Dell Technologies World 2026 | Charlie Walker & Marc Hammons, Dell Technologies

Clips
More from Dell Technologies World 2026

Charlie Walker

Head of Product for Dell Pro Max Workstations & Rugged

Dell Technologies

Marc Hammons

Sr. Distinguished Engineer

Dell Technologies

play_circle_outline Agentic AI on Personal Devices: From Advisor to Orchestrator with Dell Pro Max GB10 and GB300

play_circle_outline Edge computing brings large-scale AI capabilities to laptops and desktops

play_circle_outline Balancing Cloud and Local AI: Tokenomics, Recursive-Agent Costs, and Payback Models for On-Prem Hardware Investment

play_circle_outline Use cases: software development, research agents, personal assistants, regulated data workflows

play_circle_outline Shadow AI accelerates experimentation; managed chaos versus unrestricted shadow IT

Info
Transcript

Charlie Walker & Marc Hammons, Dell Technologies

Charlie Walker

Head of Product for Dell Pro Max Workstations & Rugged Dell Technologies

Marc Hammons

Sr. Distinguished Engineer Dell Technologies

In this interview from Dell Technologies World, Marc Hammons, senior distinguished engineer of AI commercial clients at Dell, joins Charlie Walker, senior director of Dell Pro Precision Rugged Product Management at Dell, to talk with theCUBE's John Furrier about how agentic AI is bringing supercomputer-scale compute to the desk side and reshaping enterprise token economics. Hammons walks through Dell's portfolio of AI workstations – from the Dell Pro Max with GB10 to the GB300 – explaining how each supports local agent execution for tasks like code analysis, ... Read more

explore Keep Exploring

How has AI evolved into agentic systems, and how do Dell’s Pro Max devices (GB10 and GB300) support running agents locally and at scale? add

What should people—both enterprise customers and individuals—know about the trend of supercomputing/AI capabilities moving down to personal edge devices (Windows laptops/desktops with internet access), and how should they think about or frame the opportunities it creates? add

How do tokenomics (the cost of model tokens) affect AI deployment costs, and how can moving compute from the cloud to local/on‑device execution help control those costs? add

Where is this technology being deployed, and can you describe beginner-to-advanced deployment scenarios and how it fits into the product portfolio? add

How should organizations perceive and manage "shadow AI" — is it a problem to be prevented or a feature/opportunity to be guided? add

bolt Powered by CUBE AI

Charlie Walker & Marc Hammons, Dell Technologies

search

John Furrier

>> Welcome back everyone to theCUBE's live coverage in Las Vegas. I'm John Furrier, host of theCUBE, with Dave Vellante and Gemma Allen all out getting the stories. We're wrapping up day two, coming to an end, couple more interviews, but great lineup, all the top executives and leaders and engineers coming through theCUBE unpacking the AI infrastructure as it takes center stage and the implications are going all the way to the edge, all the way down to the client side, desk side devices. We've got two great guests here to unpack the future of how agentic AI is going to be on our devices and how that's going to play out. Marc Hammond, Senior Distinguished Engineer, AI Commercial Clients, it's great to have you on theCUBE. Charlie Walker, Senior Director, Dell Pro, Precision Rugged Product Management at Dell too. So you guys have the keys to the kingdom in this future era of personal computing getting seriously personal with agentic and some of the performance with tokens so it's also an enablement. It's a whole nother wave coming. Super exciting, so thanks for coming on theCUBE.

Marc Hammons

>> Oh yeah, thanks for having us.

Charlie Walker

>> Absolutely.

John Furrier

>> All right. So explain what, lots going on, but as it means to the desk side, agentic, what does that mean and how's that playing out? Because I hear agentic, I'm thinking, oh, large corporation running agents doing workflows, but it's applying to the personal developer, AI builder, entrepreneur, business user. They're going to have their machines.

Marc Hammons

>> Yes.

John Furrier

>> Segment that out, where does everything fit? Let's start there.

Marc Hammons

>> Yeah. All right.

Charlie Walker

>> All right. I mean, maybe what I'll do is very quickly start at a high level. As you jump straight to the agentic is helping some people understand really what AI has gone through the evolution. And Jeff said it really well this morning is AI has gone from that advisor to now the orchestrator and the doer. Rather than that serial linear process where it's prompt response, prompt response with an AI model, now you have this world where the agents can go do real meaningful work for you.

Marc Hammons

>> Yeah, and that's happening across the portfolio. We like to start down with our Dell Pro Max with GB10. Fantastic device, I can run a few agents on there and they can do a few things for me. They can do research in the background. I have a clawbot these days that goes off and do research. It's really hard to keep up so you can imagine just an agent sitting there going out, reading a bunch of articles, summarizing them, sending them to you for a daily task. And then we start to scale up to more serious workloads where maybe I'm looking at code repositories on site. I want to keep things private and local, I don't want things to escape into the cloud. I may have a series of agents that are doing that. You probably saw some of that in Jeff's keynote today. And then we really scale up to Dell Pro Max with GB300 where we can put effectively a data center size chip right there at your desk side.

John Furrier

>> It's a super computer.

Marc Hammons

>> It effectively is.

John Furrier

>> Say it.

Marc Hammons

>> It's the right thing to say, because that's exactly what it is.

John Furrier

>> I ordered the code red, I mean, that is serious computing.

Marc Hammons

>> Right.

John Furrier

>> I mean, that's massive. We saw a taste of it last year at GTC. It's here.

Marc Hammons

>> Yeah, absolutely.

Charlie Walker

>> It is here.

John Furrier

>> Yeah.

Charlie Walker

>> And that's where, I mean, Dave said it this morning where, especially as you're thinking about that GB300, there's so much compute. You can actually now begin to think about, I'm going to run a company on that. That becomes a very core foundation to that AI native mindset of how do I go become agentic first, AI first?

John Furrier

>> I have to apologize for making the reference to A Few Good Men. Jack Nicholson scene where he's like, "I ordered the code red. You can't handle the truth."

Marc Hammons

>> That's right.

John Furrier

>> The truth is we are now seeing a tsunami of computing coming down to the personal edge, the machine that we know is either a laptop or a desktop computer. Running Windows, it's got access to the internet, but this now takes it to a whole nother level with super computing capabilities, which we're seeing in the large scale side on AI factories. So you're starting to see that come down. As you guys look at this, what should people know about this? Because I think one, it's going to be immediate opportunity for an enterprise customers, also individuals. How should they think about this? How should they frame it? How should they lean into it and to start getting going?

Marc Hammons

>> Right. I think what we're all experiencing today, it's some of the most powerful AI that we've ever had at our fingertips, but we're also experiencing the sticker shock of that as well, right? I mean, it's great to see the advancements that are happening in the cloud. The cloud is where frontier models go first. It's where cutting edge AI goes first. It's also where your costs are going to be buried if you don't do something about that. I think the opportunity that exists, and you've heard it, it's a story that's been said many times this week is to bring some of that compute locally on the machine and start to adjust the tokenomics of the situation. And these things are more than capable of doing that when you're talking about these open models that can support hundreds, if not thousands of tokens per second, and you don't really need every token to go to the cloud.

John Furrier

>> Token economics is interesting because people are blowing their budgets now. People don't even know what they're doing. They're experimenting, but they're starting to get visibility on real value, but they've almost specced it wrong if you think about it because they're spending all that cash so they have to then recalculate what it might look like at a steady state. So having the ability of having your tokens forecasted either on a test bed or an environment where it's local, that's where I see a lot of value. What do you guys see about this? Explain this concept because I think this is what people are looking at right now and talking about.

Charlie Walker

>> Oh, you articulated it really well of part of agentic is that recursive model. I mean, Marc talked about it earlier, you think about a research agent, it's going out, looking at, finding one paper, then it's finding something else, analyzing it, and so it's continually learning, continually developing. That burns tokens. As you do that, it becomes really expensive. As you start to pull it on device, now, really what you start to do is you pay for that device very quickly. You can very quickly pay for, people look at it and they're like, "Okay, that's a very expensive device," something like a GB300, but you pay for it within three to six months. And so it becomes very economical when you start to actually think about the tokenomics behind it versus just the hardware investment.

Marc Hammons

>> Yeah, and I think there's a good dichotomy there. You do need those frontier models in the cloud for that repository scale reasoning where it's planning at scale, doing some kind of massive organization of things, but then it can delegate that down to these machines that are desk side and really let it take over and drive the individual efforts.

John Furrier

>> I just got to say, and just chatting with you guys and then listening to the past two days here at Dell Tech World, as well as other CUBE interviews, there's a lot of things that are coming out of the core infrastructure, engineering culture, infrastructure concepts into real world. One that's fascinating to me with this token economics is a corollary to PUE, which is our data center measurement around energy, total energy available, and how much is spent on IT, oversimplified, but that's basically the formula. In a way, you're getting at the same formula. You're saying, "Hey, what are you spending?"

Marc Hammons

>> That's right.

John Furrier

>> On tokens.

Marc Hammons

>> That's right.

John Furrier

>> To what's your over available cost? Meaning it's kind of a cost of ownership calculation. If you say, "Hey, I'm going to spend $30,000 on this system," give to a team or whoever, and they're burning unlimited all these tokens. Who cares?

Marc Hammons

>> That's right.

John Furrier

>> You've already kind of won.

Marc Hammons

>> It's sunk cost.

John Furrier

>> Versus spending 30 grand on tokens, and you got a prototype that costs in time and later and other expenses, 200,000. I mean, I'm making the numbers up, but you can see that this is where this comes into math.

Marc Hammons

>> That's exactly right.

Charlie Walker

>> And part of the beauty is it's all about the more iterations you do, the better the ultimate outcome is. Part of what we've always talked about around desk side AI is you now can go have the freedom to go test, experiment, learn, pull the latest models, change weights. So you can do all kinds of things, or I'm going to go build new agents and try new things or all of that now leads to much better end outcomes rather than trying to go be conscious of, "Oh, I have a limited cloud budget. I don't want to go spend all that."

Marc Hammons

>> That's right.

Charlie Walker

>> "I need to go do these other things."

Marc Hammons

>> Right. That gives you the freedom to explore, which you're not going to get if you're just sitting there thinking about costs all the time.

John Furrier

>> I think you guys are onto this total cost of ownership because I think the number one enterprise thing I hear from CFOs, lines of business leaders is, and Jensen said it on the first day, token anxiety, which essentially just is the juice worth the squeeze?

Marc Hammons

>> That's right.

John Furrier

>> This comes down to classic IT, payback, total cost of ownership. We've lived that. How do you guys think about that? Because the enterprises are thinking, what do I do and what step? Obviously, it's a purchase involved. When you're on-prem, you're not really spending cloud tokens. You got it all there. You can bake it in the oven.

Marc Hammons

>> That's right. Go. Well, the other nice thing about the portfolio that we've got is that you think about, hey, from an infrastructure perspective, it's all in one box. I've got my storage, my network, my compute, my AI all in one thing and I can kit that up. I can actually put it into a trailer. I can ship it anywhere in the world. I've got the net benefits of data sovereignty, privacy, protection, all in one kit, and it makes it a much more appealing proposition.

John Furrier

>> All right, so take me through use cases. I want to get into this in action.

Marc Hammons

>> Yeah.

John Furrier

>> Where is this playing out? Can you talk about from beginner to advanced scenarios of how this is rolling out and how it fits into the products?

Marc Hammons

>> Yeah. One of the great things about that, let's see, A, we're based on effectively the NVIDIA portfolio, right? We've got a software stack that can scale all the way from GB10 all the way up to GB300 and across the tower series, that's fantastic. That means I can experiment down on a GB10, and then it'll scale up even within my own client infrastructure and then up into the larger infrastructure solutions group solutions as well. From a use case perspective, definitely software is where people are starting to tinker, so software development is one of these things that I think just everyone gravitates towards. The world has suddenly become code. Even if you're typing a prompt to create a document, it's writing code for you so that's one of the primary use cases that we see with customers. Research agents is another one where I've got some huge research task, I want to have it go off and do something for me, create some summary, it makes my life easier. That's how I keep up with the things that are happening in AI today. The other is personal assistants. Let's say you've got private data. I may be constrained by regulatory stuff like HIPAA or FERPA, these types of things. I can again put that in a sovereign system off on its own and I don't have to worry about that data going anywhere. And so these are some of the use cases that are pretty common across all the different industries.

Charlie Walker

>> Marc nailed it. Well, I think the only thing I'd very quickly add is, and again, I'm going to tie back to the tokenomics and the value that you extract. I know someone who does a lot of research, they will do one run on a research project, $600 in the cloud for one run. Now it takes over an hour, but it's a massive job, and so again, now you can think about, all right, now I can start to pull some of that on-prem, but that tokenomics starts to make a lot of sense.

John Furrier

>> You know what's interesting, Dave and I were talking about repatriation, and we've had many debates. I was not a repatriate fan, not fan, but I didn't agree it was actually happening from a lift and shift cloud perspective. But I do see repatriation, if that's what you want to call it, just by overall growth of the enterprise, because if you have the data on-prem and you have the engagement levels of the people doing things with the data that's on-premises, that's on-premises, and it's not happening in the cloud. So if you don't have to go there and the numbers all map out, why do it?

Marc Hammons

>> Exactly.

John Furrier

>> Okay, so that's one thing. There's not a lot of lift and shift pulling back from cloud to on-prem. I think it's on cloud and let's extend it on-prem and do more with what it means to be on-prem. Okay. That being said, I think you wouldn't disagree with that, or maybe a little bit, but my point is this is now the next question. How do I tokenize my entire company? Because if I can get everything on-prem, tokenized-

Marc Hammons

>> That's right.

John Furrier

>> I'm essentially a token factory, AI factory on-prem.

Marc Hammons

>> Then you're getting that benefit across the company.

John Furrier

>> It's a TCO calculation.

Marc Hammons

>> Exactly.

John Furrier

>> That's it. It's an IT purchase on steroids, but it's a business infiltration, not IT serving people with.

Marc Hammons

>> Exactly, and that's about modernization. That is, I think the next step for just about every enterprise out there. They're going to have to undertake this exercise where you go in and you look at the workflows across the different departments in your company, and you codify that. What are we doing? And then it's a subtle transformation because many of those processes are human to human. They're not agent to agent right now, so there'll be a transformation of those over time for, again, optimization of that agent to agent flow.

John Furrier

>> I think Jensen gets a lot of, I won't say shade thrown on him because this guy's a legend, but, in what he's doing, but people don't understand that he's not cheerleading for token economics and tokenization, he's cheering for computer science because what he's basically saying is if you tokenize the benefits you'll get, and again, the coding example I think is a real, I think I would call it the first real enterprise use case that's meaningful. Not that RAG and good marketing and good document management stuff is bad.

Marc Hammons

>> They're useful.

John Furrier

>> Very useful, but it wasn't needle moving.

Marc Hammons

>> Right.

John Furrier

>> It was, "Hey, I can sign something faster," or, "Hey, vector embeds are cool. I get a little flywheel going." But coding is you can't deny the value of whether it worked or not.

Marc Hammons

>> Exactly.

John Furrier

>> Either the code got shipped and worked or didn't work. And so people are seeing real results that the CFO can say, "Wow, my line, I did that project. We shipped it. We drove revenue. How many people did it? Three?"

Marc Hammons

>> Yeah.

John Furrier

>> I think this is where the enterprise starts to get savvy on the value creation, and then the extraction piece comes in. I mean, I'm talking around it, but this seems to be the new evaluation.

Marc Hammons

>> Yeah.

John Furrier

>> What do you guys think about that? What's your reaction?

Marc Hammons

>> Yeah, completely agree. A few things have happened fundamentally, and Jeff said it a little bit of a different way, but models move from chat to action, right? That's the first thing that really happened. And then what we're talking about here is that run times became useful. Now there's a harness that takes these models and puts them to work.

John Furrier

>> Explain harness, people that don't know what a harness is. I think of a harness like held up on it.

Marc Hammons

>> Yeah. You think about a model, right? A model really just, there's an input and there's an output and that's all there is. But you need something that puts a process around it, puts the loop around it that transforms that into something useful over and over and over so that it can reason, so that it can rationalize, so they can take action, call tools, all the things that happened that expend tokens. That's where all the cost comes from. Those have finally become really, really useful. We saw the OpenClaw moment back in February that's taken us by storm now and now everyone's like, "Oh my gosh, I could actually use this to do my job." Not just a software engineer's job, not just a let me put in a chat prompt and cut and paste that into a document, it's actually now able to deliver value.

Charlie Walker

>> It becomes a true force multiplier. I mean, that's really what it's doing.

Marc Hammons

>> Absolutely, yeah.

Charlie Walker

>> It's another tool for your workforce that every company has got a million things that they wish they could get to and they just don't have time.

John Furrier

>> And they give up.

Charlie Walker

>> And they give up.

Marc Hammons

>> That's right.

Charlie Walker

>> But now you have the force multiplier that allows you to go offload, you have agent-agent interaction, and now those people can go focus on all of those other things that are more strategic, better opportunities to go drive the company.

John Furrier

>> I was talking about this with Michael Dell and Jeff Clark and Arthur Lewis. Shadow AI is a feature, not a bug.

Marc Hammons

>> Yeah. Yes.

John Furrier

>> Because if you think about shadow IT, you had an IT department, you go around it, put your credit card in the cloud and get the prototype out there, show your boss and team, get your hand slapped and then you get promoted to run the project.

Marc Hammons

>> Exactly.

John Furrier

>> Okay. Good job. Don't do it again, but what you do again?

Marc Hammons

>> Yeah.

John Furrier

>> AI is also impacting business so people feel first party primary data of the value. It could be in the CFO office, and so they now actually feel they're not relying on IT, so it's not a dysfunctional thing. It's like, okay, they go, "Okay, that's fast. I get it." And then they start mobilizing kind of like how Dell's been doing it for three years. So I think the shadow AI experience could be reigned in a little bit. I'm not saying it's total chaos, but you can make it managed chaos.

Marc Hammons

>> Yeah, you can.

Charlie Walker

>> But a big part of AI is you just have to start. You got to be curious. You got to experiment, you got to try. And that's really the value, and so those people that have done the shadow AI have that head start.

John Furrier

>> All right. So final question for you, personal one. Now that you're in with the position you're in with the product, all your friends coming out of the woodwork saying they want a unit.

Marc Hammons

>> Absolutely.

John Furrier

>> Is that a hint?

Marc Hammons

>> Yes.

John Furrier

>> No, I can imagine. I mean, you'd probably be very popular, and we haven't seen that kind of horsepower at a level that's that agile.

Marc Hammons

>> I think we are getting that and the great thing about that is people are hungry for it. They're like, "Oh wow, that looks transformative. I want to experience that." And I think that's something that's been unique about this moment. AI is going to level a lot of playing fields, but it also has taken all of us back to square one, so we're all right out of the gate trying to do what we can to keep up.

John Furrier

>> I mean, gaming culture showed what happened with GPUs before NVIDIA became huge, and all my kid, my two sons would unplug my GPU and put a other one in, I wouldn't even know it. Like, "Why'd you do that?" "Dad, you're just running Windows. I need it." "No, I got the cheap cores."

Marc Hammons

>> I need that more than you, Dad.

John Furrier

>> Yeah, what I paid for. Yeah, but that phenomenon is happening because gaming, the performance got better quickly, and then obviously AI so I think you guys are in a good spot. All right, final question. What do you guys got on the roadmap? What's coming next? What are you guys thinking about? What are you optimizing for?

Charlie Walker

>> Yeah. Well, it's a good question. So I can't obviously go into too much detail on future roadmap, but we're going to continue to focus on how do we go make it easier for customers to go deploy. And I mean, one of the things we didn't really touch on, I will say just to tie it in, is as part of this agentic AI movement, it's the services piece as well. So back to where customers are trying to pilot, explore, learn. We also, with our services expertise, all of our learnings from the data center, we're able to go bring that down and really go help launch customers quickly so they see the value.

John Furrier

>> Awesome.

Marc Hammons

>> Yeah. What do I see next? Yeah. Better, faster, stronger. We're just getting started. Think about it. Where were we a year ago today?

John Furrier

>> I think it's a template, it's a sign of a tokenized economy, businesses start with some small teams, then it's the whole company.

Charlie Walker

>> I mean, just to build on what Marc said is very simply, AI is the dumbest it's ever going to be, today.

Marc Hammons

>> Today's the worst thing.

Charlie Walker

>> It's only going to get better.

John Furrier

>> Yeah. And I think the sorting between winners and losers is going to happen a lot faster because that gap, the sooner you get behind, it's just to your point.

Marc Hammons

>> That's exactly it. The people that are going to win, they're going to figure out what to run where and why, how to govern it and how to show the benefits, right?

John Furrier

>> Yeah. Jeff Clark called it the super user. I call the CIO, super CIO, super user. You got all this new dynamics.

Marc Hammons

>> Yeah.

John Furrier

>> It's going to change the organization, which is why everyone's a C-suite and large enterprises are paying attention. Guys, thanks for coming on theCUBE, really appreciate it.

Marc Hammons

>> Yeah, .

Charlie Walker

>> Absolutely, thanks for having us.

John Furrier

>> I'm John Furrier, host of theCUBE, with Dave Vellante and Gemma Allen. We'll be back with more after this short break.