Powering Progress: The Economics of The Dell AI Factory

Okay, everybody, thanks for joining. I'm Andrew Jordan and I'm a leader from Dell's AI solutions team. Um, our goal today is to discuss the finding of ESG's economic validation study and show how organizations can accelerate their time to value of their AI solutions and increase the overall success rate of their AI initiatives using the Dell AI factory with NVIDIA. Today, we'll dive deep on a recent study from the enterprise strategy group. They analyze the economic benefits of the Dell AI factory with NVIDIA and study the impact of the solution for organizations that typically face challenges meeting their needs with alternative AI implementations.

Today I'm excited to be joined by two distinguished peers, Nathan McAfee from uh Enterprise Strategy Group and Matt Houseman from Nvidia. Guys, do you want to introduce yourselves and explain a little bit more about your roles? >> Sure. So, I'm a a principal economic analyst. I look at why someone would use a product, why somebody would make a change instead of how that product actually works. Um, in addition to myself, we have a team of analysts and quite a bit of research that I went into to to build a a background and some some findings. >> Really great to be here.

Uh, I'm Matt Hman with Nvidia's enterprise products team. I spend most of my time working with partners and customers on putting together the the full stack AI solution. So that's AI Infra uh AI and um kind of ecosystem software uh services um putting that all together to solve real customer problems and and get outcomes faster. So thanks for having me. Glad to be here. >> Awesome. Well, thank you both for joining me. You know, we've been hearing a lot of key challenges with the organizations we work with when it comes to adopting enterprise AI.

When I speak with my customers, they're worried about data. They're worried about complexity of uh the deployment. They're worried about any risks associated with timelines that they need to hit and mostly they're um concerned about spiraling cost especially when they have a lot of different projects going on at once. Uh Matt, maybe we'll go to you first. What what are you hearing from your customers? >> Yeah, Andrew, I completely agree. I'm seeing a lot of the same things.

Uh when I think about the customers I'm dealing with, actually complexity is really the the biggest thing that comes up. So we've moved past the earlier days of how you know how do I train a model and then you train a model and you run inference and and it's done and you've detected an image you've detected uh you know a voice whatever it might be. Uh but with chat GPT and the advent of foundational availability of foundational models uh the real challenge isn't training anymore.

It's actually about how you implement it across your business how you fine-tune it um and then how you start giving agency to this. So, it's adding a complexity at at other levels that people didn't think about before. And it still goes back to some of the the basic things too. Is my data ready? Uh is my data in the right format? Uh and as I you know, people go into projects, they're now really asking upfront, you know, how much value do I actually get from delivering this?

So, we see a lot of teams that can actually build a prototype, but comes really hard is actually scaling it, putting in production. Uh again, that's where things get interesting or where they get really complex. Yeah, I I think that that's a a great perspective and we're certainly hearing a lot of that too. Um, you know, when I work with customers and we're trying to evaluate uh the cost or the productivity impacts, um, that's probably the biggest reason why we're so excited about this study that ESG did.

Um, Nathan, do you want to walk through the study and the process um, and kind of share some of your findings from the analysis? >> Absolutely. So again our our focus is why a company would change and once they they they did how it impact their business you know their ability to reach it goals but more importantly business goals around AI. In doing that, I started by reviewing everything that I could from uh from Dell and Nvidia to really understand, you know, here's how the solution works here.

Uh, you know, here's how the the two companies work together as uh, you know, as partners to to look and really see what the offering was. Then I went in and started going through our research library to dig into a lot of the the the information around the pain or the challenges because what I do is I really look at pain and pull you know pain or challenges that make me say I need to have a better solution or pull them new capabilities that say hey I need to have a you know a better solution.

Then we interview customers and we we talk to them and find out where were you, what were you using, what were your barriers, what have you found, what surprised you and uh you know that's how we end up coming up with the the findings of the paper. When we look at our research uh some of the things you've already mentioned really jumped out as what are your challenges around AI? As I was interviewing customers, things that really surprised me was just the the the lack of a clear product vision in some cases around this is what we want AI to to to do for us.

How quickly cost could get out of control when they were trying to do things on their own, whether it be a DIY, you know, process or something that they they they patched together. They looked up months later and realized, I'm not sure if we're successful or not. We just don't have the structure. we don't have the the the expertise. You mentioned data and data in so many different ways uh came out in these these customer interviews. You know, our data wasn't clean.

Our data wasn't structured. We didn't know if our data was really, you know, ready or or protected, but uh you know, whether it be to structure the data or the ability to to to protect their intellectual property, that was a huge challenge in AI. So, as we we look through this list, are there any that that jump out at at you two as as surprises or when you talk to your customers, do you see somewhat this saying that we we pulled from a survey of almost 400 respondents to ask them what their challenges were? >> Yeah, I think I'll I'll go in first.

I think that a lot of the customers um that I work with, they have data quality and data aggregation issues. So us bringing in um services and then reference architectures from Nvidia is critical to get things moving in the right direction quickly. Um from there I do see the AI sprawl, right? Customers are trying a bunch of different projects in a bunch of different SAS or API based or public cloud tools and they have these token or GPU hour costs spinning up all over the place.

Um, so us trying to centralize all of those workloads onto one common platform is usually the best way for us to show value to an organization. Matt, what are you seeing? >> Yeah, I agree with you. It's kind of a a blend of what you guys both just mentioned. So when you zoom out, like AI isn't just about one workload anymore, like said it used to be focused on deep learning, machine learning, uh, and needing acceleration for those workloads. But now we look at enterprises, they're trying to accelerate everything. got data warehouses, you've got SQL getting accelerated kind of classical simulations on HPC.

Um, you know, you've got that machine learning, deep learning, but you've also got generative AI, you've got physical [clears throat] AI. We're even seeing customers that are using, you know, GPUs to accelerate their their kind of traditional apps and even even BDI is being accelerated for for most companies. What they're what this really means is they're being now asked to build their own little mini supercomputers that accelerate all of these workloads.

It's a lot of extra stress. That's a lot of extra expertise that's needed. That's bringing new technologies. That's advanced GPUs where they're just CPUs before. Um where people were used to doing just north south networking. Now you're bringing east west networking to get the parallelism between GPUs between nodes. Um there's entire new power and cooling requirements as well as uh you need more power and you might need to move from your air cooled to your liquid cooled or some sort of uh combination.

And then when you layer on top of that the the software, you've got models, you've got frameworks, you've got orchestration layers, and all of that is actually changing incredibly fast uh on on a weekly basis in some cases. And it's just a lot and it's and it's not what most IT teams have traditionally done. And we didn't even talk about data here. Data is still at the center of all this, right? You have to get that organized. You need to get it structured.

It has to be labeled. It's got to be ready for generative AI. It's got to be ready for for use for a lot of different different uh you know techniques. And as we move into Agentic, it's got to be available and ability to to move in and out very quickly with lots of questions coming at it. What I mean by that, when we give the AI agency compared to that singleshot AI I mentioned up front, we're seeing, you know, we talk a lot about a hundred times more pressure on the compute on the storage to respond to those inference queries.

Um, and so that's as you move from a simple chatbot to like a fully generative kind of agentic based system. Um, you've got to be able to scale what you thought you needed for your inference 100x more. Um I was just at the Dell Gartner event uh AI event just uh I guess it was about two months ago now and I was talking with some of the Dell leaders uh and some of their their internal AI systems. They actually saw a 200x increase in the system load as they moved uh to a chatbot to an incentive uh chatbot.

So that's it's just a completely different architecture and it's something again that's not necessarily in everyone's wheelhouse and it's it's a big ask and it's a lot of complexity but we're we're helping people move towards that and be able to accomplish it. >> Yeah, I heard a lot of that from customers when talking to them about their pain. One question that I asked the people that I talked to that really surprised me that I couldn't get an answer was before and after.

So before you went to you know the AI factory with with Vidia before you adopted the solution what was your success rate for AI projects and they would come back and ask me how do you define success I was like well no how do you define success why were you doing this and the lack of expertise and the amount of time between right ideiation and you know impact or value was too along and that's where we saw one of the the the biggest benefits. We'll jump into benefits in a second, but that really that was a huge change in moving to the platform and I think you did a you know a better job than me than summarizing you know just the the difference in trying to develop your AI in this environment versus trying to piece mail it together yourself. >> Yeah, I see this the exact same thing with the customers that I work with.

Right. A lot of times when we're coming to them, they have an idea of what their use case might be um and [clears throat] we try to help them understand the types of data that they're going to bring need to bring into their large language model to accomplish those outcomes. And when we're starting to define a successful project, um I think about three or four things, right? So say they want to build a basic chatbot for a productivity assistant and they want to enable their sales and marketing teams to spend, you know, 50% more time with customers.

Okay, so that's the the goal we're working towards. Um we want to set a timeline on how long it's going to take us to get this project stood up. We want to deliver that project uh within the expected cost budget that we've uh proved to be, you know, within the the ROI constraints that we're working with. And the most important is that whatever data we're bringing into that pipeline, we want to make sure is uh embedded into those large language models in a secure process.

So, we're focused on performance, security, cost, timeline. Um, and if we're hitting all those, that's a successful project to me. Um, Matt, do you guys have a different way of looking at it? >> Yeah, so when I think about what is success look like with an AFACtory, um, it really comes down to speed is probably the number one thing. I know that some other folks touched on that as well. And when I think about speed, I think about day zero, day one, day two.

So day zero you can have a ton of different ideas and really enabling the ability to go from those ideas to day one to prototyping those ideas. So figuring out which ones make sense. You have the data available that you can actually you know deliver on and are going to deliver you value and show that in a prototype uh day one and then day two getting that prototype and scaling in production. So look internally. I know here at NVIDIA, we we we actually just released a case study about a couple of our internal AI solutions, but I think that we had over 700 AI use cases that our teams kind of came together and said, "Hey, this is where we can use AI, you know, you know, we can use and take advantage of AI internally." And the problem, you know, isn't having all these ideas because you can come up with hundreds of them if you sit there and you think about how you can apply this, how you can improve your business, how you can deliver a solution to your customers.

But it was really about figuring out which ones are those 700, those 600, those 700 are the right ones to go after. So to be able to have a system where you can test that quickly, pick out the ones that actually matter, prove the value that they matter, and then scale them. Uh so that's why I say with success with an AI factory, let you lot use and implement lots of different use cases at the same time. quickly see what's work working and use that same infrastructure, use that same solution to then scale up I guess the the winners or the ones uh that make sense.

Uh and so that's not really easy and again that that best practice if you can have an idea on day zero and within a couple weeks you've gotten to day one where you've proven or or disproven that uh and then you know very shortly after that move that in production or go back and and work on another idea. So that's that's kind of what I see success is that AI factory that lets you continue that cycle, that flywheel of going from idea to production and and proving things out really quickly. >> Yeah, Matt, I couldn't agree more.

And when we're working with customers, we're deploying a reference architecture on the Dell infrastructure side that's directly from NVIDIA, but we're also building our AI factories with the NVIDIA software. And I'm usually directing my customers to go to that build.envidia.com site to take advantage of the blueprints and the NIMS that you guys publish and that your user communities um take advantage of because whether they want to build a digital twin or they want to do some sort of warehousing simulation or they want to do genomic studying, there's already pre-built projects published so that they can really accelerate that time to value and get going uh much more quickly.

So, um, that's that's kind of always my my secret recommendation that that helps projects move more quickly. Um, Nathan, I know that there was a bunch of other benefits that you found in the study. Do you want to talk about what some of the other factors were that uh went into customers evaluating the solution? >> Yeah, absolutely. And when we examined customers of all sizes from from true enterprise to, you know, low mid-market, we talked to people that had very mature AI programs versus some that are were just getting started.

So we we were able to get a a a good blend of, you know, data to to put together our story. And across all those, one of the biggest we'll talk about is acceleration and the the the value of that. Then we look at cost efficiency. And what that really was was the ability to point to business impact for dollars spent. I heard so many stories of we had a huge AI budget. We started a project. We're nine months into it. We have no idea what we're doing, where the money went, where we're going to go.

And again, it is, you know, what what defines success. So I'll talk a bit about that and the improved security and compliance. And a lot of those stories, you know, range from we weren't doing AI because we didn't have the expertise to go back to our leaders and say our data is protected. And I I have a few stories that I pulled and here's one where I think you guys will probably be able to tell more stories than than I got. I got very anecdotal, but I definitely want to to you know cover a few things there.

But if we we dive into the first and really look at acceleration when I was reviewing you know what what defines success for AI. I saw before they went to the you know before they they they went to the solution that companies were nine months into a project and just not having an impact in their workflows. And that time when they they they they went to Dell Factory with Nvidia, you know, that that time was down to months or weeks just how much faster they went from idea to business impact and to to figure out what is the what what's the the benefit of that.

The hard part as an economic person is I want to build a calculation that you can put anything in and get the you know the same type of number out. I had one specific story and it was somebody said, "Hey, when we look at our sales teams, what we did, we were able to impact about 20% of our our our sales cycles, but that 20% we were able to increase our close rate 6%." So, you put all those numbers together and they had a multi-million dollar impact out of just one portion of their business.

And the reason they were able to do that is because they could get information of we need this type of capability. We need this type of change in our workflow or you know viewer or usage of data. And they were able to to turn that around much quicker with the Dell and Nvidia solution than they they were in the past. When we look at productivity gains, the the the the the easy numbers were the ones around how much more effective their people were that were trying to implement the AI solutions, but then I had examples going from, you know, this impacted every one of my employees and gave them a certain, you know, a certain number of of minutes and in one case almost an hour and a half per employee per day.

So while the examples range across the board, every person that I talked to told me how much faster they were, you know, in in trying to get a solution to actually, you know, to the the point of impact, how much more both their IT and their business people, you know, were able to achieve productivity gains, how much happier their people were working in this environment, and just overall operational efficiency. So I I'm rambling a bit, but that quote on the the the right really, you know, boils down a story.

I heard this from a few different people is we now have in this case a one year, we now have months or years of head start against our competitors specifically because we moved to this platform. >> Yeah, Nathan, those are totally great points, you know, and we see that acceleration firsthand. When we come to customers, they're looking for an outcome, right? They're not looking just for a GPU server. They want a comprehensive solution. Um, and so we get into the process with them uh designing around the NVIDIA stack to suit that solution or that set of outcomes.

And it's very important that we have multiple configurations and sizes to fit those types of workloads. Right? So whether it's a telco that's automating their entire customer support functionality or it's a manufacturer that's trying to improve quality assurance on the production line with an edge solution. We have the right sized AI factory for those needs. So Matt, with those different sizes of solutions for specific workloads, uh what are you seeing as an important part of us bringing the right technology to customers to drive these outcomes?

Yeah, this is really where the Dell and Nvidia partnership comes in uh and why it really matters. You know, Nvidia is really good at making things run fast. You know, we can give you an order of magnitude or orders of magnitude improvement in your compute with GPUs. We've got networking, we've got DPUs, we've got our Envy link uh to give you that worldclass networking to allow GPUs and nodes to talk to each other. We've optimized a whole bunch of open source models.

We've delivering our own uh foundational models and then we've we put them into our NIMS, our inference microservices, allow them to run three to five times more efficiently. So we're basically optimizing all the different pieces, the compute, the networking, the models, the the inference. But just because we make it run fast doesn't mean it's easy to implement. Again, that's where our partnership together, I think it's based on over 400,000 uh joint engineering hours over the last couple of dec decades, has brought us together with the Dell AI factory with Nvidia.

And so we're basically delivering full stack AI solutions, the infra, the software, um the the services, uh with the target of delivering, you know, outcomes, uh be it your chat bots, uh code assistants, um could just be the platform, the kind of the AI platform you want to take advantage of or the data piece, right? And as we look at that and we kind of looked at some initial customers leveraging the AI factory together versus going at it your own, this actually helps people deliver uh an AI solution about seven times faster.

So I've mentioned speed multiple times today. Uh but being able to build that supercomputer and not have to have the knowhow of all the different components and and figure it out on on your own. Uh we can help you get there about seven times faster >> and thousands of mutual customers. Right. And and that's a good point. While we're here, let's pull up that graphic of the Delli factory with Nvidia. Well, guys, this is a great representation to our joint approach to building solutions for customers.

Right on the right hand side, you can see use cases or specific outcomes that we're looking to drive, whether it's chatbot or a code assist or a digital assistant. Um, the other most important book end of this is your organization's data, right? Because with we're using a cloud or an off-the-shelf tool, we're going to have to bring in specific chunks of data into our AI workflows to drive uh the ultimately most accurate and performant outcomes that we're looking for.

Um so this is a full stack or a turnkey solution. Um at the bottom layer, it's all Dell infrastructure. So it could be a Dell server powered by Nvidia GPUs. Um, we are usually also prescribing the NVIDIA reference architecture which includes their robust suite of networking capabilities as well as Dell storage and cyber resilience capabilities. Now going beyond the infrastructure layer into the software stack is the most exciting part, right? because we're usually building uh factories not only on the NVIDIA software which is commonly referred to as NVAIE or NVIDIA enterprise but we're usually bringing in one or more thirdparty ISVS that help deliver a specific outcome uh or a specific use case for an industry and then on top of that we've built a very robust services practices all the way from deployment and implementation to strategic consulting and the ongoing development of those use cases.

Uh Matt, what do you see here as some of the differentiators of this full stack solution? >> Yeah, I mean I I see there's a number of differentiators here with our solution. I mean, I'm I'm going to double down here on the software piece. I I think pairing the right software with the services and the infrastructure below it uh is really critical uh to help you actually build, refine, iterate on your AI solutions, actually get value really, really quickly. our AI enterprise software for generative AI, Omniverse for physical AI and digital twins, right?

That's a really good starting point to get you out and moving. But these ISV partners, um, other software partners in and market, they really fill in the rest of the stack to give you a complete solution, maybe a unique solution to to what you're looking to do. Um, and I look just at Dell and kind of their offering and portfolio beyond the services capability I talked about earlier. Um we also need to think about the fact that they have both workstation and data center offering.

So AI is often developed like I mentioned before on the desktop in the data center and the cloud but having a single um you know partner like Dell that you know can give you a consistent solution for your developers that can prototype on their black wall powered desktops you can move that same workload over the data center and they can use the same software stock uh stack across that layer as well. So that flexibility is really powerful and really does differentiate uh Dell because they can let you and help you develop where it makes sense and then scale when you're ready uh and then keep everything consistent to minimize the changes between those different locations.

So when you put it all together, Dell services, Nvidia, you know, accelerated compute and software plus the ecosystem and this endtoend solution, it really gives companies a strong way uh to build and run their AI, you know, wherever and whenever they need it. >> Um, now Nathan, I know that uh, you know, this is kind of all speaking to that time to value benefit and how we can get going quickly because we're following the Dell and Nvidia playbook.

Um, but what were some of the other benefits that you found throughout the the analysis? >> Well, one thing to add to what you you two were just talking about, not just the engineering partnership. I heard from uh customers that we talked to how important the executive commitment from both of your your companies coming together that gave them the confidence to go to their leaders and say yes, we can trust this solution. So, it it was everything from the shared roadmap.

So the the the engineering I mean that that story comes through in the you know the the results and the performance but the executive commitment was really called out by by customers. So I want to make sure that was mentioned. >> Yeah I I couldn't agree more. Um and certainly that's something that we're well aware of and and we're bringing that message to to all of our customers. Um now I think you had mentioned cost efficiency. It looks like you've uh you beat me to the punch there.

Let's go into uh what the results were here on this front. >> Well, the results here were ultraconservative. One thing that was a challenge with this, you know, project where so many of the stories broke my model. You know, I've heard you say 100, 200, multiple hundred times the the performance. But what I wasn't able to to to to model was the assuredness that I'm going to end up with a success, you know, successful project that what my capabilities were in my workflows were what I intended.

So I took the ultraconervative across all the the the different stories and with that we we've talked about you know time to value 153%. Uh, you know, you one of you mentioned 7x. I heard it took us nine months to do nothing and we as soon as we we went to the Dell AI factory with Nvidia, we finished it in six weeks. So the these numbers are great, but these numbers are also ultra con, you know, ultra conservative. >> So Nathan, are you saying that every DAI factory with Nvidia has lower costs than alternatives?

What I heard was it was more trackable, more predictable. It was easier to say for every dollar that we spent, we can see the impact that for the the the the money that we're putting in, we have a higher level of assuredness that we're going to get something out of it. Now, I did not look at all the every different AI solution available and say this this is the the cheapest. Not a single person said I went to this because it was cheap. They told me I went into this because we need success in AI.

We were challenged in doing it on our own and now we have the platform, the expertise, the scalability and the ability to still to understand if we put this much investment in, we have a very high likelihood of being successful. >> Yeah. I mean, as I look over these numbers, I mean, this is this is really a, you know, a great point we're at with AI. I think we're at this inflection point just like with cloud and big data before it uh where there's a technology there's a trend people get excited about it there's some disillusionment and then people actually start getting real value from it um you one of the big things with the AI factories though is a little different on that value generation is is they really take that concept of it of infrastructure and kind of move it from a cost center concept into a value generator and so that's where we start talking about things in terms terms of tokens and we talk about the the output and the inference in terms of tokens is that every token has some sort of value that it's outputting you it's answering some question it's doing some sort of AI inference and there's a value associated with it so when you invest in the infra when you invest in the software when you plug power into it that's letting you now generate these tokens and again those tokens actually have business value uh through answers through product productivities through capabilities So, it's a new model that we're at and so these 10x these 12 uh x numbers of ROI make a ton of sense and I think we're going to start seeing even bigger numbers.

So, as we can optimize the power of the GPU software to bring that cost down, allow them to actually output even more tokens based on your initial investment and kind of ongoing power investment. Um I said we're kind of early days that 10 12x uh you know output or 10 12x uh you know value creation is is great and it's huge and we're you know those are numbers that we can stand by and we're just going to only see those continue to grow. >> Yeah Matt, I couldn't agree more.

Right. And these results are while they are early days they are uh emblematic of a great partnership right between Dell and Nvidia. We've been doing this for over 40 years and uh I think that it's all about enabling our customers with the right technology at the right time and place uh to enable some of these incredible outcomes. Um we talked a little bit earlier about blueprints and taking advantage of all of the um publish repositories that Nvidia has already brought to market.

Um do you want to talk a little bit more about that and how customers are leveraging them? Yeah, I hate to say it, but yeah, I've spent decades now in the data analytics and AI space, and I see the same pattern over and over again. If you just buy the infra, what you're getting is a blank page and actually still takes you a really long time to get value. And so, I just don't sleep on the software component. It needs to be a part of every one of these solutions.

Uh, as you build your factory, you need the tools, the components, the capabilities to actually take advantage and implement something. Uh that's why I really like a couple of offerings from NVIDIA. Number one, our NVIDIA AI enterprise software. This is really helping people with generative and agent AI uh use cases. Then our omniverse software and the physical AI uh doing digital twins. And again, this is really providing the the components so that day one customers can actually do something.

It's not just the infra plumbing. It has models. It has uh workflows. It has data. We have blueprints that take all those together to help you actually build, you know, an endto-end solution from day one. Of course, you're going to take that, you're going to customize it. You might add some other components to it, but it lets you start playing. It lets you start interacting. It lets you start building your AI as soon as you deliver the solution. Um, and this isn't just Nvidia software and and Dell components along with it.

Uh, as you look at this, this is a this is a broader ecosystem. So this is our software plus ISVS plus open-source uh projects plus the orchestration platforms plus our applica you know application uh frameworks that are available out in market and we've actually done a lot of work uh I think we're up to about 30 different software partners um that we've integrated into what we call our validated designs and our reference designs and this shows how we can take this partner software and plug it in with the Nvidia software and this full stack solution in our AI factories with Dell uh so customers can build and deploy with with confidence.

So you'll be seeing in these validated designs and reference designs kind of extended uh software partners uh you know coming into the Dell solutions and their automation platform over the next couple of quarters just making it easier to build and develop these full solutions and kind of give you that click button uh kind of flexibility to pick the right components for your solution. >> Yeah, 100%. And I think you know most of the customer deployments I work on we are bringing in uh one or more of our software ecosystem partners to drive those specific outcomes.

So uh I know that between Dell and Nvidia we're really appreciative of that ever growing ecosystem. >> Yeah Andrew I I completely agree especially on the software side this pace of change is just incredible. Uh, I know Nvidia, we track about the top 2,000 models on a weekly basis and we take about 10% of those and consistently the top 10% of those and just consistently optimize those to make sure it works with our full infrastructure stack. There's a version of those uh as as upto-date as we can uh that's going to perform really well.

But no single company even a company as big as Nvidia can do this alone. And again, this is where that ecosystem of uh of ISV and other software partners and our validated and reference uh architectures uh that that I mentioned before, this really matters because these specialized partners, they're going to bring the best tools for some specific use cases. They're going to bring optimizations uh that that are unique to them uh and may really matter for your specific use case.

And again, this is where this all comes together, where Dell takes all these components into the AI factory with Nvidia and makes it all work together and delivers this to you. And with those constant software updates, it also helps keep your AI factory evergreen. Uh, and so you're current with the latest trends, with the latest techniques, uh, with the latest models. >> 100%. Yeah. And it it really plays into uh both the time to value and the cost savings when we can bring that seamless integration into uh a customer's environment.

So um Nathan, thanks for helping us understand this cost study. I think you know from this chart and the results that we covered the advantages are totally clear and we're excited to see these results uh which are very likely to continue getting you know even better and better as time and adoption goes on. Um earlier you were talking about um a third category maybe security and compliance and what are some of the benefits that you found uh under those aspects?

Th this was a a hard one to quantify because I heard stories of one week into AI we realized our our intellectual property our IP had been exposed. You know people didn't have the the expertise to use. But the biggest thing that I heard was almost AI enablement before they went to the platform. They could not get their business leaders. they could got not get their legal to approve the project because they just didn't have the certainty and control.

So it uh I I can go through some some specific examples but most of them are the the type of we got in without the expertise. We try to do it ourselves and we have a whole lot of I don't know. I don't know what our level of exposure is. I don't know. And with the platform they this quote really says it. We we we own the risk, but now we own the control. So, is there is there something that that you could too can can add in specifically about how you're able to help people get the level of confidence and assurance that they can, you know, they they they can own their security compliance. >> Yeah.

So, one of the big things here is really just uh keeping your sensitive data inside your own trusted boundaries. So when you do that, you control the access, the policies, the models. You you control all of it. You know who's touching it. You know how it's being used. And at the same time, people aren't going to work just in one place. They're going to have some development happening on desktop and the data center, which we're talking a lot about uh today with the factory and out in the cloud.

That's just reality. It's going to be happening everywhere. There's always going to be these side projects or skunk works efforts going on, too. And and that's okay. So what really matters is being able to bring all that together so you have visibility. So you can see what data is being used, which models are being run, um how things are behaving, what what risk uh tolerance and and things you're opening yourself up to. So when you have that kind of observability, compliance gets easier, uh governance gets stronger and team actually becomes more more efficient.

So for me, it's not about saying everything must be on prem or everything must be in the cloud. I guess putting the right controls around it and and the right capabilities in each one of those places uh and and visibility to have that view. Andrew, I'm I'm curious to see how you see that. >> Yeah, Matt, thanks for uh for teeing that up. You know, when I think about security and compliance, I think about uh two really important facets. One is process and the flow of how things happen.

And the second is the tools and the capabilities that we bring to customers together. Um, so from a process perspective, we're deploying a zerorust architecture for an on-prem AI environment, right? So when you pull down a large language model from NVIDIA or from hugging phase if you want to use an openw weight model and you're bringing in synthetic data, once that's deployed onto your DELAI factory with Nvidia, it stays on that local system in perpetuity.

It doesn't go back to the cloud. it doesn't go back to any other API type tools, right? And so all of the organization data that you then embed in that model stays inside of that environment in a very secure uh typed uh repository. And then Dell has a robust uh suite of cyber resilience tools that are designed to res reduce the overall area uh of vulnerability. um they give you the capabilities to identify and monitor any potential attack surfaces and then ultimately respond to those um those cyber threats in real time to stop any sort of concerns.

So um I really think that you know by partnering with Nvidia as well as our cyber resilience tools we give customers uh an utmost advantage to address those concerns compared to alternatives. Um, so Nathan, now that we've covered um, you know, security and compliance, why don't you finish off with some of the the big picture findings from the summarization of your work? >> All right. So, we we have the details in the paper. You can go in and look, but as an an economic analyst, I've got to make sure everything that I do is conservative, is achievable, is defendable.

And that's why would the the numbers I have here, I think, can be exponentially higher. The best thing the best way for me to summarize this is uh you know somebody told me this has immediate payback and exponential value. So much of it depends on how you can impact your you know your your business and your workflows with with AI. But every person that I talked to, every every scenario that that I studied showed immediate if not very very very quick pay back on any investment and enablement.

So that that last that that last quote saying now I'm able to do stuff I've been waiting to do because I have this level of assuredness. So it uh we don't need to go through the the details each number. There's uh you can tell them in a second how to to get the paper and they can go through and look through the different details, but I want to make sure that was called out. >> Awesome. Matt, do you have anything you want to add to that before we get things wrapped up here? >> I mean, so the thing that really jumps out to me, really impresses me most about our daily factory is when I look at these numbers is um the immediate payback.

I mean, the time matters so much and and the value creation matters so much. So the ability for you to invest in AI and then actually do something with it right away and get value. I mean that that's huge. That's part of kind of overcoming that pro process as as AI matures. Um also as I look at this chart, you know, we're really looking at the year one here, but if you finance this differently, you can actually get even better numbers that we're seeing here.

So when you put the infra, the software, and the services all in place up front, right? You're seeing this value almost immediately. And then year 2, three, four, five, uh, it's even going to get better. Uh, so that's what stands out to me. Not just this ROI on on paper, uh, that you can get from from this and the tokens that you're outputting, but about how fast you can start turning an AI investment into something real >> and track your investment and match to I spent this much, now I can do this.

I heard so many stories of we poured money into a the a pit, we're not quite sure where it went. So that that that's a story I'll be sure called out. Yeah. So, I mean, the work we've done together on the engineering side, I mean, it's it's all about letting you start somewhere and grow. So, there's no oneizefits-all, but what you don't want to do is be be buying, you know, a single cluster or a single node that's stuck on a corner, and it's not something that you can build on or build off of.

Um, so you want to be able to invest in your your AI solutions, add capacity, add workloads, and scale. I mentioned, you know, maybe scale 100x more than than you initially thought. So that ability to start small, expand uh is a really big part of delivering the right solution. A really big part of the Delli factory with Nvidia and all the pieces we putting together uh we put together, it really helps us differentiate. Um so when you look across the whole stack, uh services is actually a really big piece which we haven't talked about too much today.

I mean there just aren't that many companies that do the infra that do the software and then also have the services expertise like Dell to actually help you design to build define to deploy uh your solutions. So I mean honestly for your first couple of AI projects if this isn't in your wheelhouse working with a partner makes a huge difference right it can help you pick those right use cases and go from that day zero to day one really quickly. uh you can help get those first deployments up and running really avoid I guess a lot of pain a lot of struggle up front because this stuff is not easy >> totally yeah I couldn't agree more guys and you know thanks so much for a robust conversation today um you know clearly organizations are seeing the value from the deli factory with Nvidia uh from performance security cost benefits especially that time to value um I'm sure that you guys watching out today will have more questions.

Uh so please get in touch with us. Um we do have uh quite a few ways to get started whether it is a advisory workshop where we would bring in uh members from the deli team and Nvidia to help figure out those use cases. Um of course we can facilitate uh pilots to get going on quick uh to get going quickly for things that you know you're looking to do. um as well as executive level, you know, EBC so that we can make sure that your leadership is uh in tune with the types of capabilities that we can bring your organizations.

Um and if you're looking to learn more, feel free to go check out dell.cominvidia-ai to learn more about our joint solution. Thanks so much.

This transcript was generated automatically from the video's captions and may contain errors.