← Reshaping Workflows with Dell Pro Precision and NVIDIA RTX PRO GPUs
Reshaping Workflows with Dell Pro Precision and NVIDIA RTX PRO GPUs
Inside the Dell AI Factory 2.0 with Jeremy Williford
00:52:13
NVIDIA’s Jeremy Williford explains how Dell + NVIDIA are transforming AI development—from desk-side tools to enterprise-scale AI strategy.
Loading summary
Transcript53 lines
- [00:05]
Jeremy Williford
Welcome to Reshaping Workflows with dell Pro Max PCs and Nvidia, where innovation meets real world impact in high performance computing.
- [00:20]
Logan Lawler
Welcome back. We have another exciting episode of Reshaping Workflows with Dell Pro Max and Nvidia RTX Pro GPUs. I'm your host, Logan Lawler. And throughout one want to stop just before we get into it and just appreciate and want to thank for everyone that's like listening to the podcast. So far it's kind of exceeded my wild expectations for. So for those that are listening and enjoying our content, please like subscribe, download, keep it coming. If you have feedback, we'd love to hear it. So with that thanks getting out of the way, we have a very exciting guest today. We have Jeremy Williford who works at Nvidia, and we are going to talk all about the Dell AI factory with Nvidia, not 1.0, but 2.0. So with that, Jeremy, welcome. Maybe take a second to kind of get the guests up to speed a little bit on you, your background, your role at Nvidia, and then we'll dive right into it.
- [01:11]
Jeremy Williford
Hey Logan, thanks for having us. I really appreciate being here. My name's Jeremy. I started Nvidia back in 1998 and so whenever I started we had probably about 60, 62 people total at the company. We'd already had two products that, you know, didn't do as well as the shareholders or there wasn't shareholders. How about this the investors had wanted us to do and we were really on the cusp of going out of business and so they brought me on board, we signed two large deals. The next thing you know, we took off. And so I've been fortunate enough to be with Nvidia for the past almost 28 years now. And really in, in my journey throughout the company, one of the great things about Nvidia is we've always stayed in that same culture of a startup environment. We, we really think about building ecosystems and we think about what it's going to take to reach developers, what it's going to take to, you know, educate structures in the market. And so throughout my journey with Nvidia, I've always been focused on kind of strategic growth and what we're trying to do as a company to go, you know, reach that next level. So my official title is a VP of Strategic Sales. But the reality is I, I really think about our, our business and how we're going to go partner and market to make sure that we have the right ecosystem of software Developers the right hardware structures. I, I, I am partnered with Dell to go do this mission and primarily Dell is one of our key partners in market because of breadth of, of offerings. You know it's, it's the largest, you know, PC manufacturer from a workstation standpoint. They're the largest server manufacturer, the largest market share storage company. All of the structures that really make Dell a fantastic partner. I've been given a lot of, you know, freedom from Jensen and from Michael Dell, two founders that really think about how we grow the markets together. So, so that's kind of what I do and the role I play at Nvidia. I have a global team of people that kind of help me in this space but really, you know, but that's my day to day.
- [03:14]
Logan Lawler
I'm still stuck at your employee 60 some odd and started Nvidia 98. I mean that's amazing. I mean I, I think yeah, you're the longest tenured Nvidia employee I've ever actually had. The, I've never, never talked to Jensen one on one. But I mean that's amazing, right? And I think we're going to get into that a little bit and I would absolutely agree with you. Is that what I love and we'll get into this a little bit is that, you know, unless you've been living under a rock for probably the last two or three years, right. Nvidia obviously makes, you know, GPUs, accelerators for AI and a bunch of traditional work where I play specifically is Dell Pro Max kind of ISV workflows. But Nvidia is so much more than that and we're going to get into some of the softwares and stuff like that. But I think where we should start Jeremy is I kind of teed up in the beginning which is the Dell AI factory with Nvidia. Right. And I know at DTW just a few weeks ago, you know, there were some kind of big announcements and for those we've talked about, you know, the Dell AI factory with Nvidia. It's really the partnership that Dell and Nvidia has, you know, obviously between Jensen and Michael. But it's really the hardware, the software, the ecosystem to be able to iterate prototype whatever kind of your AI journey looks like and scale that from a workstation all the way up to the data center from deployment with you know, Dell validated designs, et cetera. But we were talking about Dell AI factory with Nvidia 2.0. So maybe we can start. Jeremy, let me ask based on the explanation I said, what is different about the Dell AI factory with Nvidia, with the recent announcements coming out of dtw?
- [04:41]
Jeremy Williford
Well, it's a great question, Logan. I think the first generation that we had for AI Factory, you know, it really had a lot of component structures kind of built in a way that helped customers especially that were already starting a training type workload. But now, you know, the market for AI has really evolved. Now you have a structure where you're needing to provision these workloads. You need to think about not only, you know, how the GPUs interact with one another, but how the nodes work with one another. And so whenever we think about structures like GB200 and in that protocol now, no longer are we putting server boxes together. We have an entire rack structure of infrastructure, not GPUs tied to each other, but also CPUs tied to the same GPU network through an MV link interface. And those structures are amazing protocols of how PC architecture and acceleration level works in getting the most out of every single device, be it a GPU or cpu, so they can all share a structure of memory and be able to work from that. Not only is that portion of a 2.0 now, you're also thinking about how the storage interfaces with those compute structures, right? How are we getting real time information? Because in a world of agentic AI now you're working on how each one of these processes works from an inferencing standpoint and then reaches back into the storage to look at that same structure. They just build to do checks and validations against that. And that really comes down to a software structure on top of those hardware, compute, network, storage fabrics, all the stuff that's working together. Now you're thinking about how the software operates on top of that. And Dell and Nvidia have done a recent announcement where Dow Dell can now be a provider of Nvidia AI Enterprise. That software stack now comes directly and supported directly from Dell. That's a huge deal for our customer base because in the AI Factory 2.0 it means that our customers can do transactions with Dell without having to go through, you know, end customer license agreements. With Nvidia, they have their own agreements and structures with Dell. Dell's able to provision that directly. And in the world of AI Factory, most of the structure will be based on and predicated on what software structure are you using, what models are you using, what ISV elements are you having in place. So as we think about how the customer's evolving and what they're doing with these factories, you Know, we have to think about a full end to end solution. You know one of the great things about Dell is you guys have not only the full, you know, compute, networking, storage, all the structures that are required inside of data center, but you also have products like, you know, now you're going to release the Dell Pro Max with GB10 and this is a desk side representation of that same factory but now you have it in a desk side what looks like a Kleenex box the size of so small, but it's literally the same performance that we built whenever we built our first eight way server back in 2016. So it's an amazing vehicle to help AI developers build, refine that software, stack on a desk side system and then deploy on a full AI factory 2.0. So that being said Logan, I feel like there is a, a lot of enhancements from an AI factory standpoint that now as we both grow the technology together, we give so many more vehicles, so many more capabilities to our customers.
- [08:16]
Logan Lawler
I think that was well said. I mean at the end of the day, I don't know if anyone listening this has bought AI Enterprise, you know, Nvidia Enterprise from Dell before. I will say that I've had a few customers that have and bringing it kind of into the ecosystem, I mean that's a really big deal. I mean not that it wasn't easy to get before, right? But being able to service it and have that all kind of tied in manageability and all that makes total sense to have it. But you, you mentioned kind of the system on chip and designs we're going to get on GB10 but kind of peeling a little bit back, right. Like in terms of, and I'm not a poweredge expert by any stretch of the imagination, but you're thinking Nvidia kind of H1 hundreds, H2 hundreds on the workstation side. You know previously you were thinking 6000 ADAs, now you're thinking RTX Pro Blackwell 6000s. So but recently you've kind of moved into more kind of a what I'll say a system on a chip design when it comes where there's the integration of kind of the memory, the processor all in one device, maybe talk kind of give a tad a bit of a, I mean you can go as technical as you want but kind of explanation of that. But then why kind of the, the move from more of a traditional GPU into that. Not that that's being replaced but you know, but why, why are we moving into that and why are we seeing that?
- [09:27]
Jeremy Williford
Logan, that's a great question. I mean, the reality is, it is fascinating to me because from a PC architecture point, Nvidia was kind of born out of the idea that we believed in accelerated compute architecture. We believed in a parallel pipe, you know, structure. And so as I look back on the history of Nvidia, we've always had this mindset of how to do large computation construction. How do we do, you know, more pipeline of memory into our processor so we could fill up these parallel pipes in a bigger, faster clip. And so really this is just the natural progression of what we've always done as a company. We recognized early on. In fact, if you think about why we wanted to acquire ARM at one time, right, we wanted to because we're using their architecture and that ARM architecture we've been able to refine in such a way that we offer a wider availability of data structures to flow into our GPUs. We're able to put those same ARM CPU architectures on our NVLink fabric, which allows us to get more bandwidth into the GPU. You're looking to feed those GPUs faster so we can get real time information back. We can have token production at a higher clip. We can go in and do reasoning stacks much better whenever we have the fabric and the data structures and data flows coming to us at the right times. So whenever we think about how we did Grace Blackwell Solutions and the GB2 hundreds, now coming out with the GB3 hundreds, we really put a heavy emphasis on how all those structures work together, basically to make it look like a giant gpu. I mean, it's just this massive infrastructure, full rack configuration, ridiculous amounts of wiring to make it all work correctly. But the reality is the more we can have it operate as a single architectural flow, the better we are going to be able to do these real time inferencing reasoning structures that we need to go do. Now boil that down. What does that look like whenever it's desk side and whenever a user needs to understand how that workflow does. Now we're no longer in the x86 environment, we're in this Grace Blackwell ARM GPU environment. And we need to showcase and demonstrate how that structure works. Well, there's no better way to do that than to set it desk side and allow a user to plug it into a wall and go into air gap solution and be able to do and work with his stuff without having to go, you know, go get cloud time or cloud access. And if you could do this for under $5,000 for each of your AI developers, I gotta tell you that that's like free less than one month at aws, I guess for some of these guys. So that being said, we think about the architecture level, we want to shrink it down, bring it to developers so they can use it, they can work with it. And that's how we think about the structure and why it's so important to be able to, you know, operationalize that GPU CPU interaction and make sure that everything has the widest bandwidth possible.
- [12:28]
Logan Lawler
Sense, right? I mean really if you boil it down, I mean two things, one is speed for a system on chip design and you can feed the bees faster basically and you could get more packed into a tiny space. I mean you kind of mentioned it before. The Dell Pro Max with GB10 is launching soon. Fingers crossed, coming up, you know, hopefully in the next few months. And then we have GB 300. And for those that don't know, that is a Delpro Max system. It's a kind of video system on a chip. Grace Blackwell architecture design has 128 gigs of integrated kind of CPU, GPU memory and it's exactly that. It's designed, it runs and hopefully I'll get the nomenclature right. It's the Nvidia Nvidia DGX OS which is kind of the Nvidia flavor of Linux. And it's really targeted at developers at the desk side. I mean basically what you've said. But what I want to, what I want to ask, and you kind of made a really good point that I want to like tag onto is that I can't tell you how many AI conversations that I've had with different startups, different companies and you know, they might not be using Dell and they probably should be, but they're in the cloud. And I remember one very specific where they were using the cloud and then they got their first bill and they were like sticker, sticker shock. And that's really what the GB10 is designed is to equip someone who's a data scientist, someone works on neural networks to be able to unleash them with this really powerful box. It could be a standalone, you could daisy chain it together, it could work in supplement to a system that you already have. But one thing I want to ask because I know that we, you know, obviously Travis Wells who works for you, I work with him a lot at Nvidia. Is that really the only use case and kind of Persona that GB10 or you know, the Nvidia Spark kind of founders editions going for? Is it Just kind of AI developers or are there other use cases outside of that?
- [14:07]
Jeremy Williford
It's a great question Logan. I mean, because I've, I've been approached by a ton of guys that were like, hey, it's a great product. We should stick this on the edge. This thing weighs less than £3. It's, you know, it looks like I could put an industrial area. Let me start with this. We, we really designed this product so that customers can do exactly what you mentioned up front. Right, right now the vast majority of AI developers will go straight to a cloud instance because it's, it's the easiest, it's the only thing they have. Logan, one of the problems that we probably have is that industry is we don't have enough accelerated GPU, you know, accelerated computing or GPUs in standard IT infrastructure, right? So if you're a developer on prem, even at Dell and you said hey, you know, five, six years ago if you were a developer on Dell, you would say hey, I need some GPU access. And IT guys would say well I don't have that. And you, you just go off to the cloud. One of the great things that Michael Dell and Jeff Clark did was they really decided, okay, we need to build our own on prem infrastructure for our own developers. We need to think about how they want to develop and we need to build an AI factory for them to come to. And it really made a tremendous difference in how Dell launched their own AI use cases and how their developers interact with that. But many customers don't have that, that same provision from their IT guys because they lack the, you know, the vision of being able to do that. So one of the elements we've been able to do is put this, you know, this GV10 together with Dell and it's going to be a fantastic vehicle for AI developers. But to that point we built it very specific with a Linux structure. It runs a DGX OS and it's, it's very, you know, proficient at what it does for an AI developer. They can run the full Nvidia AI enterprise stack and it allows them over access to over 200 different applications in there runs, most models can fit in there, can still run. So it, it's perfectly suited for that. It's not really meant for an edge type deployment or activation, which is unfortunate because it's, it's a great form factor in size. We have other products that are really meant for the edge that are still system on chip. So we have a, an Orin product and that Orin product Is is perfectly suited for as an SOC for edge deployments. That market's super fragmented though. So you know, if we work with Dell we have to, we have to find a way to get that into the system. Many edge deployments look at just using standard servers or workstations for that matter that have been ruggedized and have that same feel and they run on x86 and they do all the remote management stuff you need them to do. So really right now it's just for AI developers. I think to your point, as we look forward, you're going to see more and more use cases and applications, you know, suited for those things.
- [16:46]
Logan Lawler
Yeah, no, absolutely agree. I mean I've had, I've heard use cases around potentially rendering, you know, the edge use case, you know, I've heard a lot and we've kind of danced around it. But the Dell Pro Max with GV10 small mini box, I mean literally like yeah, great example, like Kleenex box. But then we also have the Dell Promax GB300 which is, you know, you go from a thousand AI tops at FB4 up to 20,000, which is basically a server.
- [17:12]
Jeremy Williford
You could run a trillion parameter model.
- [17:14]
Logan Lawler
It's insane.
- [17:15]
Jeremy Williford
Let's just sit on that for a second. Trillion parameter model, desk side plugged into the wall. That's incredible.
- [17:22]
Logan Lawler
It's incredible that performance there. But that's kind of where my question was going. It is such insane performance and it'll be coming out in kind of second half later this year, early next. But when it does come out, what is the market that that's really designed for? And what I mean is, is that, I mean that is server level performance in a case like a normal size kind of workstation fixed case. Right. Who is, is that targeted more at those that might not have, you know, a data center that are doing AI development where that box can kind of be split up piece and parse across, you know, different developers. Is there a different use case? It just, it's just seems so powerful that it, it should be in a server farm. Do you know what I mean? I'm curious to hear your answer.
- [18:03]
Jeremy Williford
So it's interesting if you look at the new world economy, Logan, and you think about customers. I love Larry Feinstein was on stage with, with Michael Dell during day one of keynote for Dell Technology World. And you know he said that JP Morgan Chase was going to spend $18 billion. 18 billion on infrastructure. And that's required for a GenTech AI workflow. Right. If you want to do, you know they have Thousands, tens of thousands of developers, JP Morgan Chase and they, he was describing the environment he wants to bring those guys into and how they wanted to develop agentic workflows in their own structure. So that means they're, you know, they will build an entire data center full of these GB3 hundreds rack servers and you know, the storage and everything that goes with them. If you think about the investment these guys are making and everybody's making this type of investment in order to, you know, to really augment how much productivity they're able to have. Think about how agents will help each one of us do our daily tasks and our daily jobs and how to use data as a structure to really move your company forward. If you think about the investment they're putting in there, investing in these desk side GB3 hundreds to help the developers to build out on that architecture and influence what goes through that data center. That's a no brainer if you think about how much data scientists, you know, make to go build these structures inside of the companies and how valuable they are. Man, you're talking about giving a tool and an asset to that developer that allows him to do, or he or she to do any of the work they need to go do on their desk side environment without having to go off and get provisioned GPUs and provision stuff. That is a tremendous tool. And the value, I mean, gosh, it's hard to put a price tag on what, what that gives a developer in house. Remember they're able to do that and not have to go off that. So, and really what you're getting with that GB300 infrastructure from a, from a Dell Pro Max perspective is you're getting essentially what's in that server. So you're actually working on almost the exact same architecture at a smaller scale, but you're doing it desk side. So you're really able to really fine tune it for how you want it to run at scale whenever you get off of there. So I, I don't know, Logan, I, I suspect that that is going to be a drop in the bucket for these guys that are making large investments for these data centers.
- [20:28]
Logan Lawler
It's true. I mean in don't know the pricing on that, you know, all that'll shake out in the wash. But you're right, I mean it's, you're not going to go out and buy a $500 chef knife and give it to me because I can barely, you know, cut a peanut butter and jelly for my daughter. Right?
- [20:41]
Jeremy Williford
Do they make those?
- [20:42]
Logan Lawler
Yeah, I think they do. I get ads for it all the time.
- [20:45]
Jeremy Williford
I didn't know they make those. I just assume the rusty knife that's in my wall, that's the only one they make.
- [20:50]
Logan Lawler
No, I think they do because I've been getting ads for it all the time. They think I'm like a pro chef and I can't even make a peanut butter jelly sandwich. But you're right, like equipping the right tool for the right person, I mean it makes total sense and in the grand scheme of things at, you know, what it costs to find a very good data scientist AI developer, it makes total sense and you kind of touched on it a bit. But I mean that is really the connection, I think. And not saying it hasn't been there before within the Dell AI factory with Nvidia, right, but we had workstations, right? And traditional, you know, precision Dell Pro Max, right? And you scale up and down to the data center. But now there is this really cohesive fabric, whether it is Dell Pro Max with GB10 or you know, GB300. But then you layer in kind of the Dell or the Nvidia AI enterprise stack on top of that and it is truly death side deployment back down, fine tuned back up in a very seamless way that I don't think really. I mean it existed but not as in a seamless way that it does now. You know, after dtw and I think that's a big call out, we've done.
- [21:52]
Jeremy Williford
A good job collectively on the Dell workstation line to be able to do Linux on Windows and to be able to do that, that entire structure where you can open a Linux environment and be able to run that and operate on Windows, I was blown away. But you know, even the, the Windows, you know, the Microsoft CEO came out and said, hey, we're in big, big support of this and, and it is a vehicle that we have today that people can use on these, these notebook environments and these desktop environments. But I would say to you Logan, handing someone that something that's out of the box, ready to go and ready to be deployed and has the full software stack on it, there's something super valuable about that, especially at these low costs that we're talking about for GB10 and for the high performance structures that we're Talking about with GB300, I think it just brings a new paradigm into the market. I will say this though, I think that it's a different buying market than what is traditionally a workstation buyer. So if you think about all enterprise customers have, you know, Application support and if you're running any AEC application or any media type application, you have a workstation because you're a professional, you know, you're a podcaster, so you have a professional setup with that. Those, you know, if you go look at where the, the full AI developers sit and where they are, we're going to have to make sure that collectively we go reach them with this new product. Because it's not a traditional kind of workstation cell, it's more like a, hey, let's make sure these developers know this is available. Let's make sure it knows how to get these in their hands. Right. Because it's not a, it's not a normal workflow. So we're have to do a lot to kind of get the message out there to make sure that people understand what this does, what it's capable of and, and who are the right audiences and target audiences to use it.
- [23:35]
Logan Lawler
Agreed. You know, I've been a part of several product launches within Precision and Dell Pro Max. This one's the most unique. 100% agree because it, and maybe someone correct me, I've said this statement along, so I don't know if anyone correct me at this point, but it's kind of the first purpose built system where you know, I personally use a precision 3680 with a 6000 ADA. Once I get my hands on Blackwell, I'll do that and I'll upgrade to that. But you know, I can do the traditional ISV kind of the Catia, the Autodesk M and E Adobe substance type workflow, but I can also use WSL to do some basic kind of AI development as well. But it's kind of the first purpose built system. So there's been a lot of time spent around not only educating customers, but really educating our salesforce and everyone on what really GB, GB10 and GB300 is and it isn't. So it's a really good point that at the end of the day with this purpose built system we have to go out and educate the market and with that you've kind of, we've kind of alluded to kind of some software, but Nvidia is more than just a GPU company. So there I said it. I think everyone knows. But what are the two kind of big software applications that Nvidia? Obviously you don't have to tell us something that you have in the works, but it currently has that you think is that has kind of a big impact or the biggest impact within the Dell AI factory. With Nvidia, I mean AI Enterprise is probably one, but what would be that second one?
- [24:52]
Jeremy Williford
That's a great question. If I think about the software overall architecture and how everything's laid out, one of the big elements of software that's really, that requires a ton of emphasis for enterprise customers is that whole journey of organizing your data structures. So if you think about how enterprises will interact and build their own AI use cases, one of the most important things they have to go do is build, you know, have a strategy for the data structures they want to use for AI. So you need to make sure that you have a full vector database associated with those. Maybe it's cold storage right now, maybe you need to build a data lakehouse for that to be retrieved. And so we think a lot about what we call Nemo Retriever, things that go in and how it thinks about Stack ranking, the stuff that you're going to look at. So instead of doing, you know, right now, if, for example, Logan, if you were going to go look for a file right now, you'd really have to have the name of the file or something that was detailed of that file to be able to pull that out retrieval. Now with AI you should be able to do a semantic search. You should be able to do a contextual, hey, I'm looking for a paper that describes this and it doesn't have to be a word in there, but contextually. You should understand what's in that, that data structure and you should have it vectorized and you should use things from Nvidia AI Enterprise that look like retreat, a Nemo retriever or look like a stack ranking feature. These are all tools that sit inside there and they actually sit on AI data platform. So you build an AI data platform with it with Dell first through an AI lake, hour through a data lakehouse and then attach that to your workstation. So you could start doing iterative, you know, now, now we can get all the data we need. Now let's build a chatbot that interacts with that data in such a way that we could put that as a forward customer facing event. And in that let's make sure that we have a digital avatar, digital twin or digital representation of a human that's talking to that person using that data structure. So those are all Nvidia AI Enterprise software features to go build out the outcome you're looking for. So we think a lot about the outcome and then work our way back on. Here's all the software structures that it takes to accomplish this.
- [27:09]
Logan Lawler
It makes perfect sense for example, I mean, not related to kind of the AI factor. But you mentioned kind of Nemo. Right? It's something that I've used recently. And, you know, part this piece isn't within the AI Enterprise, but I use Nemo to release the Parakeet model, which is really used to take an audio file and then transcribe it with great punctuation, not missing any words. And I use it actually in the context. I'll get the audio file from this podcast and I need to go and review and make sure brands align and all that great stuff. But what I get right now is not super accurate, but installed kind of that model with, with Nema, it's able to give me down to the second accurate transcript of this entire episode, which makes things super easy. And that's one great thing, I think, within AI Enterprise is that, yes, it is kind of the fabric that holds everything together from a software perspective. But there is so much that's included and that you can use, like Nemo Retriever and other things or Omniverse that are all included within. And so it is really designed. Whatever the kind of the use case is or the pain point is, there is usually a software in there from Nvidia that would address it. And that actually kind of comes to my next question, is that, you know, you've been in Nvidia quite a while. What are. You know, you talk to customers all the time. You know, you're out on the road. Even though I'm on the road right now, I think you're back at home. But what are kind of the two biggest pain points that you're hearing from customers this year in, say, 2025, from an AI standpoint? Is it. Are they still confused about, hey, what are the AI use cases I should go tackle? Is it the amount of compute? Is it getting investment? What are kind of the big pain points that you're currently hearing kind of over and over from customers?
- [28:39]
Jeremy Williford
Okay, so I think there's two areas that we're. We're really thinking that could probably need help in the. In the space. One is you have a proliferation of AI activity. So you do have a ton of people. But look, the fact, Logan, that you're able to build your own structure, it tells you that it's pretty. I mean, it's not. It's not. I'm not saying that it's easy because you could do it.
- [28:58]
Logan Lawler
No, I'm an idiot. I'm an absolute idiot. Like, it's easy. It's. It's not hard.
- [29:03]
Jeremy Williford
But My point is, you're exactly right. We, the tool structures are out there to make it achievable for many different practices. And so what we see is, you know, a proliferation of what we call random acts of AI. So in other words, people are users, they're like oh here's my company, here's what I can do and they will put up. I, I think Jeff Clark said it well in his keynote at DTW. He said inside of Dell we probably had 800 random acts of AI, right. Thousand thousand paper cuts. But, but the point was that you have a lot of well intentioned people that really want to do a lot of interesting stuff. However, the enterprise customers themselves, the people, the customers that are really wanting to use this to move their businesses forward, it needs a real focus point to say this is who we are as a company. This is the most valuable piece of data we have. We need to use and have focus on that. And what we're seeing in the market is developing that focus is really a problem set for customers right now because they lack the strategic vision to say this is exactly what we do. We think financial services, they seem like they're a little bit ahead. They have a construct of data that requires a lot of privacy. They, they think about how they do computational work before you know, their competitor can get to it. So, so they are a little bit ahead. But the reality is I think all customers are in this, what's next standpoint. We, we've got a lot of acts here. How do we focus on which ones are going to be done? And then the second biggest thing is it really requires, as we said before, one of the biggest things is a mentality around how you structure your data to get the information you want. I think what I've heard is 80% of the world's data is cold structures right now. In other words, there's great data out there, but how do you bring that data forward in all these different repositories and make sure that you could use it to bring valuable outcomes for your company. And so those are the two areas, focus and a strategy for data. Those are the first two things we gotta go solve and help customers to solve. And Dell's gonna help them. With Dell consultant services, you're gonna have a ton of great global GSI's that will go help like Accenture, Deloitte, EY, all these guys will help customers get there. But that focus and those, you know that, that line and then we'll be able to bring in a lot of great tools to Help them get where they need to go. So I think those are the two big elements right now in the market. But it's clear, Logan, there's either you're either one or two companies, you're either Blockbuster or you're Netflix. So you either embrace this stuff hard and go fast at it, or else, you know, somebody else will do it before you and they will out operationalize you in a microsecond. Because the reality is there's too much technology here that's too valuable and if you're not adopting it in a quick way, you will be left out.
- [31:57]
Logan Lawler
Yeah, I mean, I couldn't agree more. I mean, at the end of the day, AI is here to stay. I mean, it is making life kind of easier. And you're exactly right. And that, and I'll bring it back to a point where I did a talk last year at Adobe Max, and you talk about data, right? And this is a very knocked down example of this, like a very low end example of this. But just to train a Laura to be able to recreate my dog using generative AI, I have thousands of photos. Which one was I actually looking at? I had to resize them all. But just think about that on like a corporate, you know, an enterprise scale, right? That was just my wife loving my dog and taking a bunch of photos and. But the enterprise scale of that, is that to set yourself up? I agree. And that's what I keep hearing, you know, over and over and over. It's really the, the data piece of it because you can't really bring AI without having the data structures there. And you know, that kind of leads next to the next question. Is that the companies that are doing it right, like say the Netflix or you know, the companies are evolving. Can you share? I mean, you don't have to say any names or, you know, give any identifying details, but was there any maybe one or two use cases that you've either seen in practice or have heard about kind of using Dell AI Factory, you know, AI Nvidia technology that just kind of left you like, wow, like I can't believe that I just saw that.
- [33:13]
Jeremy Williford
Well, I will say this, let's break it down to a couple of pieces. One, you know, I think everybody in the world using code generation as one of their tools. And by the way, if you're not Nvidia is 70% software engineers. I mean, you talked about it earlier with your, the comment of a GPU company. The reality is we're a software company and, and we have great architecture hardware architecture to build upon. But our 70% of our company, that is, you know, software architects, all of them are using generative AI and code assist structures to go help with their code development. It's improved our, our own output tremendously. And what I've seen is every company that's adopted, that has scaled quickly with that solution and it's amazing how little infrastructure it takes to go build that. You were talking about pictures of the dog and stuff. You know what all your code base is. And the last thing you probably want to do is put all of your IP and your code structure into a cloud to go get out of it. So, so that's a, it's an easy grab for people in a low hanging environment to go get that. I think everybody needs to be using chatbots internally first. What we've seen, Logan, is in the marketplace. Once any customer has an AI use case, they quickly refine it, they quickly make it better and better and they're like, oh, now we could use it for this, this and this. So it's very much a domino effect. Once you have an instructional set that you're working from and you're like, oh, this is how you do it now let's, let's expand and go hit new products and everything else. I will say this, the most amazing things I'm seeing right now are probably related to the AI work that we're doing for Digital Twin and the stuff we're doing for Omniverse. I mean, customers in their, in their spaces that are able to do robotics training and simulation and do all that work in a virtualized environment that they've created with all of their own assets. That is mind blowing to me. And it is amazing how quickly that market's, you know, really adopting. Dell's a tremendous partner for us both on the workstation side, on the data center side. Jensen says it at every keynote. And I love the way Jensen does this. He tells the entire world our strategy at every keynote. He does. And he does them one, he usually has one every couple months, if not even more frequently than that. And he will build in that robotics, that Omniverse kind of dialogue that demonstrates what it's going to be and where that that field's going. And watching that from simulation to robotics in physical real world situations and what they're doing, the advancements on that are crazy, Logan. So I, I, I'm enjoying that a ton.
- [35:55]
Logan Lawler
Yeah. I mean if you watch every Jitson, Keena, I mean, whether it be GTC or even at ddw, right? Yeah. They're definitely the talk of the generative to the agentic to the physical, like AI. And it is really cool when you see Omniverse, it's kind of hard to wrap your, your head around to think of like how do I make a digital twin of a space and why would I want to do that? But to be able to train in real time in that environment so you don't have robot knocking stuff. Do you know what I mean? Like, it's like there's such an advantage, but until you see it in action, it's kind of, kind of hard to grasp to be honest. But it is, it is really show stopping or being able to create a prototype car in 3D that doesn't exist, right? There's a lot of really, really cool use cases around USD and all of that. But that actually kind of leads me to my next question is that, you know, let's look some forward facing questions and the just two for you because we're getting kind of towards the end of the hour is AI adoption, you know, at least in my humble opinion is accelerated, but it is still new, right? Like we are in the infancy of it. We still got kind of a long way to go. You know, if you think about how things have kind of worked, you know, like for example, with personal, you know, computers, right? Computers have been around a long time, but not just Dell, but others, you know, other OEMs, you know, shrunk a computer where you could put it into something that would fit in your backpack, which is exploded the possibility or the Internet coming into your home, right? What do you think kind of the real seminal event will be for AI that will make adoption like mass very quickly accelerate. What do you think that event will.
- [37:20]
Jeremy Williford
Be the biggest factor for the mass adoption is? Listen, I think what we've done with the first couple stages here, you know, if you think about the stages of AI, you know, we look at recommendation engines, right? And, and we all saw the difference, right? Whenever recommendation engines became good and then all of a sudden you're watching Netflix and it's showing you things that you're really interested in for the first time. Before that it was, it was just random acts of hey, you know, things that they would show to you. But once they started using recommendation engines through a parallel compute structure, all of a sudden it, it became good and you could feel it and you could sense it as a user. You didn't know what happened though, really. I mean, really didn't. And so, and then whenever chat GPT comes out, you're blown away. You're like, it's unbelievable. It could write my papers for me. And but at the same time you're like, oh, but there's hallucinations and there's, you know, and this is not, it's not, it's not appropriate, it shouldn't wrote this, you know, and then you're, you're kind of seeing, they're like, ah, that's not right. But then all of a sudden now we have multi step structures that allow you to put guardrails in that allow you to check it versus solution. Now you're not just using one agent to go off and, you know, build the paper for you. Now you have a multitude of agents that are looking at the output and coming back to you with better and better and better features because it goes through a full stage of agentic AI workflows. And now all of a sudden it becomes, you're not just asking a question about, you know, what the output of this certain requirement is. Now you're saying, build me a business plan for this. And all of a sudden you're, you're looking at a fully thought through process that's taking information from everywhere. And if you look at what corporations will do, they will now have systems that go into, you know, right now if you think about their IT infrastructure, it's really siloed. It's like, okay, this runs my SAP workload, this runs my ERP workload, this runs my HR workload. What we're going to see, Logan, is we're going to see these structures. You're going to have agents that work in every element of all of these databases to be able to work real time with you, to go do things that you would normally it would take you a while or it would take your team a while to come back with. And now these agents that sit in these structures will be smart enough to go work on that in the background while you're doing something and come back to it with a fully represented representation of what's going on. And it will be overnight, Logan. The companies that use this, well, you will feel that you will, it will feel that way in your job every day. And I think that that pivotal moment is just coming into view as we do these reasoning level stacks. So I definitely think that pivot moment is coming because we're now seeing, you know, even our own companies use this. And the closer you get to it, the more you start using it as a partner of what you're doing, then all of a sudden you'll feel the shift. And companies have to create that environment to give that to them. And they have to be mindful of what it takes to create the environment. Yes, you could probably, you know, if you want to save. I don't want to spend that money on a infrastructure on prem. Well then, then you're gonna have to be willing to put all your data somewhere else. You have to be willing to pay somebody to build all the calls and everything else. So that's why we're really thinking about how people can bring, you know, bring the compute to their data and have a sovereign AI structure that helps them to deploy on prem. So we think about those things a lot. But that agentic AI workflow, Logan, that. That really is the next step.
- [41:00]
Logan Lawler
I love your answer and I've never really thought about it like that. You're right. Like in. In terms of a seminal event. Right. I mean you think chatgpt, you think of other things and. But you're right. The recommendation agent, I. Yeah, I mean it is a lot better and I've never thought that it was just gonna kind of happen and you could tell the difference. So you're right. It will. Things will happen in the background for me and sometimes I have to make a lot of decks and the day that I could take thoughts and I get a 99% where I want to go, that my boss loves deck, it's over. That's what I want. That's my next. That. That's my. That's my what?
- [41:32]
Jeremy Williford
And customize it for every. For every person you're about to go talk to.
- [41:36]
Logan Lawler
Exactly. It could be three different decks. It could be one for Charlie, one for riding, one for Kevin. Oh, it'd be beautiful. Instead of three different decks.
- [41:44]
Jeremy Williford
Don't just tell me what my product does. Tell me what their mission is as a customer and how I can fit into their mission and tell me how my products build me a plan that says this is their core competencies, this is where their focus is. Let's talk about how to change those focuses in the AI workflows and what that's going to take. And look, that that's something we could all do through research, but there is so little time in the day to research at that level that you need to be able to do that. So I, I know that's going to be a shift in the market and people are always concerned about. But the reality is companies will spend once they are able to produce more, they will spend more on people to go run this workloads with these, with these Adjective Ki Helpers.
- [42:27]
Logan Lawler
Agreed. So kind of towards the end, got two fun questions for you just based on your time and your tenure experience at Nvidia is one, is that you said you worked in video. I think you said 28 years. Is there one, as you think back over the course of those years, like one pivotable moment that happened and that really changed the direction of Nvidia. And maybe you didn't know it at the time, but when you look back on it, you're like, wow, that I'm so glad that that really happened or I'm glad that we went down that path. Is there one that maybe sticks out to you?
- [42:58]
Jeremy Williford
Let me give you a commentary on that because that is a great question and one that I really appreciate the answer to because at our company, we really, at one point in our company, we decided with our parallel architecture that we wanted to make it programmable and we wanted to give that feature to, you know, the artists, the people that were using it for graphics design. But we also wanted to make it general purpose programmable. And we made that decision, you know, early in the process of our company, before there was really a market available for it. LOGAN and, and what we did, it wasn't a pivotal moment, it was a moment that said that Jensen came through and said, you know what, we're going to make this programmable and we're going to spend a lot of money on it because we have to hire all the architects, we got to abstract the layers in our chip, we got to make sure that people can write to it. We're going to speak to the developers, we're going to make this happen. And look, before the chat GPT moment came around, right? We spent the better part of a decade, over a decade selling this as a, as a feature to developers. Hey, would you like to get more performance than you're getting out of your serial x86 execution pass? And the answer to that was, I'm okay. Seems like a lot of work. I don't want to do it. But it was, it was the, it was an architecture that we were solving customers problems but they were not ready to lift. And we spent a decade trying to sell this. And in that decade, we had, the investor community was just hammering us on it and we had a CEO that was committed to it. Jensen was all in. He was never going to, he was never going to say because he had the vision that said, you know what, I know what Moore's law is. I know whenever they're going to run out of compute cycles I know that the market will always need faster compute, they will need more performance, they will need this stuff. And he stayed committed to it. And the time he delivered that first system, a first eight way server system that Nvidia built because we couldn't get anyone else to build it for us, you know, and that's a, that's a pretty broad statement in the industry that you can't get. We, maybe we had one or two partners that were building it, but not enough. And whenever he built that first system, he gave it to, you know, he actually went to San Francisco, drove it himself in the back of his car to a nonprofit organization and he handed it to him in 2016. And then in 2020, that company, OpenAI launched ChatGPT. And that was the moment. But that moment was built on 15 years of Nvidia and Jensen staying committed to what we're going to build and why we're going to build it, and why the market needed it. And that led us to the largest market cap in the world. And it is phenomenal to look back on and see the resolution he had and the emphasis we put on it as a company for so long and what it, what it turned into.
- [46:01]
Logan Lawler
I mean, I'm speechless. I actually did not even know that story. I mean, I can't believe it. It's much work to do with the. I did not know that story. I mean that, that's amazing. I mean just. And I think that that really sums up Nvidia overall, right? Like just the dedication and the focus to the product. You know, everyone that I've ever met at Nvidia is so dedicated to the mission, their product, you know, helping customers succeed. I mean, that is just fantastic. And I mean, I'll be honest, I don't know about Dell, but you know, I think a lot of companies probably after a while, from investor pressure, probably would have given up. And I'm, I'm very, very glad that Jensen held strong and everyone in Nvidia held strong because we probably wouldn't be on this podcast if he didn't.
- [46:38]
Jeremy Williford
It'd be tough to see. But I got to tell you, you know, working for a founder led company, I get the, I get the privilege of working for two founder led companies. I get to work for Jensen directly, I get to work for Michael in a way that, to help, you know, the joint mission we both have. And I will tell you, it is so fundamental to have great leadership in your company to help you, you know, navigate these waters. And I couldn't have picked two Better people in the world to really, you know, put my life mission into. In workforce. It's such a blessing for me to be in both of those programs. So, Logan, I'm telling you, it is something I think I reflect on a lot, how grateful we are to be in this time and. And to be part of this industrial revolution. It's simply incredible.
- [47:22]
Logan Lawler
It is so at that, Jeremy, it's amazing. I mean, I just looked down and we went almost an hour, and it feels like we've just started talking.
- [47:29]
Jeremy Williford
So I'm sure you can cut them this. I will leave you with. I gotta leave you with one more, though. I'm gonna give you one more funny moment. All right, So I will say this. I am in. It is circa. Circa 2000. We have one of our last corporate events. Jensen asked all of us to bring our families. We're still a small company. We're. We're probably about 500, 600 people at that point. We're still small. We are public, but we're small. He comes in, brings us all in for a big, you know, holiday festival. And he stands up there to give a speech. He tells everybody, families included. Hey, I just want to let everyone know your parents are working, are doing a great job at this company. We're really happy. We are about 30 days from going out of business, though, so I'm really going to need to see some emphasis this next year. And we're all used to it, you know, we're all. Yeah, that's what we do. We live on that. But whenever you have spouses in the room, all of a sudden that becomes a different dialogue. They're like, whoa, what, What?
- [48:24]
Logan Lawler
What?
- [48:24]
Jeremy Williford
What are we doing?
- [48:25]
Logan Lawler
No way. No way, no way.
- [48:28]
Jeremy Williford
It's okay. Hey, look, that guy's always been real, and he loves the community and the family. He always. He's always bringing them together. He used to have, you know, dinners at his house for, you know, guys that didn't have anywhere to go over the holidays. So he is generous to a fault on that deal, but he was also honest to a fault on that deal. And hats off to all the families that stuck around for, you know, past, you know, 25 years.
- [48:50]
Logan Lawler
That's amazing. I can't imagine what my wife would say. That'd be an awkward car ride home for sure. That's so amazing.
- [48:57]
Jeremy Williford
Is your resume fresh? He's like, are we sure?
- [49:00]
Logan Lawler
Are we sure? Like, okay, Adi, we need to talk. That's amazing. So. Well, Jeremy, you've been absolutely incredible. Really enjoyed the time. But before we go, you know, give us any closing thoughts that, that you have, that you want a couple of takeaways, that you want people to walk away listening, you know, to walk away with and we'll, we'll go ahead and wrap it up and get you back to your family. And I don't think Nvidia is going anywhere but. Well, we'll keep them, we'll make sure.
- [49:22]
Jeremy Williford
Well, I, I'll say this Logan. You know, one of the things that I, I, I emphasized a lot during this discussion was how important the software is and, and what companies are doing to own that software. That's really the mission of what we're building with GB10, to allow customers to own their own destiny. I think customers get into structures that are, you know, that long term are, they're tough to be viable right from a standpoint of pricing and how they think of a cost flow. As you develop your AI use cases, as a customer, really think about how you own your own software outcomes. Right. And that's why we're doing the GB10 and the GB300. That's why we're doing full Nvidia, a enterprise stack through Dell. Those are all critical to the mission of building an AI strategy and you know, long term supportable kind of scale out structure. And so that's why I would just encourage all customers to think through that. And that could be, you can still run that stuff in the cloud. Just make sure that you understand the software structure that you're running and that's portable, that you could bring it back on prem, you could scale out to whatever you need to do, but it's your structure as opposed to, you know, being someone else's protocol.
- [50:34]
Logan Lawler
Yeah, exactly. I mean, I think that that is well said. I mean at the end of the day, I mean I'm, I'm very excited about the Delpromax, you know, with GB10 that's coming here shortly and then obviously the follow on is GB300. But yeah, I mean at the end of the day, you know, wherever you start your AI journey, it's important to not go down that journey and then have to start again. So put that thought and kind of focus up front on kind of the architecture, the software, you know, where is your, you know, where is going to the training, inferencing, where's all that workload going to reside? How does it interconnect with each other, you know, how is it powered? Because you can save yourself a lot of issues kind of down the road.
- [51:09]
Jeremy Williford
Well, I'm excited about buying my first Logan app, where I got my Logan Lawler, you know, video editing, slash audio filtering structure that's built into my, you know, my next podcast.
- [51:21]
Logan Lawler
Well, you know what? I might have to come up, hit you some funding for Nvidia, you know, and have you guys funded for me, because it would be amazing. But Jeremy, this was great, man. I really appreciate the time. This was great. I know the audience is going to love every second of this. So with that, you know, this is Logan Lawler with reshaping workflows. Until next time, keep your AI workloads running on Del Pro Max with Nvidia RTX Pro GPUs, and we'll see you on the next one. Do what you want.
- [52:03]
Jeremy Williford
Do what you want. This podcast was produced in partnership with Amaze Media Labs.