world and the stochastic world coming together and that's why not all SaaS is dead, although some SaaS is going to get disrupted. Bad software always dies. But you, because you now have perhaps the leading market share in CPU. let's say half the market.

Derek Dicker

>> The trajectory looks good and it's in that range. It's big.

Dave Vellante

>> And I'm curious as to how, we talked about this a little bit last month and how you see that evolving. Will the traditional sort of general purpose functions get absorbed into the sort of AI world, bringing together that deterministic and non -deterministic world together. I see that CPU heritage as an advantage to bridge. If it doesn't happen, that's fine. If they're, you remember the Gartner hybrid IT, the old stuff and the new stuff, the bimodal, right?

Derek Dicker

>> Bimodal. Right, right.

Dave Vellante

>> So it may be bimodal, okay, if it is, that's fine.

Derek Dicker

>> Multimodal.

Dave Vellante

>> But I think it's more likely to sort of come together. and you can be a bridge there. How do you think about that? What are customers asking you to do or partners asking you to do in terms of affecting that bridge? Is that a conversation that's going on?

Derek Dicker

>> We fortunately, because of the technologies that the company has available to it, as well as the recognition that's come from the CPU acceleration over time and share gain, we've become a trusted advisor to a number of different enterprise customers. What we see happening over time today is a lot of the agentic workflows are a mix of existing applications. It's web searches, it's database. There's a lot of pre -processing that gets done before something goes and hits a model, gets processed, gets kicked out, and then it needs to be processed again. And if you look at the choices in architecture and where the optimization points are, it's a combination of both, but it's pretty heavy CPU, versus what we might have seen in the chatbot era, which was maybe largely you'd have a CPU whose role was just orchestrating workflow into the use of a given model on a GPU.

Dave Vellante

>> So, shifting dramatically. So I guess the question is, is it like stovepiped, bimodal, here's CPU, here's GPU,

Derek Dicker

>> it's not. No, not at all.What you're saying is it's a new computing paradigm. Absolutely true. So you use that word. It's a new computing paradigm, but it's stitching together a lot of what existed before, which is a very interesting combination, as they keep going.

Dave Vellante

>> I live in an old house, and we do a lot of stitching together. you don't just tear down the house, even though sometimes I'd like to, but I can't. And you have the good rooms that you hang out in.

John Furrier

>> That's exactly right. And the old rooms that look great when people come over. That's kind of like IT today. Derek, I want to get your thoughts as we wrap up because one of the big discussions, it's not always mainstream, but it's pretty much top of mind, is reskilling and new skills. If you look at the business schools and the engineering programs, even the comp sci programs, you got a lot of young talent coming out and the world that they were trained on four years ago has changed. If you were to write a case study and say, hey, this is what you need to think about, what would you say to folks who were coming into the workforce, because it's very much system architecture, systems engineering, software, hardware, you got to know how things work. People are walking into an AI world, and you're seeing the results, the AI native companies are booming. They have no problem working with models and systems. The NeoClouds, they're standing up literally hyperscale level performance. It's pretty incredible how far they've come in the short term. If you have the right mindset, how would you describe that case study as a lesson to be ready to compete and contribute to this new AI infrastructure?

Derek Dicker

>> It's an awesome question. It's actually timely for me at a very personal level. So my oldest son is in school and he's studying engineering at Texas A&M. And the dialogue that we have is wrapped around what are the core tenets that you need that likely won't change too much that will help enable you to be effective in the workforce. The first one is understanding how to solve problems and at a very fine-grained level of detail. The second one is being able to work with people. The third one is being able to communicate. But wrapped on top of all of that is understanding how to be agile and to use the tools that are available today in the AI space. You cannot not be focusing on where this world is going. And it can't just be a check in and play with tools for a little while and phone it in later. This is changing so quickly, you have to stay on top of it. I think it's not clear exactly where all of this eventually trends to, but if you have those core tenets available to individuals, I think that's what's going to be required. Just to highlight some of the tooling advances, because I think this is one of the key points.

John Furrier

>> Tooling has advanced so much in the past 24 months.

Derek Dicker

>> Yes.

John Furrier

>> What areas jump out at you that you say, wow, that's game-changing? Even physical AI, we had that discussion earlier. this is huge. These tools are very effective. They are teaching context, not just tools for putting stuff in a database for search. these are like building work into technology.

Derek Dicker

>> I'll tell you the piece of technology that I've played with the most at a personal level is Ollama, and then also looking at Hermes as an alternative play. And I've been astounded that once you connect it and begin the process of training it and asking it to do things, some of the insights that come back are completely unexpected. unexpected, the ability to actually have something go out, help manage calendar infrastructure, help book appointments, these are very much consumer-centric models and uses, but as soon as I saw that several months ago and projected it into what the working world looks like inside an enterprise today, it's like this is going to be quite a transformation.

John Furrier

>> AI scales superpowers, scales intellect.

Derek Dicker

>> Without a doubt.

John Furrier

>> And that's the superpower.

Dave Vellante

>> And everybody's stressing about the return, et cetera, but we talked about the bottom-up, like the PC era, the personal productivity. What's happening, I see, I wonder if you see this as well, is the projects are very focused. They have a clear outcome, they're getting done, and they're adding value in their little way. But eventually those are going to come together across an enterprise software stack, and it just takes time. If you're a multi-billion dollar company, you can't just rip that out and do it overnight. If you're a small company, you can, but otherwise you have to come at it from the edge.

Derek Dicker

>> Right.

Dave Vellante

>> And that's really hard. That's why it takes the better part of a decade to play out. But the impacts, I feel like, are going to be enormous.

John Furrier

>> Well, we haven't even talked about sovereign. The sovereignty angle now with AI is a revenue opportunity in countries and geographies. So, there's so much going on.

Dave Vellante

>> Well, by the way, a key piece of sovereign is it has to be open. Because otherwise, if you're locked into a particular technology stack and you don't have an open source alternative, then you're really not purely sovereign. Now, you don't necessarily have to go all the way to this extreme, but it's a consideration for customers.

Derek Dicker

>> Absolutely true, and part of the reason why we're getting pulled into all of these discussions in that space is exactly for that reason. People are looking for an open solution, they're looking for multiple options to go host on-prem in their own infrastructure and on their own soil.

John Furrier

>> Derek, thanks for coming back on theCUBE and sharing your insights. Congratulations on all the momentum at AMD. Love the new Instinct portfolio you guys have at the system level. It's all coming together, looking forward to catching up. Again, we could have gone an hour, but we don't have an hour. Thanks for coming on, good to see you again.

Derek Dicker

>> Thank you guys for having me again.

John Furrier

>> It's a pleasure as always to see you. Corporate Vice President of Enterprise Business Group at AMD, I'm John Furrier with Dave Vellante here in San Francisco live. Thanks for watching.

Derek Dicker, AMD

search

(INTRO)

John Furrier

>> Welcome back to theCUBE's live stream here in San Francisco for AMD's Advancing AI event. It's a global event. It's open, free for practitioners, packed house, sold out. I'm John Furrier, hosting with Dave Vellante, my co-host, breaking down all the action as we've been looking at all the advances of AI being injected in companies up and down the stack from the semiconductors to the application layer, agents and physical AI, all happening, the concrete's being poured, the roads are being paved for a new generation. Derek Dicker here, theCUBE alumni, corporate vice president, Enterprise Business Group with AMD. Derek, great to see you again. Thanks for coming back on theCUBE.

Derek Dicker

>> Thank you for having me.

John Furrier

>> It's a pleasure to see you guys again.

Derek Dicker

>> Hasn't been that long.

John Furrier

>> Yeah, right. So I wrote a post on LinkedIn, Dave wrote a post as well. We're kind of in a historical view right now, and one of the big takeaways I was preparing for the event is looking at all the different waves of the computer industry.

Derek Dicker

>> Yes.

John Furrier

>> the PC wasn't a mainframe, cloud wasn't a data center, mobile wasn't a desktop. Everyone had that kind of thing, but we're in a whole other world in the enterprise. You're seeing a wake up call, tokenomics, instant time to value, the value equations changing in the enterprise. What's your take? How do you see the historical relevance of this shift?

Derek Dicker

>> I think you caught it really well. We're in a very, very interesting period of time for the industry. the demand that we see being placed and coming from the enterprise right now is at a monumental level, and actually more broadly across the entire industry. There are a few areas that we see fairly regularly in discussions with customers. One key trend is definitely an interest in silicon choice, silicon diversity, and openness. So a desire to see an ecosystem get built out. The second is a lot of the data centers that currently exist right now, they're all being tasked with taking in AI, but they're bursting at the seams. And so modernization requirements become a real issue. And then lastly, you mentioned it in your intro, the whole total cost of tokens and people figuring out how they're going to build optimized solutions is an open question.

John Furrier

>> The role of compute, storage, networking, data, databases are all changing. And what's interesting is if you look at the modern era, and I want to get your reaction to this, it's not hard for anyone to have a cloud now. Because hybrid computing has kind of won. You see the role of cloud native, Kubernetes. that's fueling massive data center build outs, net new. Pre-existing data centers are being modernized to connect to the cloud, and you got the edge emerging with intelligence being injected in. How do you guys at AMD view that? Because one, you have a couple nice things happening. The bet on hybrid and cloud worked. The modernization of the on -premise customer facility, the edge where the action is for autonomous and physical, all kinds of boiling up.

Derek Dicker

>> Absolutely true.

John Furrier

>> How do you guys view this from a company standpoint?

Derek Dicker

>> Well, the company really specializes in building silicon products that optimize around some of the toughest problems the world sees. That's how Lisa describes the mission of the company, addressing the world's toughest challenges. And one of the starting points is, if you're sitting inside a data center today, how do you end up creating room for artificial intelligence? and that comes with building great TCO focused solutions. That can come from how you architect a device, a silicon device, to how you work with folks to build a server to make sure that you're building the most power efficient solution, as one example. And that's sitting just in the core of the data center infrastructure. The other side of things, if you just look, is enterprises are rolling out artificial intelligence. Exactly as you said, there's a need for edge and edge computing, building GPU centric solutions that sit inside edge servers is one where you can see our MI350P solution there. And then lastly, the very notebooks that we all use today, being able to run models locally inside an actual device with an AI PC is something that we continue to see people deploying and building out, gaining momentum now. Because developers are now using their PC to run Ollama, all these new tools, unbelievable. Well, I'd say, and by the way, the other piece is just the deployment of the GenAI and what's that doing to architectures inside the data center.

Dave Vellante

>> I want to get your take on how wrong the conventional wisdom has been in the last three or four years, and how enterprises are dealing with that. just writing down here, the assumptions that the market made is going to be one dominant player, token costs are irrelevant, RAG-based chatbots are going to deliver a ton of value, US frontier models, they're essential, they're going to dominate everything, cloud, same, they're going to dominate deployment, on and on and on. All of that has been upended.

Derek Dicker

>> That's right.

Dave Vellante

>> It's totally the opposite now. You're seeing Chinese open models, you're seeing other open models, new startups coming out. How are enterprises dealing with that chaos and what role does AMD play in helping them navigate that?

Derek Dicker

>> Well, first off, our objective is to be a trusted advisor to the enterprises with whom we engage, but we can't do it alone. We have to do it through an ecosystem. So a lot of the work that we do is with ISVs as part of a broader ecosystem, channel partners, global system integrators. There's a whole community. It takes a village to go pull this off. That's the first thing. The second thing is understanding what the use cases are of these enterprises. Everybody's on a very different journey. Some folks are far down the path. They have a hybrid deployment that actually takes into consideration how do they run local open models internally. And some of them are all fully frontier model and they're looking at the bill and they're going, oh my goodness, this is probably not going to hold on for the long term. We're going to have to find a way to adjust. So, the way that we approach this is very much through a consultative partnership with a customer base and with partners, and then addressing the enterprise markets. The big thing we find that they're looking for is solutions, and stitching together solutions across a multitude of those parties.

John Furrier

>> You guys have such an installed base from the historical perspective of AMD, got the puzzle pieces, Xilinx, Pensando, AMD IP, ZT Systems. I'm on Wall Street now with the NYSE studio with theCUBE, and I hear a lot of Wall Street, they're kind of short, medium term thinking, not long game like the Valley. They think it's a GPU story, and my theory is it's an infrastructure story. We are in an AI infrastructure build out, so AI really is about infrastructure. If you look at all the things that are happening, now the application layer could be anything.

Derek Dicker

>> That's right.

John Furrier

>> Coding, agents, work, physical AI, we see that clearly here. As an infrastructure supplier of silicon and software, How are you guys packaging that up to the enterprise? Because like Dave said, they got a real wake up call, which is their budget cycles, how they procure, and even how they do TCO is changing.

Derek Dicker

>> That's right.

John Furrier

>> They got to bring that to the front lines on a quarterly basis, not a one to five year payback and all kinds of analysis to justify what an IT spend was. Now they're in value creation mode, not a support role. What's your take on, what's your angle on that?

Derek Dicker

>> Well, first off, there's a massive infrastructure build out that's occurring right now, exactly as you described. We'd say it's of epic proportions, being the CPU guy pun intended.

John Furrier

>> Pun intended, pun intended. Well done.

Derek Dicker

>> A big part of it, interestingly for me, and it goes a little bit back, Dave, to your comment earlier, the world's changed a ton in the last year. If you looked at GPU being the center of a lot of these workloads, what's happened over time as Agentic has unfolded is a realization that it's actually very much as much a CPU workload as it is a GPU workload. And so we're now in the process of working with a ton of the ecosystem players to go build out fairly dense GPU racks where the value is on core count, for instance, the number of threads that are being run at a level deeper than maybe we want to go to, but a lot of the things that AMD's been building in its roadmap for the CPU side of the house is super well -tuned to where this is heading. And the only way that we deliver these products? Partnering with the OEMs. You'll see a ton of announcements, a ton of rack infrastructure here that's designed to exactly go after that segment of the market.

John Furrier

>> I love the rack scale because Dave and I were saying, back in the day, servers, PCs were categories. The rack is basically a server with a bunch of servers in it.

Derek Dicker

>> Absolutely true.

John Furrier

>> The complexity to make that happen is pretty high. So how would you answer this if someone's like, well, the rack and stack days of IT are over, because they kind of are. Top of rack switch, oh, we need more compute, put another box in there, all good, old way. The new way's a lot different.

Derek Dicker

>> Yeah.

John Furrier

>> AMD is not doing that. You guys are clearly here saying, we're not your rack and stack set of servers. It's more of a system play. How do you address that? How would you communicate that you're not just a set of servers and you're the system? What's the key value points there?

Derek Dicker

>> I think if you just look at Helios as an example, and I know you'll have several others on the show over the course of the next day or two, what the company decided to do is instead of building systems themselves and competing with our customers, we took the approach of open and building a rack scale architecture, publishing a spec, engaging the ecosystem, bringing them all together so that the industry can actually realize the best of breed on all the different technologies that are required. But in order to do that, you got to go build an actual full system and infrastructure. infrastructure, and if you have the opportunity, if your viewers have the opportunity to come see it, to check out what Helios represents. It's a coming together of the entire industry to go develop a solution, it's a perfect example of exactly what you described. We're not operating just the same way as before, here's a reference design, go build it.

Dave Vellante

>> So Derek, you would agree then, you have to build that whole system. You don't necessarily have to sell it as a whole system, you have to build it, and I want to follow up on John's question because the critics would say, oh, it's just a server farm, they'll use that term. What makes it not a server farm, and what is it? What makes it a system, and what are the salient aspects of that system?

Derek Dicker

>> If you look at the system, you're essentially integrating 72 GPUs into a high performance rack, which means that you need high performance networking, you're going to need compute to go orchestrate across all of it, you have to integrate all of the GPUs into a system, be able to deliver it in a way that's serviceable also. And all of those different pieces, it really, if you want to do it as a single company building it, absolutely possible to go do so, but then you don't get the advantage of all of the different best of breed technologies that can be built to deliver an entire solution.

John Furrier

>> We're a decade and a half into Open Compute Project, which was kind of founded with Facebook and others. We were there at the beginning, John. I remember when the 19 - What a journey it's been. The rack spec, Cole Crawford actually did the rack spec on that. If you fast forward to today, that open standard has helped a lot. But also now with AI, vertical markets are now wide open. You're seeing the intelligence bringing on-prem back. So you're seeing healthcare, I'm going to put a monster server rack scale maybe in a hospital if it's big enough. Or it might be a small telco, CO or tower, whatever the use case is, you're not always going to have the same form factor footprint. Absolutely. This has become a huge design concern. Talk about the vertical market opportunities that you guys see for AI.

Derek Dicker

>> Well, it's really interesting. There is not a vertical that exists today that isn't trying to figure out how they apply AI to their infrastructure. It could be an FSI, it could be telco, healthcare, retail, manufacturing, pick any one of them. The challenge is to figure out what are the unique use cases that they have, and then partner with both the software vendors that play in that space, the hardware vendors that play in that space, and then put together solutions. At the end of the day, no enterprise has really come to me and said, hey, I want to buy a CPU. They're looking for a business outcome. And so a big focus of ours is identifying what are those business challenges that they're facing and how do we build it with the best of breed technology we have here or with partners in the industry. I think you can see that in spades.

John Furrier

>> Talk about the velocity of turnaround time on solutions. Let's take a telco, FSI, they're always fast, they have big budgets. Healthcare, obviously you're seeing a lot of action with AI in healthcare. Healthcare and telco I would put as two nice verticals because they represent what I call slow moving verticals from an adoption standpoint. They all want rapid deployment. How are some of the turnaround times on the velocity of identifying a reference use case, turning that around? Have you seen an uptick in the velocity or is it the cadence of the silicon? How would you describe it?

Derek Dicker

>> What I would say is that what's really, truly interesting is we're watching the industry use more and more advanced technologies that are rooted in AI to help go build solutions for AI. And as a result, some of the development times that are required are compressing. I wouldn't put my finger on any one particular vertical that I could see and say, hey this one's moving extremely quickly. But I would say in each one of them, there's a pretty amazing set of use cases that are being developed. Whether it's, as you suggested, in the healthcare space, or even in retail, looking at how people are combining video analytics into an existing store and delivering insights to ownership.

John Furrier

>> The top question that comes up in the past six months has been the SaaS-pocalypse and legacy vendors, in some cases partners of AMD, getting caught in the noise of, oh my God, stock's down, they missed a quarter, maybe they don't have AI right, or AI can replicate that. As you look at the AI success formula, the enterprise spigot doesn't seem to be fully open. The floodgates haven't opened, at least from our data. They're encoding, they're seeing the costs. So we're seeing a mid -tide, early shift. We think 2027 will be a pretty monster year for enterprise. What are some of the issues that you're seeing in the enterprise right now? Is it change management? Is it personnel? Is it budgets being sucked in from the tokenomics? Is it infrastructure expenditures? Why aren't the businesses moving faster?

Derek Dicker

>> I think that the unfortunate reality in enterprises is they tend not to move as quick as something like a hyperscale infrastructure where they have full control over their entire stack. In the enterprise space, what I find is different enterprises are in different stages of awareness of what they want to do. So in some cases they're caught up in trying to figure out what the solution is. In other cases, it's actually building out a plan to deploy solutions in their environment. I agree with you that the current momentum, if you were to just look at it at a point in time, you'd say, hey, this doesn't feel like it's as far along as you might expect to see given some of the size and scale of the numbers that are being put forth. But if I use our own infrastructure as an example at AMD, and how quickly we've actually seen token consumption increase, how quickly we've come up with solutions to go deploy to run on-prem versions of models on open source so that you can delineate between the two like John was describing. It's happening rapidly.

John Furrier

>> Well, it's almost reversed.The user adoption is far outpacing the speed of deploying solutions.

Dave Vellante

>> Yeah, well, it started top down. The CEO said thou shalt do AI, and then everybody started saying, This is great. I could download this stuff and now it's bottom-up, kind of like the PC was in the productivity boom. I want to come back to something you said, Derek. You said agentic is basically a CPU and a GPU story. This is something you and I talked about last month. And John, you mentioned the SaaSpocalypse. People talk about the deterministic world and the stochastic world coming together and that's why not all SaaS is dead, although some SaaS is going to get disrupted. Bad software always dies. But you, because you now have perhaps the leading market share in CPU. let's say half the market.

Derek Dicker

>> The trajectory looks good and it's in that range. It's big.

Dave Vellante

>> And I'm curious as to how, we talked about this a little bit last month and how you see that evolving. Will the traditional sort of general purpose functions get absorbed into the sort of AI world, bringing together that deterministic and non -deterministic world together. I see that CPU heritage as an advantage to bridge. If it doesn't happen, that's fine. If they're, you remember the Gartner hybrid IT, the old stuff and the new stuff, the bimodal, right?

Derek Dicker

>> Bimodal. Right, right.

Dave Vellante

>> So it may be bimodal, okay, if it is, that's fine.

Derek Dicker

>> Multimodal.

Dave Vellante

>> But I think it's more likely to sort of come together. and you can be a bridge there. How do you think about that? What are customers asking you to do or partners asking you to do in terms of affecting that bridge? Is that a conversation that's going on?

Derek Dicker

>> We fortunately, because of the technologies that the company has available to it, as well as the recognition that's come from the CPU acceleration over time and share gain, we've become a trusted advisor to a number of different enterprise customers. What we see happening over time today is a lot of the agentic workflows are a mix of existing applications. It's web searches, it's database. There's a lot of pre -processing that gets done before something goes and hits a model, gets processed, gets kicked out, and then it needs to be processed again. And if you look at the choices in architecture and where the optimization points are, it's a combination of both, but it's pretty heavy CPU, versus what we might have seen in the chatbot era, which was maybe largely you'd have a CPU whose role was just orchestrating workflow into the use of a given model on a GPU.

Dave Vellante

>> So, shifting dramatically. So I guess the question is, is it like stovepiped, bimodal, here's CPU, here's GPU,

Derek Dicker

>> it's not. No, not at all.What you're saying is it's a new computing paradigm. Absolutely true. So you use that word. It's a new computing paradigm, but it's stitching together a lot of what existed before, which is a very interesting combination, as they keep going.

Dave Vellante

>> I live in an old house, and we do a lot of stitching together. you don't just tear down the house, even though sometimes I'd like to, but I can't. And you have the good rooms that you hang out in.

John Furrier

>> That's exactly right. And the old rooms that look great when people come over. That's kind of like IT today. Derek, I want to get your thoughts as we wrap up because one of the big discussions, it's not always mainstream, but it's pretty much top of mind, is reskilling and new skills. If you look at the business schools and the engineering programs, even the comp sci programs, you got a lot of young talent coming out and the world that they were trained on four years ago has changed. If you were to write a case study and say, hey, this is what you need to think about, what would you say to folks who were coming into the workforce, because it's very much system architecture, systems engineering, software, hardware, you got to know how things work. People are walking into an AI world, and you're seeing the results, the AI native companies are booming. They have no problem working with models and systems. The NeoClouds, they're standing up literally hyperscale level performance. It's pretty incredible how far they've come in the short term. If you have the right mindset, how would you describe that case study as a lesson to be ready to compete and contribute to this new AI infrastructure?

Derek Dicker

>> It's an awesome question. It's actually timely for me at a very personal level. So my oldest son is in school and he's studying engineering at Texas A&M. And the dialogue that we have is wrapped around what are the core tenets that you need that likely won't change too much that will help enable you to be effective in the workforce. The first one is understanding how to solve problems and at a very fine-grained level of detail. The second one is being able to work with people. The third one is being able to communicate. But wrapped on top of all of that is understanding how to be agile and to use the tools that are available today in the AI space. You cannot not be focusing on where this world is going. And it can't just be a check in and play with tools for a little while and phone it in later. This is changing so quickly, you have to stay on top of it. I think it's not clear exactly where all of this eventually trends to, but if you have those core tenets available to individuals, I think that's what's going to be required. Just to highlight some of the tooling advances, because I think this is one of the key points.

John Furrier

>> Tooling has advanced so much in the past 24 months.

Derek Dicker

>> Yes.

John Furrier

>> What areas jump out at you that you say, wow, that's game-changing? Even physical AI, we had that discussion earlier. this is huge. These tools are very effective. They are teaching context, not just tools for putting stuff in a database for search. these are like building work into technology.

Derek Dicker

>> I'll tell you the piece of technology that I've played with the most at a personal level is Ollama, and then also looking at Hermes as an alternative play. And I've been astounded that once you connect it and begin the process of training it and asking it to do things, some of the insights that come back are completely unexpected. unexpected, the ability to actually have something go out, help manage calendar infrastructure, help book appointments, these are very much consumer-centric models and uses, but as soon as I saw that several months ago and projected it into what the working world looks like inside an enterprise today, it's like this is going to be quite a transformation.

John Furrier

>> AI scales superpowers, scales intellect.

Derek Dicker

>> Without a doubt.

John Furrier

>> And that's the superpower.

Dave Vellante

>> And everybody's stressing about the return, et cetera, but we talked about the bottom-up, like the PC era, the personal productivity. What's happening, I see, I wonder if you see this as well, is the projects are very focused. They have a clear outcome, they're getting done, and they're adding value in their little way. But eventually those are going to come together across an enterprise software stack, and it just takes time. If you're a multi-billion dollar company, you can't just rip that out and do it overnight. If you're a small company, you can, but otherwise you have to come at it from the edge.

Derek Dicker

>> Right.

Dave Vellante

>> And that's really hard. That's why it takes the better part of a decade to play out. But the impacts, I feel like, are going to be enormous.

John Furrier

>> Well, we haven't even talked about sovereign. The sovereignty angle now with AI is a revenue opportunity in countries and geographies. So, there's so much going on.

Dave Vellante

>> Well, by the way, a key piece of sovereign is it has to be open. Because otherwise, if you're locked into a particular technology stack and you don't have an open source alternative, then you're really not purely sovereign. Now, you don't necessarily have to go all the way to this extreme, but it's a consideration for customers.

Derek Dicker

>> Absolutely true, and part of the reason why we're getting pulled into all of these discussions in that space is exactly for that reason. People are looking for an open solution, they're looking for multiple options to go host on-prem in their own infrastructure and on their own soil.

John Furrier

>> Derek, thanks for coming back on theCUBE and sharing your insights. Congratulations on all the momentum at AMD. Love the new Instinct portfolio you guys have at the system level. It's all coming together, looking forward to catching up. Again, we could have gone an hour, but we don't have an hour. Thanks for coming on, good to see you again.

Derek Dicker

>> Thank you guys for having me again.

John Furrier

>> It's a pleasure as always to see you. Corporate Vice President of Enterprise Business Group at AMD, I'm John Furrier with Dave Vellante here in San Francisco live. Thanks for watching.