Dell Technologies Inc. ($DELL)

Please welcome Michael Dell.

Good morning, everyone, and welcome to Dell Technologies World. Thank you for joining this extraordinary community of innovators, dreamers, builders and leaders. Together, we are driving one of the most important transformations in history. Abundant intelligence is here. It's not coming. It's here right now. Intelligence is becoming infrastructure. And just as electricity transformed the world when it left the power plant, AI will transform the world when it leaves the screen. We are making intelligence real. local, secure and useful on the oil rig, in the ambulance and on the factory floor. Dell Technologies is building the distributed infrastructure that turns isolated insights into intelligence in action. And we are opening the door to a new era of progress across every field of human endeavor. The models are smarter than all of us, and they're improving exponentially. Since ChatGPT was launched in 2022, we've gone from a chatbot that could write a decent SA to self-improving AI agents that write code. They run workflows and they operate around the clock. Coding is now AI native. You describe it, verify it and ship it. And the barrier between imagination and execution is collapsing, and this is unlocking human creativity on a scale we have never seen before. For organizations, AI is no longer a feature. It's becoming the operating model for the modern enterprise. And that is where we are going to spend our time together this morning on the journey through the modern enterprise. All right. So here we are in Austin, Texas in our boardroom, where modern enterprises begin with a conversation and a commitment and a vision for what comes next. It's where we learn about each of you about your journey, your opportunities from health care to manufacturing, to finance, to agriculture, from small businesses to large enterprises and governments around the world. This is buying all-time favorite visual. I use it every year, and I'm just wondering, can we make it just a little bit bigger? There we go. That's this is what I am most proud of. It's the trust that you place in all of us that we take so seriously and the relationships that we've been able to build and the work that we're able to do together. It's incredibly meaningful. This is where trust begins and where our partnerships start to reinvent industries. AI becomes real when it shortens the distance between discovery and impact. And when you're at the frontier of modern pharma, the stakes could not be higher. So now let's look at Eli Lilly, where science becomes life-saving medicine. Please welcome Diego Rauf. Diego, great to have you here. Thanks for being here. And look, I mean, Lilly has an incredible history and story of remarkable scale and impact in the world in medicine over many generations. And you've been at the center of sort of bringing this super important company into the modern age with the latest technology. Tell us a little bit about how technology is shaping Eli Lilly today and what that's look like.

Well, fantastic. And what a journey it has been because Lilly actually just celebrated a birthday party on Thursday last week. We turned -- thank you. really don't feel like a day over 140 years old. But Lilly is not a tech company at our core. We were started by a kernel from the civil war in the United States. -- who saw people needlessly dying on the battlefield, not getting medicines that they need. And we're actually getting sham medicines, and he had this crazy idea that why don't we actually go use the scientific method figure out what medicines actually work and then figure out how to make them at large volumes. So from the beginning, really was always a company about scale. So in 1923, when insulin was discovered, we had a breakthrough medicine that could save so many lives, but nobody could figure out how to make it at volume as where Lilly came in or actually penicillin in 1928, as when it was discovered. We needed it for World War II when people were dying on the battlefield. U.S. had ramped up supply by 1942, 14 years after penicillin was discovered, Guess how many patients the United States could treat with its supply.

I don't know.

10, Not 10,000. That doesn't sound in No, it doesn't. So the United States government. It's definitely not good. The United States government called in Lilly and said, "Hey, can you do that thing you did with insulin? Can you do that again with penicillin -- and that same thing happened again with the Sapolio vaccine. It happened again with the first COVID antibody. It's happening again today with GLP-1s because medicine isn't just about discovering something new -- it only works if you get it out in mass numbers to help the people that they need. The only way you ever get scale in medicine is to actually have technology underneath it all. And so that's why we have always been huge technology adopters at Lilly. -- early in bid. We were 1 of the first to adopt the IBM mainframe back when it took up a space probably like this size to run. And -- we were 1 of the first adopters -- actually, we're the first commercial adopter of the CRA 2 supercomputer. And this year, in February, we introduced when online with a pharmaceutical paraceutical industry's largest supercomputer -- we've always gone early, and we've always gone big because technology is how you get that scale for medicines.

And give us a sense for what the impact is how AI is making a difference today in Lilly?

Sure. Well, it makes a difference across the board, of course, because I think if we took it away from people, they would quit right now, right? Nobody wants to go back to the pre-AI days. But one of the exciting places, I think, is in manufacturing we probably don't talk enough about what happens in the world of manufacturing. -- to be able to scale up all these medicines -- exactly to get that scale of medicines again. And one of the things that's really cool that I like is when the medicines come down the filling line, they're like a social media influencer getting glamor shops the whole way from all kinds of different angles. It's not like 70 different photographs with budgets of milliseconds to detect -- it's something that's super cool or actually the optical inspection mature quality because we have humans do that and that's what humans mess up. Like these machines, they don't. And so it's amazing what you can do. Digital twins, I mean, how long have we been talking about digital twins, that's been like forever. Now we are actually really doing it. And we've taken processes that humans have said are completely fully optimized. And what we've done is we've replicated the machines that are in there as digital twins move them around. And guess what the AI the humans. It's not completely optimized. So manufacturing is just a fascinating world. But the thing about it is, we're so dependent on technology now that it just can't go down. That's the difference about this technology revolution from anything that we've ever seen before. And that's why we've been very happy about our partnership with Dell because -- when we open a new site, we don't have to go out and figure out what we need to do. We've already got a plan together. We already have the architecture. We already have the stack, everything has been validated, and it just works. It's coffee exact. Exactly.

So that's the operations side. What about drug discovery and research and accelerating really the plot Eli Lilly at its core.

I love the challenges in research and discovery because that's another kind of scale challenge just like we have never seen before. I mean I think -- when we think of labs and if you're not in this world, you picture people running around in white cos with notebooks, taking notes and maybe looking for a microscope. Now today, a modern lab, it's all technology in there. And every single 1 of those things is connected. -- scientists aren't looking at some fuzzy image and trying to make something out. They've got 4K resolution, 120 frames per second of whatever they're taking. Everything in that lab is just showing us with data, petabytes of data as it goes. -- and it's fascinating, it's overwhelming, but at the same time, that's the data that we need to use to power AI. That's the whole reason we have such high-performance compute needs.

And what happens in a couple of years when you get orders of magnitude, more computing power to the research and drug discovery at EI Lilly. I mean how do you think about the effect of that on cancer and other sort of challenges that exist in the world.

I love that because we're today -- we're already there. We're solving problems that like -- that we weren't thinking about 10 years ago. The problem was of 10 years ago are just frankly boring today. So everybody knows about alcohol, right? Super cool. You figure out genes and you figure out what protein would come out and what cheap it that makes Well, you know what, that's boring, Nobody really cares about that now. It's like, who cares what the protein looks like, what is it going to do for me? That's what you need to know, like, okay, so you've got a protein now. How is it going to interact with another protein. They're all moving rapidly in time and space, orbiting each other, bumping into other molecules. -- that's how we're actually modeling things that are much more advanced or we're -- and by the way, that's actually factor cancer and things like that. That's actually how you build something that's so selective. It can only attach to one class of cell. And if you can do that, you can drop off a payload, chemical or nuclear actually or other things like that to kill a very specific cell. That's the kind of stuff we never could do before. It's just an absolute breakthrough of what we have in technology now.

So another example where we live in the most incredible times and RNA vaccines, gene editing, all these things are incredible. So tell us a little bit about how Dell is supporting all these efforts and sort of what it means to be able to have the technology to power this.

Well, it all the scale, everything I've been talking about only happens with technology and all that scale is actually what we are getting from Dell. We are still proud of our AI supercomputer that we have or...

1,000 GPUs.

Now 2016. The newest generation. By the way, if you haven't seen 1 of these things, it's crazy there. like it's 105, 110 decibles. I clocked the wind speed coming off the fans at 30 miles an hour. -- like you're...

We're trying to make them quieter.

They're amazing. But the Dell infrastructure underneath. -- is actually what's powering all of that. But I want to make it -- I want to make it clear because it's not actually about the hardware. What we did first is we tried this out in a couple of smaller versions. And Dell was there with us as we were learning. We weren't just learning about how to set up the hardware. Dell was actually helping us with the whole management process. And what we really -- what we really benefited from was not the models that we trained but actually the training we got from you. And so I really would love to say thank you to you and to all of the Dell team members because that's actually what helped us get the skills that we need to be able to build the supercomputer that we have today. And that's how we're going to make the kinds of breakthrough discoveries that we've been talking about. Because you know what, in 150 years from now, we're going to have new things that we can't -- we're going to have breakthroughs that we can't possibly imagine. I think we're on the verge of maybe being able to end disease as we know it. I mean something like that was completely unimaginable 20 years ago. But today, we can imagine it. And I think it's what you said at the beginning, -- now the challenge is how do we shorten that distance between the discovery of the imagination and the execution.

Diego, amazing. Thank you for a life-changing partnership and capabilities that we're really privileged to be a part of or inspired by everything that you and Eli Lilly are doing with technology. Thank you so much for being here. Wow, more from health care soon. So thank you, Diego. It's -- what you just saw is not a chatbot. It is intelligence in the physical world. It's a faster path from science to medicine. And now every organization faces the same challenge. How do you turn intelligence into impact at speed. And I hear it from the CIOs, the CTOs, engineers, operators, CEOs and boards. It's coming from the top down. I can tell you it certainly is a Dell. I see Jeff Clark. That's because we understand how AI native companies work, how they operate, what they look like and what they're truly capable of. This is an existential moment -- the companies that redesigned their work around AI are going to compound advantages faster than at any time in history. And every leader in this room feels that pressure right now. It's exciting. It's exhilarating. It's sometimes exhausting, but it's also unavoidable. And I want you to know that you're not alone. We are by your side with the expertise, the global supply chain and the execution to deliver when it matters. The supply chain is modernizing too. And some of our most important partners and suppliers also are our most important customers. For our next modern enterprise, let's look at the intelligence behind something we all rely on pretty much every single day, modern semiconductor manufacturing. [Presentation]

In companies like Samsung and across every industry, AI is accelerating from proof-of-concept into production. And it's flipping the traditional buy versus build equation. For years, the trend was off-the-shelf software and public cloud. But now AI is collapsing the cost and time to express your competitive advantage through software and so guess what, there's going to be way more software everywhere. A recent Dell technology survey on AI adoption shows that 67% of AI workloads already run outside the cloud either on-prem, on device at the edge or a colo and 88% of respondents are running at least one AI workload on-prem. The CIOs are aggressively pivoting to hybrid AI. The risk is not the cloud. The risk is losing control of your data, your cost, your security, your intellectual property and your speed. In the agent era, walk-in does more than slow innovation, it actually limits what your company can become. As soon, every company will deploy fleets of agents, composing workflows on infrastructure that they control. It's estimated that the worldwide AI infrastructure spending could hit $3 trillion to $4 trillion by 2030. AI is fueling a renaissance in enterprise hardware, a shift from bits back to Adams. The question is, how do you deploy the world's best models, where you need them with security and governance built in. We give you model choice without infrastructure chaos, you get the open models, the frontier models, the specialized models running where your data lives, available starting today in the Dell Enterprise hub on hugging face representing major architectural leaps across the AI ecosystem, new models are available, including Minimax M2.7, DeepSeek Pro, Deep Seek V4, GLM 5.1 and Kimic2.6. These joined the GemA4 family of models, the NVIDIA Nemotron, SuperT-series, the RC Trinity large thinking and mistrial small. In addition, ServiceNow customers will be able to leverage the Dell AI factory to operationalize, their AI-focused business outcomes. And we have some very big news, you'll be able to bring the latest frontier models to your enterprise data on the Dell AI factory. Google and Dell are bringing the Google distributed cloud with the Gemini free models on-prem with the PowerEdge XE-9780 servers. You can run advanced AI workloads in a confidential computing environment using the new Gemini models. And open AI and Dell Technologies are building a solution based on open AIs coatings. -- to bring their latest agentic harness on-prem with the context provided by our Dell AI data platform using a set of connectors and skills. You can securely connect to the latest GPT 5.5 and GPT Codex models. And SpaceXAI-adell Technologies are delivering rock advanced reasoning and multimodal capabilities at secure, enterprise-grade AI systems, deployable on-prem or in a hybrid approach. Palantir's foundry AI platform are also coming on-prem with ontology deployed on our object scale and PowerFlex platforms, you can connect all your data sources and optimize operations with the full weight of AI connected within the boundaries of your own environment. And reflections open source frontier AI models are now coming to the Dell AI factory. Regulated industries, including governments and sovereign entities to deploy AI in a fully controlled environment. Look, every nation is waking up to a new reality. AI capacity is a strategic asset like energy, communications and secure supply chains. And we're delivering the secure air gap nationscale AI infrastructure. without lock-in and without worsening your data into somebody else's black box. AI can become the most concentrating technology in history or it could become the most democratizing technology in history. We are choosing democratizing. Powerful AI wherever it's needed, open, secure and yours. This is hybrid AI. It's not a compromise. It's a competitive advantage running at your Dell AI factory. Now how big is the demand for intelligence, it may well be the largest market that's ever created. By 2030, estimates are that token consumption will explode, increasing by 3,400%. And as organizations scale their AI workloads, there's a new economic reality, cost curves, compute demands and data movement patterns are all being rewritten. AI is not just changing technology. It's changing the economics of technology in favor of enterprise infrastructure. And now is the time to decide how you can most cost effectively generate the tokens that you're going to need for the long term. We now have 5,000 customers running their Dell AI -- their workloads, AI workloads on Dell AI factory, turning their AI ambition into scale production from data to tokens to outcomes. Customers turn to Dell Technologies for the design and innovation, the engineering, the supply chain, the services, the support and the financing, all working together. Time to first token is incredibly important with investments of the scale. And we can deliver hundreds of AI racks a week to a given customer up and running in hours, generating outcomes immediately. Now manufacturing, particularly advanced manufacturing is where AI stops being theoretical. And it has to work with super complex and precise machines, sensors, supply chains, safety systems and make real-time decisions. Join me in a visit to modern industrial innovation. Please join me in welcoming Suresh Venkata. Suresh -- Yahoo of Honeywell. Suresh. Rest so great to have you here. When some people think of Honeywell, they may think of the Honeywell for in the past, but we know Honeywell is a different company today. modernizing and far beyond the roots in industrial controls, Tell us a little bit about what the modern Honeywell is today.

Sure, Michael. Before that, true to be here. This is my first ever then being with you right on stage, fantastic to be here. Honeywell under the 140-year-old company and one of the most important transformation we are going through at this point in time. We are positioning 3 independent companies, advanced materials. We have already spun them off, which called -- we have coming up with aerospace and then new Honeywell company, which is going to be a pure-play automation company. And then what do we do? It's a controlled system company, which runs industrial oil and refineries, commercial building, utilities, data center, we do the best, connecting to their installed base, assets, controls platform. That's what we do, optimizing and building application all along 140 years of automation. Few things that's happening right now and why I'm so excited about the future Honeywell is the customers are looking for improving their throughput much bigger and better. Not only that, their systems are fragmented, siloed systems all over the place. What -- the current technology and our exploration and innovation is going to -- there's an opportunity. So we have come back and build this whole forge as a platform to integrate their entire backhaul to optimize their operation. So the future about automation autonomy is going to be so real, and I see the possibilities for us to really serve them better.

We've been working together for decades, and we've been honored to be a part of your systems. And now tell us about Forge and what capabilities that's going to bring to customers, what is it going to enable? How are they actually going to see it in their environment, be meaningful and impact them?

Michael, I'm going to be breaking down into 2 chapters. Chapter 1, 2016 to 2024, I would say. We will force like any IoT platform, connecting to our installed base improving the visibility of our operation assets and we built enterprise SaaS-based software application. I think we brought in a lot of improvisation in terms of service upgrades, software upgrades, -- and we brought in what they call visibility and transforming their operation. Chapter 2 for the last 18 months, we said it's time that we actually not only have a connectivity tissue to the operation, but you have to contextualize and understand assets, processes, systems much better. I think with an advent of all the foundational models and AI at scale at this point in time, we are venturing into what we call a Chapter 2, which is all about you connect your operation, you have to contextualize process data and system. And then if you are able to simulate and emulate you have a better prospect in optimizing the operations. So we believe that time has come in to drive something at scale. Let me give 1 example that will give you a context. We work with a number of customers across the board. Let me take an oil and gas customer in Middle East. We have done for the last 10 years. Having our Forge platform, connecting to their assets, we would predict their uptime, downtime, any downtime for an oil and gas refinery remains impact to their throughput in year, we've done a lot more work right there. But we believe that there's a second phase, which is all about contextualizing all types of assets that are close to 100-plus asset types your ability to contextualize asset characteristics and names in terms of oncology and then taking their process units and contextualizing them, can you predict the eat, you predict the throughput assume that if you have a feedstock availability, can you predict what those outputs would be. That's pretty much what we are venturing at this point in time, Michael.

It's in a way analogous to the entire sort of AI macro theme that's going on in the sense that in the past, all these things were individual silos. But now you're bringing them all together with data and intelligence and helping drive yield and safety and better outcomes. And how is Dell engaged in this and give us a sense for how we can help in making all this real.

Two things. I think we partnered with you and NVIDIA in the last year or so. We said there are 2 things that's required. One, you've got to extend your intelligence beyond cloud and Smart Edge. You got to extend it across your enterprise network. So we actually believe that introducing this concept called a physical server at each site level, you have an ability to really balance out your intelligence workload across the board because the decision-making for any autonomous future, latency matters, data transaction matters, so we started creating this concept call, forge cognition with Dell and NVIDIA. I think we're happy to say we are piloting at this juncture, getting great results at this point in time. For me, partnering with Dell and NVIDIA is not just about getting an infrastructure, high-compute services, but this whole AI stack which is a combination for me is right from an ontology simulation, getting this entire model, as you talked about the there's a proliferation of models and then getting an agentic AI platform that you could run at scale, secure and ability where our customers can trust. I actually believe that we are going to be bringing this whole industrial AI at scale that is so essential to transform whatever we're talking about, which is automation to autonomy.

Awesome. So customers are going to feel something very different. Sort of give us a vision of 5, 10 years from now, what does this likely look like for 1 of your customers?

The first thing that I would actually say is autonomy for us means it's not about taking human beings off. We still believe there are untapped potential in every operation we touch. I'm sure there's a pressure from our customer side. improved throughput, help us to expand the margin. So we believe that the new age of AI industrial AI is going to help the workers perform their task much better. The assets to extend their life cycle to their peak performances, the operation processes, we expect that to actually be expanded. So I think we have brought in a category at this juncture where we believe it's no more connected intelligence systems or operation, is going to be a self-learning and evolving operations of the future, which we believe that we have an opportunity to create along with Dell and rest of the partners moving forward in the world of industrial side, utility side and the commercial building side, and we are extremely happy and thankful for everything that you all have done. I'm sure I said this in the past that Dell had helped us to really deploy our control systems for the last 20 to 25 years, Experion EBI in Niagara, which we call a winter strategy. And then now you bring in a GPU accelerated and extend this across the value chain. I don't integrate something interesting for the world.

Fantastic. Thank you so much, Suresh. -- it's a great -- it's really a great honor a privilege for us to be part of the solution that you're delivering to so many important customers and activities around the world. Thank you.

Happy to partner.

So not long ago, AI meant assistance that could write faster and summarize better and answer questions. But that was sort of the age of 20% to 30% productivity gains, it was valuable and kind of amazing, but really only the beginning. Now we're deploying agent autonomous agents that plan and reason and execute and adapt and close the loop. And agents are not something you just simply bolt on to your legacy systems. That's not going to get the optimum results. They are digital workers. They have memory and credentials and access and the ability to take action. And this requires a new architecture for work itself, trusted data and governed action and infrastructure close enough to be able to make real-time decisions. The old systems and processes and the patient decision-making from the past are no longer enough. And it is time to completely rethink and reimagine your company's workflows for agenetic automation and for recursive self-improvement. And that is going to lead us to gains of 20 and 30x in terms of productivity improvement. So who gets there first, we'll rapidly distance themselves from all the rest. And the companies that do not become agentic AI-driven businesses, I think they'll struggle to survive. There is a massive AI investment boom that's already underway. And a productivity boom is beginning. And in some companies, including ours, see you, David, it already has. The rate of change has gone parabolic, and it's not slowing down. This creates monumental opportunities, and it also creates new responsibilities. The agents have credentials. They have access and autonomy. Security must expand beyond the human users to also govern the nonhuman workers operating at machines feed. If a bad actor can influence an agent or if an agent is trained on bad data, the blast radius is no longer contained to just a single system. It can propagate across all your workloads and your infrastructure and, in fact, the business itself. This is already changing the boardroom conversation. It's not just are you secure, but do you understand what your systems are doing on your behalf. You can't protect what you can't see and you can't manage. So we give you visibility and control, no matter how fast these threats are evolving. Zero Trust principles keep you adaptive and resilient and are built in from the endpoint to the core. Speaking of the endpoint, you can see this into the deepest layers of the PC with our Dell trusted device platform. You can store the key credentials. In fact, the keys to the kingdom, if you will, in control vault so that only authorized identities, whether they're human or machine have access to your systems. Agentic AI is only as good as the data that it can trust access and act on. And if your data is siloed, your agents are blind. They can deliver the value that you expect. Everyone has access to the same models. The differentiator is your data, the unique proprietary hard one knowledge inside your business. The agents are hungry for your data. Now imagine tens of thousands of them going to millions of them, running 24/7 thousands of times faster than any person. The Dell AI data platform with NVIDIA has been engineered to address the data bottleneck. Within the platform, the data orchestration engine is the intelligent control center that terms that raw fragmented data into production-ready AI. You can transform and clear your data with NVIDIA's close to now 1,000 models and QX libraries integrated natively and you get 12x faster vector indexing, 6x faster data query at 19x faster time to first token. And we keep pushing the limits of what's possible. Today, we are super excited to announce that we are working with Starburst to enhance the analytics engine with NVIDIA with their vera CPUs for 3x faster SQL and with black well GPUs for 6x faster SQL, leading institutions like Bank of America, who already have a partnership with Star Bert, NVIDIA and Dell Technologies are going to be able to use these new capabilities to support AI-driven analytics while meeting they're important governance, regulatory and resiliency requirements. And for extreme scale AI and high-performance computing environments, we're introducing Dell exascale storage a unified RAC architecture supporting power scale, object scale, PowerFlex and the light new file system. Exascale is a 4-in-1 storage platform built to deliver up to 6 terabytes of throughput per second per rack. And we all know that AI devours data. And our family of unstructured data solutions including power scale, object scale and the lighting file system, the fastest in the world, are being deployed extensively by thousands of customers around the world, including some of the AI native companies like Core weave and end-scale iron fluid stack, Boston and many others and many other leading firms that demand incredible performance like McClaren Racing Hudson River trading, Quadrature, 2 Sigma, Optiver and many, many more. Now of course, all that data needs to travel at blinding speeds. And so AI factories need AI grade networks. The new power switch Spectrum 6 gives you the massive scale, speed and efficiency and reliability required for AI. And we've added the NVIDIA QuantumX 800 liquid cooled with co-packaged optics also to our networking solution. Now all that storage and networking feeds the beast of accelerated computing. The PowerEdge XE9812 built on NVIDIA's barorubin NVL72 delivers 10x lower cost per token than black well for massive scale genic AI inferencing. -- and the new lineup of PowerEdge X servers built with NVIDIA, HDX, Ruben, NDL 8 are an industry leader, supporting up to 144 GPUs per rack with 100% direct liquid cooled compute nodes, 5.5x more powerful than the previous B200 GPUs. And to bring it all together, we're excited to announce Dell Power Rock, a turnkey rack solution where the AI compute, the networking and the storage are all engineered together designed, tested, validated and delivered as 1 system. With power, you can deploy a fully integrated system where the thermal design, the power management and all the software have been engineered and tested to work perfectly as one. Now one reality we're all facing is that AI is energy intensive, even though the amount of energy used per token is decreasing quite rapidly. A single rack of NVIDIA robin GPUs can draw over 130 kilowatts. And as the industry builds hundreds of thousands of these racks, the pressure on the power grid is real. And we are committed to being a part of the ongoing solution. Our Power Cool Heat Exchanger delivers up to 60% reduction, including energy consumption and cuts the annual energy cost by 1/3. And our Power Cool CDU is capable of cooling varirubin-NDL72. It delivers more than kilowatts of cooling capacity. We designed for performance per watt, so you could deploy more AI within your existing footprint. Now of course, the most efficient token is the one that is generating closest to where your data is. So we're expanding the Dell AI factory with NVIDIA to accelerate a genetic AI in the enterprise. We're announcing Dell decide AgenticAI, which brings together the highest performance Dell Pro precision workstations, NVIDIA Nemola and Dell Services. You can test, build and fine-tune agents locally while running the latest open weight models in the $70 billion to $250 billion parameter range, up to $1 trillion parameters. And you can do it without unpredictable cloud costs, bandwidth costs or risk of IP leak. This is unmetered intelligence, and you can break even versus public cloud APIs in as little as 3 months. NVIDIA Open shell is also now supported across the entire AI Delhi factory, giving developers a sandbox to secure and fine-tune their AI agents from workstations to servers. And Dell technology support for NVIDIA, AI 2.0 blueprints gives you a tested foundation to deploy multi-agent workflows. We are providing a consistent path to accelerate the move from deployment to pilot to production at scale. From the PC to a full rack of 144 GPUs, Two, the biggest data center you could imagine in the world, all of those racks, we are engineering the accelerated architectures to work together as one system agenetic and autonomous. The Dell AI factory with NVIDIA, the backbone and the brain of the modern enterprise. All right. Now no company does this alone. And the next era of infrastructure is going to be built by deep partnerships between companies that are advancing accelerated computing and the companies that know how to deploy it across the real world. Please join me in welcoming a great partner friend, a true leader and visionary of the AIAge NVIDIA's Jensen Huang. Thank you, Jensen it's great to have you back here. At Del Tech World, we've been talking about agents.

I'm here every year selling Dell.

We appreciate it, man. We appreciate it. So give us your perspective on where we are in this genic recursive self-improvement world. It looks like every time we wake up, there's been some leap in the model and the capabilities and sort of it doesn't seem to be plateauing. Give us your perspective from the front lines.

Well, 2 years ago, when I was here, we had just started with the agent -- excuse me, generative AI journey, right. Generative AI can, of course, generate content -- but remember, it can also generate content to think with, generate thoughts, which led us to reasoning, which led us to planning, which led us to agentic systems. So now we have -- we now have, for the very first time, useful AI, which is the reason why your demand, my demand is going parabolic, utterly parabolic because, right, it's going parabolic because Agentic systems -- the AI has to understand, has the reason, has to think it has to plan, use tools, look at the results of the tools, think about it some more, come up with maybe an improved plan -- and so it iterates until it can get the job done using a whole bunch of different tools. Well, the amount of computation necessary because it's running autonomously for so long, instead of just responding to a query, the amount of computation has grown 100x, 1,000x -- and depending on the work that you're doing, sometimes we'll kick off a software programming job, it doesn't finish for a week. Of course, it did a week what would have taken a whole team a month to do. And so big deal in productivity but gigantic leap and computation requirements. So that -- let's just say that computing went up by 100 or 1,000 times. Meanwhile, because it's so useful -- the number of people using agents now all over the place. Every enterprise, all company, your company, everybody is using agents all over the place to do software development, devops, SE, all of our CICD work, QA testing, the amount of software work that we do in the company now supported by agents is incredible. One engineer, a really good engineer today is working with an agent, but a really great engineer in the future is going to be orchestrating a whole bunch of agents who are going to be orchestrating a whole bunch of sub agents to do work. Well, between the amount of computation and the amount of demand use going up, the product of that, that's our demand. And so we've now arrived at the era of useful AI, which is really just really exciting for all of us because until now, it's been novel, interesting, incredibly exciting. But really at the enterprise level, many of you have said before last couple of years, the impact of AI is the potential is incredible, but the actual use was minimal. Now it's taken off. And we're starting to see now -- we didn't -- a couple of years ago, we started -- we didn't have 5,000 enterprise customers. Now we have the biggest companies in the world like you saw with Lilly at Samsung and Honeywell, they're piling into this in a big way. But really, that's just starting, right? And to reimagine their workflows to be able to understand the trajectory of how this is improving and how it can affect ultimately what -- the companies could become, that really is just an idea. I mean we haven't seen that in any scale. I mean there are some companies that are doing it, but -- it's a very small number today.

No doubt Well, you're experiencing, and I'm experiencing, our company has always gone fast. But it's gone -- it's really going fast now. And just the amount of content that's being created internally, the progress that's been made, people said that AI is going to make us more productive. There's no question about that. What took months now takes weeks, what took weeks, now takes days and what takes days, now it takes hours. And things that would take an hour, you and I pretty much expect it instantly now. And so what has really changed is that our ambition has changed. There's no question my ambition has changed. I want it to be somebody to do something, make a contribution. But that was the old Jensen -- the new Jensen, I got big ambitions now --

You got to all like to ask ourselves the question, how I is up. Well, it's pretty high, right? Yes, yes. So we've got all these great products -- we're embedding AI in everything and enabling this distributed inference and intelligence. This is the best room I've ever been -- what is your favorite? Well, it's hard to I got -- I love all my kids. How about the one on the end. You've got to love that one, most others, yes. Well, this is incredible. And so we -- Michael and I have been working on a whole new line of computers to run agents. The way to think about agents is this. There's a large language model. It's gigantic. It's the most computationally intensive piece of software in the world's ever known. And then we've run in that system towards the end. That NVLink 72, the world's largest scale-up single domain computer, okay? It's just 1 giant system operating as 1 computer. And that system has a large language model in it, 1 terabyte, 10 terabyte of parameters, no problem. And so that's 1 giant system. That's the brain. However, an agent starts with the harness. -- the harness has to sit in a secure and governed container, we call it a sandbox. And so the NVIDIA open shell open source sandbox is the security system that just about everybody in the industry is using, inside, we have a reference harness we call NemoCloud. NemoCuaruns on. And that CPU could be a CPU and there, a CPU here a CPU in there. It can also run Nemotron or any of the models, open source models that you would create your own specialized agents for your own companies trained for your own special domain of data or skills. And that would run locally, if you like, and then the large language models could run in the cloud. And so you have this hybrid AI. The thing that's really cool is that NVIDIA's architecture is the only architecture in the world that runs every Frontier AI model and lately -- recently, in the last year or so, entropic has been really leaning into the NVIDIA architecture. So now we support every single Frontier model. We support every open source model and they -- we support them in the cloud or locally. As you can see, our computer is the first 1 in the world that runs in every cloud, but it also runs locally. And if you have your models that you're quite sensitive about, our systems are built with confidential computing. -- that's way you don't have to trust the operator that's operating the data center with your secure data. And so all of these architectures now run at every cloud runs every model, runs hybrid AI and runs agentic systems. And so you have your harness that runs on a CPU. The CPU also uses tools and the CPU we created, I think you mentioned it earlier, called Vera Yes. The CPUs of the past, the CPUs of the past were built for hyperscale clouds. And so you're renting the CPU cores and so you're optimizing for us many CPU cores as possible. Well, agents in this new world, you're generating tokens. You're not running course anymore. You're generating tokens and marketing that. That's the economics of this AI Vera. And so the AI wants to generate runs work and generate as many tokens as possible as quickly as possible because that's the output of the intelligence. Vera CPU has the highest single traded performance of any CPU in the world. It has 3x the memory downwidth of the fastest CPU in the world. And as a result, Starburst, Duck EB, all these databases run incredibly fast because the agents are pounding on the databases, so the CPU is better be super fast. The agents want to get through its work, so the CPU has got to be super fast. Otherwise, the big machine down there is waiting for the agent to get its work done. And so now you have the harness running on the systems here local AI models running on the systems here and giant models running in the cloud or in your own data center on the big machine down there. Well, let's go check it out. But as we do that, what's exciting is to think about in the past as humans, we would do work, right, and we would pass it on to the next human. But now we have all these agents that we're managing, right? And an individual can manage I don't know, 1,000, 100 agents. And so the possibilities that, that unlocks in terms of human creativity and what humans are going to be able to do. And I love this idea of the unmetered intelligence, right? -- where you've got the power in your own PC in your own data center and you can use it with your own data, that's sort of super cool.

They don't have to struggle with token anxiety tell you on out of tokens or get that retail. But Jensen since you're here, would you do us the honor auto graphing this latest.

All right. This is -- what's the date today? 18, May we're not going to sell this one. This 1 is -- you got to sign it, too. I already signed it. You did Well, you got it right here. here you go. Michael has already signed up right here. And so this is -- Michael, this is 100x larger than this one, the station, right, exactly the same architecture.

The GB 300.

GB 300 in here. That's 100x larger than this one. This is the only computer beside computer in the world that can run a terabyte a $1 trillion parameter AI model okay? So this is doable.

Would have been unimaginable even a year or 2 ago.

know it. One silly, I mean, this is basically be a cloud just 2 days ago. And so -- so this is the station, and this is 100x bigger. So yes, this one, and that's 5, 6x larger than this one. And this is one of my favorites. This is the smallest same architecture is that one is that one. It's incredible, right? One architecture, Incredible.

Well Jensen, really appreciate you being here. We treasure the incredible partnership with NVIDIA, all we've been able to do for customers around the world. And look forward to...

And we grew up together.

Yes, we did without 31 years, we've been doing.

Guys keep up the great work. Thank you, everybody.

Thank you, Jensen. Well, thank you, Jensen. His enthusiasm is infectious. And his passion for innovation and its belief in accelerated computing and what that can bring to the power of AI around the world. Now let's look at that power in action. In modern health care, intelligence is not measured in tokens. It's measured in time, in accuracy, and compassion and in live safe. [Presentation]

Ladies and gentlemen, Dr. Chuck Fraser is here with us today, an incredible pediatric cardiac surgeon. Thank you, Dr. Fraser. This is why we do what we do, why we are so passionate about solving your toughest challenges. For too long, the AI conversation has been trapped inside the screen. The real story begins now. AI is moving into hospitals into factories into schools, energy grids, laboratories, cities, homes and yes, even into orbit, solving problems at the scale of humanity. Over the next few days and in tomorrow's keynote with Jeff and Arthur and in our technical sessions, our trailblazers and the solution showcases. We want to inspire you to be bold and to move fast. The road ahead is bright, and it is beautiful, and we will be with you on it every step of the way. Thank you, and enjoy the show. [Presentation]

Crazy train. I know the team doesn't know what I'm about to say, but they have no idea that brings back such fond memories. This song hit the United States in February of 1981 about ready to date myself. I graduated high school a few months later. On August 12 of that year, right before I entered engineering school, something very important happened that's changed my life forever. IBM launched the PC. And I've had the privilege of spending most of my career Dell, nearly 40 years now with a front row seat wave after wave of innovation from PCs to servers, the storage, the Internet, mobile, and here I am today in front of all of you, getting to talk about artificial intelligence and agentive workforces. What an incredible ride -- welcome to Day 2 of Dell Technologies World. Michael set the stage yesterday and as usual, set the bar high. So I have a bit of work to do. He talked about what the modern enterprise looks like. Today, we're going to talk about how to build one. Last year, you might remember, I said token use was likely under called -- and I ended my keynote by saying we are on the forefront of the Agentic era. It turns out I was a bit too conservative, which seldom happens, but I was crossed a real inflection point. The models, the software, the hardware, all improved faster and more capable than anyone predicted. -- what was thought to be a 3-year journey happened in less than 12 months, happened in less than 12 months, utterly amazing. And you all know I love a good top 10 list. I'd like to start these conversations with them. The team has continued to try to break me of this habit, yet again, they did not succeed. I have 10 things that I want to talk about that changed our world since the last time we were together and they're profound. The first one, number one, AI went from an adviser, Michael talked about that yesterday to an operator. AI executes the answer, manages the exceptions and escalates what it can resolve. Number two, model prices have dropped roughly 40%, excuse me, 80% but token consumption is up 10x, surpassed 100 trillion tokens in a year in 2025. Equally important, the context windows across millions of tokens, you can now in our entire code base a year of contracts, a full operations history all done in a single pass. Number five, training built the models, inference runs the business inference workloads account for nearly 2/3 of all AI compute this year. The compute curve bent and quite honestly, it's never going back. Number six, generative AI software and enterprises, the spend tripled in 2025 to $37 billion. Companies that are spending aren't experimenting anymore, they are building one of my favorites, physical AI left the lab, robotics, autonomous systems and bodied agents are now showing up in factories, warehouses, hospitals and farms. One of my personal favorites. The PC is part of the AI stack. Power at the desk side matters because workloads like software development, media workflows have all moved to the end point. And quite frankly, the talent conversation has flipped. It's no longer where do I find AI engineers. It's how do I get everyone to think and work like an AI engineer. And then lastly, the conversation changed. Every CIO I meet has stopped asking me, should we and have started asking how fast can we things, 12 months, any one of them in any other era would have been the headline of the year. This time, they are the context of what comes next. Think about that. So what I wanted to do now is share with what I'm seeing inside Dell and so our customers inside the 1,000 environments that we're helping customers implement today, 3 patterns, 3 distinct patterns that I think needs to sit with each and every 1 of you as you leave this morning. The first observation kind of buried in the top 10. Token costs are falling fast. Token use is exploding. Token costs are down roughly 80% year-over-year. Tokens alone for reasoning is up 320x. Think about that. Model cost, token costs down 80% explosion of use 320x, so much that you see the unit prices collapsing leading to this volume is exploding. And if this pattern sounds familiar, it should. We've seen this before in bandwidth, storage, compute for decades now. Cheaper units unlock so much new usage that the total spend goes up, not down. and we've never seen it move this fast at this rate. And we sit here at Dell. We've had groups of engineers who have consumed a month's worth of allocated tokens. Yes, we were allocating them in the past. I'll get to that in a bit. and they consumed it all within a few hours. And not because anything was broken because it's working. This is what success looks like with agents and you better be ready for it because it's coming. Observation number 2 AI productivity is extremely nonlinear. Inside almost every company deploying AI today, 5% of the people are driving 95% of the value value. Why? Because when someone learns how to use agents effectively, they use them for everything. -- research, coding, modeling, strategy analysis. Their day looks completely different from the individual right next to them. We call them super users of intelligence, a small slice of the workforce driving most of the games compounding their advantage week after week. So if you're measuring AI ROI, mouthful, by averaging across your entire organization, quite frankly, you're doing it wrong. That's like measuring system performance by averaging the idle cores. The right questions are who are my super users? What are they doing? And how do I get everyone to become one1 of them. Observation number three, if it already has it, tokens are about to be a line item in your P&L. Agents are doing in minutes and hours what took multiple sprint teams of multiple people days and weeks, writing code, testing code, deploying it, planning the next iteration. That's not a 30% productivity gain. That's a 10x gain, that's 100x gain and oftentimes even greater than that. That's structurally a different company. As the agents take on more of this cognitive work, cost migrates from headcount to tokens. Let me say that again, it's very, very important. As agents take on more of the cognitive work, cost migrates from headcount to tokens. Tokens are going to be much greater than a rounding error in the future. And the companies who are planning for this shift now are going to have a big advantage over everyone that waits. If I take those 3 observations together, infrastructure demand is exploding as model prices collapse. The gains are concentrated in a handful of super users who are pulling away from everybody else week after week. -- and the cost structure of cognitive work is inverting in front of us. Cognitive work, the cost is inverting in front of us. Every enterprise in the world was built on 1 assumption, cognitive work scales the human hours. You want more analysis, hire more analysts, more code, more developers genetic AI has broken that ratio forever. And here's the part that nobody wants to talk about outlot. Most of our operating models that are built for the world that we lived in yesterday are now broken. You can feel it, your teams can feel it. Nobody has completely figured this out yet. I know we certainly haven't. But what's clear to us is the companies that win in the next decade will be AI native. What is AI made of mean? I want to be very clear with this. You don't have to be born in the AI era to be an AI-native company. It's not a birthright. It's an operating model. It's built on the premise that intelligence is a utility that it flows through every workflow, every decision, every product, every interaction. In my home state of Texas, we often hear the transplant, say to Dorman of them. But the transplants do say, I wasn't born in Texas, but I got here as quick as I could. That's my challenge to you. just because you didn't start with AI, does it mean you can't get there. And the question you're going to have to ask yourself, are you really ready and willing to disrupt what you've built -- are you ready to tear down to the ground, the old ways and rebuild in a new way? Or as I like to say, breakthrough ship and start over. That is ultimately what we're talking about. Things have to fundamentally change. This takes me to the next part. So as we think about those observations and the impact they're having is going to have a profound impact on how we build an AI-native enterprise. So I have 5 imperatives, things that you need to do right now and you got to get it right or you're not going to be prepared for what's coming. The first thing, Arthur will expand on this in a bit later, build an AI-ready data foundation. Michael said this yesterday, AI runs on data. However, in most enterprises, the data is scattered across dozens of systems. 80% to 90% of it is unstructured. None of it is connected in a way that an agent can use effectively. To power agents at scale, you need a real-time connected knowledge layer. And here's the structural and architectural decision you're going to have to make as a result. Don't move the data to the AI move AI to the data. That's fundamentally a different approach and, quite frankly, a decision that you need to make now. Number two, build a distributed AI infrastructure. training and inference are very different workloads with very different physics. You heard Jensen and Michael speak briefly about this yesterday. Training needs massive GPU clusters. -- centralized by necessity. And I think about inference where the growth is, you have reasoning models executing multistep change. You have agents calling models over and over planning, evaluating, iterating these workloads are 10 to 100x more compute-intensive. Jensen said yesterday 1,000 times, 10, 100, 1,000 times more compute intensive than what you were running 18 months ago. an AI-native enterprise has to be built for both. Number three, secure autonomous systems. Here's the thing about agents. They just don't call the model, they call your CRM they call your ERP, they call your financial systems, they call your customer databases. Every one of those touch points has to be secured, logged and must be reversible. Because when an agent takes action on your behalf, changes the price updates a customer record, initiates a procurement workflow, you need to know what it did, why it did it and how to undo it if it got it wrong. -- and an AI workforce, every action needs a receipt. And that's not a compliance check box, that's how you build trust into the system that will act on its own. Imperative number four, integrate the enterprise stack. The agent becomes the coordinator, the stack has to let it. Agents need to plan task, call tools, execute work, handled exceptions across your entire stack. That means an API-first architecture, workflow orchestration and agent frameworks that can do multistep execution. If your stack can't do that, your agents are going to be siloed and expensive. And that's simply not a conversation you want to have with your Board. The last imperative, restructure for intent AI and Toconomics. -- to the question isn't whether consumption grows, is it absolutely will. The question is, are you running the right tokens on the right infrastructure? A routine summary doesn't need a frontier model, a sensitive financial analysis shouldn't leave her building. and a complex reasoning chain might need the most capable model wherever it lives. You need to think the right workload, the right model on the right tier, to be edge data center cloud -- and if you do that correctly, you're going to get the optimum performance, privacy and cost efficiency, run everything in 1 place because it's easy be ready for a surprise, and that surprise is a large bill that's only going to get larger. And Michael said this yesterday, it's fundamentally something we believe the notion of token routing where to put that token is going to be 1 of the most important infrastructure decisions you will make. 5 imperatives, 1 direction, the AI native enterprise isn't a vision anymore, it's a blueprint. And the reason I'm confident you can build it today is because the ecosystem is here to support you. So now let me show you what that looks like with one of our most important partners. [Presentation]

Thank you Jim and I, on premises, air gap if you need it, Dellis Customer Zero, available today. Pretty exciting. So I've been talking this morning about the AI native enterprise and what it has to look like, those 5 imperatives, the data foundation the distributed infrastructure, the security, the stack integration and the toconomics. And Thomas just shared what happens when that comes together with a world-class partner extraordinary and more to come. Now we're going to shift the conversation a bit because building the modern enterprise isn't just an infrastructure story. It's an agent story, too. I want to bring someone up here who has been living this from a completely different vantage point. Our next guest is the Co-Founder of Offline Ventures, Co-Founder and Board member of Open Qua Foundation, one of the most important open-source projects in enterprise AI right now. built on the simple principle. The enterprise should own its AI not rent it. Everyone, please welcome Dave Dorman. Good morning, Dave. How are you doing?

Good to be here.

Glad to have you here. So we spent some time backstage this morning talking about what you're doing. It's pretty exciting. You've had a front row seat on some of the changes in mobile, social in your career. You had this at Tiffany with Open Claw. Tell us about that aha moment and why you realized it was going to change everything.

When I was 7 years old, my grandfather put a personal computer on my desk. And the first feeling that I had with it was I could create anything, but it was mine. -- and I could hack it, I could turn it into whatever I want it. And throughout the rest of my career, I've been chasing that feeling. I went to college and I took a Dell computer with me, and I built my first business in my dorm room using that Dell computer. After that, I went on to work at Apple and Facebook. And in each of those places, we were trying to give the power of computers to people so they could create new experiences for themselves. Over the holidays, I was at home visiting family and a friend of mine take me and said, "Dave, you got to check out this open clock thing. And I said, "Oh, it's the family time, it's the holidays, I don't have time to check this out. I decided to install it. And within 24 hours, I had that feeling again. I felt like this is the first AI experience that I've had since ChatPT launched or maybe even since the iPhone that I felt like I had a computer of the future in front of me on my desk. And Open Cloud was the first time that I felt that way. And ever since then, I called up the founder Peter Steinberger and we started working together every day to bring that feeling to everybody in the world.

That's pretty cool. I've been using Open Claw in my dashboard at work. It's made things much easier the team is a little surprised I'm using that. But if you think about the capability of an agent and its applicability to enterprises, we were talking backstage that you've done some research on the use cases of Open claw in the enterprise. How do the 2 merge from what may have started as a personal assistant now into an enterprise class capability?

Yes. I mean I think when you put an agent into the enterprise, the first thing that comes to mind is security and observability. You need to understand what does this agent know? We all know that when AI wakes up, it doesn't know anything about you. And so you have to fill it with context. You have to give it access to critical data to help people do better, work better in their workflows. In order to do that, you have to know what it has access to, and you have to also know what it's doing. And so what we've learned in building Open Cloud over the last several months is figuring that out, giving the agent read-only access at the start and then also building observability layers in order to understand every single action and every single thing that the agent does helps you figure out where it can go inside of your enterprise and how it can help people either in the cloud or on the edge.

We talk Backstage a little bit about one of the biggest challenges as connecting data.

Yes.

I'm sure your experience is maybe with our audience today of how connecting the data and the challenge with that, but also the benefit once you get it right, the efficiency and productivity you get.

Yes. We've seen 2 different things. We have people building massive multi-agent architectures in the cloud, trying to figure out how do I build adapters for each different type of data. It can be a really intense task and people are going about it in a lot of different ways. One of the cool things we're doing at Open Claw and our own workflows is we actually build crawlers that go and grab data from the cloud and pull it on to the edge device. And store it in SQL Light or other lightweight database and simple data structures so that the agent can rip through it really quickly and answer questions locally on device and make it really fast, really easy to use. And you can also know that it's happening on that device, the data, the workflow right there right next to the person, you know it's secure. And so we're doing this internally. It's how all of us at the central core of Open Cloud, the maintainers are key engineers. We use workflows like this every single day.

Pretty cool. Maybe we should wonder over to the devices.

Yes, I saw the thing over here.

We sent a GB 10 to -- we're going to get to this one. I know what -- you would have seen a SP29155098 How about here. Essentially want to jump our G10 device, tell us your experience with it, particularly as we think about moving more of the capability to the edge.

Yes. I mean these things are amazing. 128 gigs of RAM, -- you've got a black well in there. I've got 1 set up with Open Claw on it. It will run any open cloud workflow. You can use 30 billion parameter models, run data right on the device. You can also set it up as a satellite of your core cloud -- so I have a really advanced cloud that I've been working on for 3 months. It runs all of my key workflows. It has access to all of the foundation models, but it now also has access to my 10 -- and so when I want to run a really sensitive workload, something that's important, secure, I don't want it leaving my own machines. I use this as the local inference layer and have my main claw talk directly to this and run load.

But from our conversation, maybe you want a little more. And why would you want more.

I used to to think that I wanted a bigger monitor, but now I know I want a bigger computer.

And why?

Well, what's interesting is that this thing puts you about 1 year in the future. This thing it puts you 5 years in the future as an individual, and this thing you could run an entire company on. It depends on the size, of course, -- but a 3,800 under your desk can run a $1 trillion parameter model. You're talking Frontier level performance under your desk and the ability to run it 24/7 multi-agent workflows, absolute frontier engineering. I mean I want to put one of these underneath every single 1 of our core engineers' desks so that we can be running in the future today.

Available on wwwdil.com. Maybe a last part of your question. If you think about a thought to leave the audience today, what would it be? And maybe tied to our conversation about the notion of needing more capability at the edge and the role the edge plays in the future with agents and AI?

Yes. I mean I think give people the tools to try this. If it needs to be at the edge, do it, start here, start here. models need context in order to provide the best, most powerful possible answer workflow -- we're seeing that people are coordinating their work in vastly more efficient ways, put an agent into your workloads, give it read-only access to the data to start and just get it in the game, see if it can organize a simple workflow, help people do things faster, more locally. And just get started, don't be afraid.

Get started, don't be afraid. Thanks for joining us today. My pleasure. Thank you -- so confirm something I've been saying all morning, -- the enterprise has to own is AI, not rent it. But it raises a very, very practical question. Where do I run this stuff? Yesterday, Michael announced the Dell deskside agent AI -- the principle of it is very straightforward. The most efficient AI token is the 1 produced closest to the data. And most of your data is not in the cloud. It's on-prem, it's at the edge. -- that's where the AI needs to run. For every developer and IT leader in this room, that means a secure sandbox where your teams can build, test and fine-tune agents locally. No unpredictable cloud costs, no bandwidth issues, no bandwidth penalties, no IP leaving the building, and I'd like to show you what that looks like. [Presentation]

Pretty cool. That's right, that 45-year-old device, the PC, a free unmetered token generator. That's the future. 1 developer, 10 agents, 1 billion tokens in 24 hours in the cloud, $3,400 on a workstation, 0. That's not a demo. That's a deployment model we're helping customers build right now. We kind of stole the thunder a little bit earlier with Dave, but I want to show 3 products that actually build upon 1 another to do exactly what we just described. The GB 10, 1 petaflop performance. It's optimal for models in the $30 billion to $200 billion range. It generates 40 tokens in a second. This is where most teams should begin. If I go to our Dell Pro precision towers, this is a T4, 7 petaflops of performance. Scale to a 500 billion parameter model, generates 400 tokens a second. This is where serious developers should start or move next and then the big boy, my personal favorite, the GB 300, 20 petabytes of performance Jensen said this yesterday, up to 1 trillion parameter model, generates 700 tokens a second. This great black well Ultra Super Chip drives over 1,500 watts of power. You can't cool that by air. So we engineered a closed loop liquid cooling system that doesn't exist anywhere else. This one makes me smile. This is special. Big boy is fascinated. That's our new portfolio of Del Pro and Dell Pro Precision shipping available today. And as I mentioned today, wwwdel.com, you can get the GB 300, and we've already shipped the first last week. So I've been talking about what it takes to build the AI native enterprise. The desktop is where your developer start, but to scale it from the desk to the data center to the cloud, you need infrastructure underneath it. That's Arthur's world. I'm going to let him tell you what he's built because, quite frankly, some of the stuff they've done has even surprised me. Everybody, Arthur Lewis.

Don't ask me about the. Good morning, everybody. It's great to be back here at Dell Technologies World. Jeff just laid out a clear and compelling demand signal and 5 very important imperatives. There is little question that the Agentic workforce is emerging and that the token explosion is real. And we have a lot to cover today, and I want to jump into something that I think is incredibly important, which is building a data foundation. And let's start with a very important fact the vast majority of enterprise data resides on-prem. This data is your strategic advantage. Yet no frontier model has seen it. It is sitting cold, it is sitting dark, invisible to AI generating little to no value and leveraging ever-evolving techniques like active learning, fine-tuning, reinforcement learning across the entirety of your data is not a nice to have in the world of agented it is an absolute must-have. Not only will it optimize the results, it will drive significant token efficiency. For example, a task that's run against a frontier model that consumes 4,000 tokens in chain of thought and 3 tool calls can be done on a fine-tune model with 400 tokens and a single tool call -- and of course, this calculus multiplies daily across millions of inference. So how do you build this muscle? How do you build this engine? And how do you do so at scale? The answer is simple, the Dell AI data platform. This is an engine that is fueled with your data shaped by your intelligence and tuned to your outcomes. And it comes in 3 simple layers. -- layer #1 data preparation. In order for your data to be AI ready, it must be discovered, cleansed and labeled and transformed into something that the AI can actually interact with. And this is exactly what the Dell orchestration engine was designed to do, built from our data loop acquisition and in close partnership with NVIDIA, it can handle structured, unstructured multimodal data turning raw enterprise content into rich curate and govern data sets ready for training and inferencing. This orchestration engine comes with a built-in marketplace with over 200 applications, models and data pipelines, including NVIDIA NIMs as well as no code pipelines for RAG, active learning and fine-tuning. You can go from an NVIDIA blueprint to production deployment without writing a single line of code. And powering all of this are the Dell data engines designed to transform inquiry data and to do so at great scale with CUDA -- with NVIDIA CUDA-X libraries natively integrated, customers are seeing 12x faster vector indexing, 6x faster queries and 19x faster time to first token. Let's take a look at the solution in action. [Presentation]

So layer #1 data preparation. Layer #2, disaggregated inference. Once your data is AI ready and available to agents, speed now comes down to the model's ability to access and query data to generate new tokens. And this is exactly what lightning with its container-based and high availability architecture was designed to do. Lightning is the fastest parallel file system in the world, delivering 150 gigabytes per second of throughput per rack more than twice that of our nearest competitor. Lighting also provides the context memory extension and storage access needed for disaggregated inference, delivering 20x the performance of flash-only scale-out alternatives. -- coupled tightly with NVIDIA Dynamo, Lightning is capable of bypassing the prefinal stage of inference to intelligently route KV cash data between GPU memory and storage disks leveraging RDMA in real time. The infinite storage capacity that Lightning brings to bear enables agents to retain long-term context over extended durations something that is impossible to do with limited GPU memory. And the third layer is the underlying storage platforms. power scale, object scale and something you may have heard about this yesterday. Over 1,500 customers today deploy GPUs on power scale. It is NVIDIA-certified super pod validated and supports GPU Direct and RDMA. And when compared against other NVIDIA reference architectures for saturating up to 16,000 GPUs, power scale is incredibly differentiated. 80% less rack space, 8x fewer switches, 72% less energy saving customers $1.5 million along the way. Next, object scale is the world's most cybersecure object storage, delivering 40 gigabytes of throughput per node, more than twice that of its nearest competitor. And our unique implementation of S3 over RDMA and GPU Direct shows significant gains versus traditional S3 only implementations, 70% less CPU utilization, 30% more bandwidth and 80% less latency. And when coupled with advanced features like S3 tables and S3 vectors, object scale is primed for a broad array of data sets for AI training and inference. And this new thing was announced yesterday, exascale a 4 in 1, a unified rack infrastructure built to support power scale, object scale, lightly and PowerFlex Exascale is the only 4 in 1 storage built platform specifically designed for extreme scale AI, high-performance compute and enterprise workloads, delivering 6 terabytes of per second per rack unit. So whether you choose a software-defined architecture or an appliance architecture, whether you choose file, block, object or disaggregated inference Dell has the broadest portfolio of data storage in the industry. Your data is prepared. It's moving at speed and it's stored at scale. A holistic workflow designed to drive down your token costs. Now that we've built a strong storage foundation, let's talk about security. Agents don't just analyze data. They access systems, they execute, they modify data. Thus, security has to be a prerequisite. It cannot be a secondary concern. -- and this high level of protection must exist where your data lives at the edge, in the data center or in the cloud. Our research shows 94% of of brand somewhere attacks seek to compromise backup workloads. But more alarming is the success rate, 57%. Why such a high success rate you ask? The reason is that the vast majority of backup workloads today land on generic infrastructure. This is a huge problem. Number one, it's a horrible TCO. But more importantly, the software and hardware are not jointly integrated and hardened. Rather, security controls are layered in after the fact to engine instead of being engineered in from the beginning. Security must start very early from the supply chain and extend into jointly integrated and hardened software. Backup workloads must land on purpose-built systems like power protect data domain systems that are specifically engineered for security, for performance for recovery under attack. This means customers get factory to site verification silicon root of trust, Zero Trust controls, end-to-end encryption and built-in immutability, these are table stakes in a purpose-built appliance. Equally important is the software stack and the ability to securely and effectively manage a complex environment across an entire estate. And this brings me to another announcement. Power PROTECT 1 cyber resilience simplified. PowerProtect One is a unified integrated architecture that brings PowerProtect data manager and power protect data domain together. PowerProtect data remain, of course, the world's leading target appliance rated #1 in customer satisfaction and supporting 650 exabytes of data across more than 15,000 customers around the world. PowerProtect ONE will enable you to scale components independently to improve security and drive down operational costs. With PowerProtect One customer should expect 0 to first back up in 1/4 of a time that it takes with a nonintegrated appliance. 50% of less management -- less daily management overhead, 75:1 data reduction, enabling 150 petabytes of logical capacity per node to faster restore time and 2x faster replication. This is cyber resilience for the AI era. Now that we've built a strong data foundation and we've secured it -- let's talk about modernizing your infrastructure and freeing up critical resources with the Dell Private Cloud. As customers think about building a modern private cloud, they typically think about 3 things. Number one, they think about flexibility because we're living in a multi-hypervisor world, spanning virtual machines, containers and bare metal. Number two, they think about simplifying IT operations and significant advancements in automation and application orchestration. And number three, they simply want to know that their architecture is future-proof and cost effective. And with the Dell private cloud, they get all 3. And the savings are significant, upwards of 65% TCO when compared against legacy hyperconverged infrastructure. And today, we are expanding our offers across broad ecosystem of partners, including the latest software from VMware and Red Hat, as well as power store support for Nutanix and Microsoft Azure local, we are giving customers real choice in automation-driven simplicity. And of course, at the center of any private cloud is storage. And today, we're incredibly excited to announce PowerStore Elite, a new class of modern data platform. This is Well, I haven't gone through the spec shift. This is a new class of modern data platform, a full third-generation refresh of hardware and software -- we've added more powerful CPUs. We've added faster memory and significant enhancements to our software stack and the results amazing. Power Store leak delivers 1.5 million IOPS, which is 3x that of the previous generation. It delivers 80 gigabytes of throughput per node more than 4x the previous generation. We've improved the world's leading data reduction guarantee from 5:1 to 6:1, tripling the density generation over generation, enabling 6 petabytes in a compact 3U form factor. And we've extended PowerStore scale-out capabilities to include transactional file, enabling customers to consolidate even more workloads. Let me be clear on this. Power Store Elite is the gold standard. No competitor in our industry can deliver this level of density, performance and functionality in a single product. And it's not just about the capabilities that we deliver today. Power stores, container-based architecture means that everything is modular and upgradable which means that it will evolve as workloads evolve and adapt as new technologies emerge. PowerStore is the definition of future-proofed. Now -- let me -- and every single Dell private cloud is managed through the Dell automation platform. And today, we are extending it with AIOps driven intelligence and an ingentic layer, enabling customers to proactively observe, manage and optimize their infrastructure. [Presentation]

So freaking cool. The London Stock Exchange, now known as chose Dell to build a new, secure, modern private cloud, integrating our servers, our storage and our automation platform and LSAG using this solution can now deliver capabilities related to data services and trading operations faster than ever before. Now let me bring it all home with the Dell AI factory. We launched the Dell AI factory 2 years ago with a very simple premise, a full stack modular validated architecture the compute, the network, the storage, the data management, the software, the ecosystem, the services, design, delivered and serviced as a single product. Built to support workloads at the edge in your data center and in the cloud. And the adoption speaks for itself 5,000 -- over 5,000 customers have deployed Dell AI factory. In the last year alone, you've seen 160-plus product releases. And perhaps most importantly, customers are seeing first year ROI of 269%. And the model is working. It's working exceptionally well. And today, I want to introduce you to the next generation of infrastructure that will power the Dell AI factory starting with the compute and let me begin with the workhorse of the portfolio, the recently authored signed 9812 built on NVIDIA's virorubin-NVL72 platform. This absolute piece of a platform delivers 10x cost per token reduction, gen over generation and 260 terabytes of GPU to GPU throughput, which is more bandwidth than the entire Internet. The new lineup also includes 3 new liquid-cooled PowerEdge servers built on NVIDIA's HGX NVL 8 supporting up to 144 GPUs in a single rack and delivering 5.5x the performance of the 200 GPUs. And as we've said all along, AI requires a heterogeneous mix of accelerated and general purpose compute. And this week, we are announcing the broadest refresh in PowerEdge history with our 18th generation portfolio including a new series of ultra-performance servers supporting CPUs from NVIDIA, AMD and Intel built for high core count density and memory bandwidth to support everything from databases and virtualization to AI workloads. The entire portfolio, 1 common platform of management. the entire portfolio post-quantum cryptography compliant day 1. Now compute alone will not solve you need a high-speed network. The most powerful GPUs in the world require a communication super highway. And agenetic technology will tax the network like never before. Agents exchanging context, scheduling, making decisions in real time, your network must be lawless adaptive and fast where GPUs will stall. -- and a stall GPU is an incredibly expensive proposition. To build this communication superhighway, we are expanding the power switch portfolio to include NVIDIA Spectrum 6 and Tomahawk and Broadcom Tomahawk Ethernet technology, enabling 409.6 terabit per second of switch capacity with co-packaged optics. -- which will enable 5x more power efficiency 10x more reliability and 5x more uptime in AI workloads. We are also introducing NVIDIA Quantum x800 InfiniBand solution liquid-cooled co-packaged optics enabling 10,800 gigabit per second post connections and a 9x boost in performance. But network hardware alone is not enough. -- your network operating system must be purpose-built for AI. It must be high bandwidth and low latency. 20 years ago, Linux became the standard operating system for servers. It was open multi-vendor and a global community continue to improve it. The same thing is happening today. Sonic is to networking as Linux was to servers. And we have enabled the entire power switch portfolio, we have enabled Sonic across the entire power switch portfolio, giving customers access to the same high-performance fabrics that hyperscalers use with enterprise-class support in life cycle management. And now that we've talked about a lot of the individual components, it's important to not lose sight of the fact that bringing -- that the complexity in bringing all of this together is immense. -- and we make it simple for customers. We design and validate the full stack, the compute, the network and the storage as 1 system. The rack is now the product and this engineering and design philosophy, coupled with the expertise and capabilities of our services organizations means that we can deploy rack scale infrastructure in under 6.5 hours and maintain uptime of 99.9% and the market has taken notice. IDC ranked #1 in rack shipments last year, shipping more than 2.2x the number of racks of our nearest competitors. The Dell AI factory is the proven path from AI investments to business outcome and the model is working. Hudson River trading trades millions of shares a day across 200-plus financial markets deployed their AI research platform on a Dell AI factory with NVIDIA. Power scale as the underlying data foundation, coupled with liquid cooled power edge GPUs, the results HR's AI research now keeps pace with the markets in which they serve. SanDisk is another great example. They went all in building an advanced AI and generative platform on adult Aitor with NVIDIA and and the results were simply amazing. Factory costs were down 32%. Energy costs were down 46%. CO2 emissions were down 45%. Defect rates were down from 800 parts per million to 100 parts per million and factory operation -- light hilt factory operations was up to 95%. Ladies and gentlemen, this is what good looks like. In closing, Jeff gave 5 imperatives. Number one, build a strong data foundation. We've built it. Number two, distributed AI training and inferencing at the edge of the core of the cloud. We've built it. Number three, secure autonomous systems, we've built it -- number four, a full-stack solution. We've built it #5 in agent a full stack solution ready for an agentic world, we've built that, too. The Agentic enterprise is no longer a vision. It is being built right now and Dell is leading the way. Thank you for your time this morning, and Boss, back to you .

Arthur Lewis, everybody. So home stretch. I'm going to get a home quickly here. 3 years ago, we found ourselves the thousands of shadow AI projects running inside Dell. People across the company experimenting on their own with whatever tools they could find. It wasn't a failure of governance. It was actually a signal of demand. So we got disciplined. We identified 5 use cases with real outcomes. We created an AI office to kind of gather it all together. We went to 90-day sprints that turned into 60 days prints that turned into 30 days fronts. My challenge to the team is now 3 days heading towards 3 hours. Today, our service assistant is alive across our global services organization. It's actively closing cases. It's reducing dispatch rates. It's lifting customer satisfaction. All deployed on-prem, all on Dell infrastructure, all in an existing data center. No additional power, no additional cooling -- the ROI was in less than 3 months. And our current infrastructure deployed at scale agents, ROI in less than 3 months. and every observation I shared with you today showed up our own data. We Blue pass every forecast we had for token consumption. We pivoted and are now running our agent workloads on our own Dell AI factory because the economics demanded it. We can name our super users, a small group of engineers and operators who became force multipliers and reshaped how we're thinking about AI adoption entirely in the company. And we've restructured how the company runs in real time. We're mapping workflows. And even more important, we are tracing the data path and the logic that powers our operations from the supply chain, to services, to sales to software development, all of it. This is not a pilot. This is how our company is running today. So let me leave you with 3 goods. 3 things I really want you to think about as you exit the building, so to speak. One, go look at your AI cost model, ask honestly, whether you're budgeting for both compute and token consumption. If you can't answer that, you're going to have a very uncomfortable conversation with your CFO in about 6 months. Number two, go find your super users don't count them, go find out what they do, study them, build alongside them. The gap between your super users and everybody else is the gap between your future and your past. And then lastly, decide on whether you're going to lead the operating model change, the disruption or be organized by it. Both options are in front of you. but only 1 of them is for yours to choose, I believe. So that's the modern enterprise. It's not a destination. It's a permanent operating model. We're doing the work ourselves. We're working with thousands of customers right now the opportunity is massive. Let's get to work. Thanks, everybody, for joining us today.