Keynote by Antonio Neri – Architecting AI starts with your network

Good morning. Wow. That's a big room. It is great to be here with you. HP Discover is where we showcase what is next and shine a light on the ambition and innovation shaping our industry and our lives. Today, we are witnessing one of the largest technology platform shifts in history. Workloads and applications are moving from being driven by end users, but now being driven by both end users and AI agents. Agents that will fundamentally transform how we design and build, how we serve our customers, and how we operate our businesses. I have always been drawn to how things work, how systems are built, and how they evolve over time. In fact, if I had not become an engineer and a CEO, I would have become an architect. Architecture like engineering teaches you to think in systems, to build for today and for the needs of tomorrow. You don't design a building around a single room. You design a structure that allows the whole system to flow and adapt over time. Architecting for AI demands the same focus and discipline. Fundamentally, AI is only as strong as the data foundation beneath it. If the foundation is not robust, nothing else holds. Across networking, cloud, and AI, HP is delivering the essential building blocks that make your AI-ready foundation possible. Networking to connect your infrastructure and workloads at scale. Cloud to enable you with a hybrid operating model to run your workloads and applications where they belong. And AI to turn your data into intelligence and put it to work. Architecting for AI starts with your network. For years, HP Aruba Networking has helped you deliver secure connectivity across campus, branch, and the edge, creating a digital on-ramp that connects your users, devices, and data. With the addition of Juniper Networks, we have extended that leadership into the data center and across the critical networks connecting the AI era. Scale up, scale out, and scale across. With our combined HP Networking organization, our goal is to deliver the best user and operator experience possible. We will do this through our next generation of secure, self-driving networks across every domain. They fix problems securely before they impact the experience. We make new rollouts faster and easier, and fundamentally transform how you manage your network. Whether you are on the HP Aruba Central or HP Mist, you get the full benefit of our accelerated innovation on both. Nobody is left behind on the road to self-driving. But the phrase self-driving may have caused some confusion with the Mercedes Formula 1 drivers. Let's take a look.

I had the opportunity to talk to both drivers this past weekend and they told me they had fun doing that video, but the reality is that they really are very, very keen and interested to understand technology. So, congratulations to the Mercedes team for a fantastic year so far. Every AI architecture needs more compute. In the AI era, your servers need to operate more efficiently than ever. With HPE ProLiant Gen12, you get the performance to run everything from enterprise workloads to inference in a much smaller, more efficient solution. As AI moves from generating content to taking action, the demands on compute are increasing. Agentic AI requires fast orchestration, continuous evaluation, and real-time access to data, which shifts more pressure to the host CPU. That is why we are expanding our ProLiant portfolio with the new HPE ProLiant Compute DL394 Gen12, powered by NVIDIA Beta CPUs. Beta provides the low latency memory access, bandwidth, and coherence required for agentic AI, reinforcement learning, and other CPU intensive workloads. It does so with the security and management you expect from ProLiant. As AI moves beyond the data center, compute needs to move with it. We recently expanded our ProLiant Edge portfolio, bringing secure AI-ready compute to rack and distributed environments, so that inference can happen closer to where decisions are made. AI also needs to reach the always-on systems at the core of digital business. With HPE NonStop, we are bringing AI-powered fraud detection and autonomous operations into high-volume payment processing. This helps financial institutions detect fraud faster, automate compliance monitoring, and keep transactions moving safely. The most advanced servers in the world are only as valuable as the data they can access, which makes storage a critical part of your AI foundation. The HPE Alletra Storage MPX 1000 makes your data accessible, context-rich, and ready to continuously feed your AI data pipelines. Across the full lifecycle from ingestion, training, inference, and continuous learning, the X1000 now supports native file and object storage on a single architecture. It is also the first object storage platform validated through NVIDIA certified storage for enterprise AI. And for the mission-critical applications that run your business, the Alletra Storage MP B1000 continues to deliver. The B1000 is the fastest-growing all-flash block storage array in the market. Sitting across the top of the stack, of course, is software. As AI advances, enterprise environments are becoming more distributed and complex. The spine, virtual machines, containers, AI infrastructure, and public and private clouds. At the same time, rising virtualization costs are pushing many of you to look for more flexibility and choice. With HPE Cloud Ops software, we have brought together HPE Morpheus, OpsRamp, Zerto, and a broader cloud portfolio capabilities into one unified operating experience. This enables you to modernize on your own terms while simplifying how you provision, observe, and protect your hybrid multi-vendor environments. Security and resilience must be built into every layer of the stack of your architecture. As AI changes the speed and the scale of cyber threats, new risks continue to emerge. Resilience is no longer just a set of isolated tools. It is a converged strategy across your systems, your data, and your network. With HP iLO Silicon Root of Trust, we provide secure attestation from the silicon to cloud, helping verify that your infrastructure is trusted before your workloads run. And with Zerto and Cyber Resilience Vault, we help protect your critical data so you can recover faster and minimize disruption. Networking and security are also converging. As AI becomes more distributed, the network often is the first place to see what is happening across your enterprise. With zero trust architectures and integrated SASE, the network becomes an active security layer, enforcing policy, detecting threats, and reducing risk from edge to cloud. We bring all the elements of your AI foundation together with our GreenLake cloud. GreenLake gives you a unified cloud-native experience across your entire hybrid estate with the flexibility to run workloads across public and private clouds, colos, and at the edge. With GreenLake intelligence, we bring intelligent AI to hybrid IT operations, helping you see across environments, act faster, and continuously optimize performance. From simplifying network operations to streamlining virtual machine migrations, GreenLake intelligence makes your infrastructure more adaptive, more autonomous, and easier to manage. So you can spend less time managing tech and more time managing and advancing your business. So, let's take a look.

It's an amazing advancement we brought with HP GreenLake. But, by the way, what you saw here is just a beginning. In the Discover Showcase, you can see how GreenLake Intelligence brings agentic AI operations to life. So, I encourage you to go and experience it yourself. Architecting for AI takes more than technology. It also requires the right people, processes, and partnerships. Our services team is here to help you through your AI transformation. With HP Financial Services, we help you modernize with confidence and better economics, including lower upfront capital investment. And with our IT lifecycle management program, you can retire legacy multi-vendor technology and turn that value into funding what is next. This week is an opportunity to explore our full-stack AI foundation firsthand with demos, sessions, and conversations with your peers. Discover is one of the few moments where we have the full power of the HP community together in one place. During the rest of our time this morning, I want to elaborate on two core tenets of our strategy. First, the networks that are the heart of today's AI data center build-outs. And second, how we are enabling your transformation into an agentic enterprise. In the AI era, the network is both the essential enabler and the main bottleneck for performance. Nobody understands what is at stake more than our customers who are at the leading edge of AI. Customers like Vultr, the world's largest privately held hyperscaler. Let's take a look at what HPE and Vultr are building together.

Thank you to the entire Vultr team for being such a great partner. I'm excited to share that this partnership continues to expand as we work together with Nvidia to support Vultr's next phase of growth. What Vultr is building at hyperscale points to a truth that every architect knows. That there is always one core element of your infrastructure that touches everything. With AI, that core element is the network. The performance of your entire architecture depends on it. Every byte, every token, every decision, all of it crosses the network. Which is why today we are bringing the HP Juniper network into our AI data solutions, enabling more efficient, high-performance AI environments. Whether you are a hyperscaler service provider or a neo cloud or a large enterprise, you have more choice in how you connect and secure your largest AI investments. Let me show you how it all comes together starting with a model training use case. An AI data center is designed around one purpose, turning data into intelligence as quickly and as efficiently as possible. At this scale, performance depends on how tightly compute and networking work together. For customers building AMD-based systems with Helios, we are introducing the industry-first HP Juniper networking scale-up switch, purpose-built for the AMD Helios architecture. The QFX 5250 brings scale-up performance into an open Ethernet fabric designed for AI at scale. It connects 72 GPUs into a single rack, delivering 260 terabytes per second of aggregate scale-up bandwidth. You get the low-latency performance required for large-scale AI workloads with the openness of standards-based Ethernet, SONiC OS support, and Juniper AI automation. But one rack is just the beginning. The largest models are trained across hundreds, even thousands of racks operating as one cohesive cluster. Multiply a small delay across hundreds of thousands of GPUs over weeks of training, and your network can mean the difference between training a new model in 90 days or 30 days. Think about that. It is the difference between chasing a breakthrough or making one. That is why the scale-out network is so important. The Juniper QFX family is built for this next generation of AI scale-out connectivity. Our newest addition to the portfolio is shipping today. The QFX 5250 is the world's highest-performance 100% direct liquid-cooled ultra Ethernet transport-ready switch. Powered by Junos OS, it moves data across massive AI clusters. It achieves this through low-latency congestion control and operational simplicity required to keep hundreds of thousands of GPUs working together. Increasingly, AI data centers are expanding beyond single sites. They span multiple data centers and regions, sometimes hundreds or even thousands of miles apart. That puts new pressure on the network between these environments. That is where the HP Juniper PTX routing family excels. PTX is built to carry massive volumes of traffic across the data center interconnect, core, and edge networks that connect today's distributed AI infrastructure. Our PTX 12000 series is an ultra-dense routing platform designed specifically for AI fabrics. It enables 800 gig routing, 1.6 terabit ready scale, and ZR plus coherent optics to connect data centers across sites without compromising performance. And to protect that connection, we have our HP Networking SRX family, including our most popular firewall, the SRX 4700. It is one of the fastest quantum-safe firewalls on Earth, delivering up to 1.44 terabits per second of security performance in a single rack unit. It helps you secure modern data centers without slowing down the applications and AI workloads you depend on. But when we talk about a complete portfolio for AI training, we support every layer of a modern networking architecture. From scale up and scale out to scale across and secure data access. All in a single coherent architecture that is secure and fully automated. With the introduction of the HPE AI grid with Nvidia at GTC in March, we extended this integrated network even further. Built for service providers, the AI grid combines Nvidia accelerated computing and AI networking, including Spectrum-X, ConnectX, and BlueField with ProLiant compute and Juniper routing, security and unified orchestration across the full stack. Together with Nvidia, we are enabling a wide range of new real-time AI services from conversational agents and interactive media to hyper-personalized experiences across hospitality, healthcare, and retail. But the real value of AI increasingly comes from inference. When intelligence moves closer to your users, applications, and data. That requires a network built to extend AI to edge locations like regional data centers and service provider sites where the Juniper MX family of edge on-ramp routers is top of the class. Our MX 301 brings the proven performance and flexibility of the MX family into a small form factor, a 1RU power optimized platform. It is purpose-built to move AI inference out of the cloud and closer to where the data is processed for inferencing so we can accelerate decision-making. Powered by Juniper's sixth generation Trio silicon, it has near infinite flexibility to meet your networking needs today and into the future. To build your inference environment, you also need high-performance switching. That is why today we are introducing the new HP Juniper Networking QFX 5140 inference switch. Purpose-built for distributed AI deployments. Also in 1RU, the QFX 5140 delivers up to 16 terabytes per second of switching capacity, connecting GPUs and inference infrastructure with AI optimized load balancing and end-to-end congestion control to maximize performance. The 5140 gives every edge location the local intelligence to host AI workloads closer to where inference is needed for faster AI responses and better experiences. Look, the bottom line is to win in the AI era, you need a network built for the full AI lifecycle from training at the core to inference at the edge. AI is also transforming the demands of the campus and branch networks. They still need to securely connect people and devices. But now they also need to support AI powered workflows that depend on real-time access to data without compromising speed, security, and reliability. That level of complexity cannot be managed through reactive troubleshooting alone. Your network has to see more, understand more, and do more. Self-driving networks move IT from reactive troubleshooting to proactive assurance, understanding experiences, identifying root causes, and resolving issues faster. In the race to self-driving, HPE continues to lead. In fact, we were recently recognized as a leader in the Gartner Magic Quadrant for both wired and wireless LAN for the 20th consecutive year. Positioned highest in execution and furthest in vision. What matters most is what these capabilities mean for customers like the Milano Cortina Winter Olympics where HP helped deliver flawless network performance across a very complex environment spanning 15 venues hundreds of miles apart. HP Mist adapted the network in real-time, helping ensure seamless secure connectivity for everything from broadcast, live streams to event operations, and fan engagements. Every moment could be viewed by millions. I hope you watched the Olympics while organizers operated with confidence knowing the self-driving network was working behind the scenes to maintain rock-solid performance. Today, we are extending this self-driving experience across our Aruba networking portfolio with two new announcements. First, Aruba CX switching is coming to HPE Mist. You get AI-native assurance, faster troubleshooting, and automated operations across your campus and branch environments. And second, which is what we talk about cross-pollinating, right? With Ravi. Marvis actions is coming to HPE Aruba Central. Marvis is the first network assistant in the industry to bring conversational AI to networking. So, your network can move from reactive to self-driving with AI-native operations that are continuously improving. So, you can see how much progress we have made with Juniper in such a short period of time. We are really proud of the progress we're making for you, our customers and our partners. But, across industries, customers are making the switch to HPE networking and discovering that the self-driving network is a quiet network because it just works. That is the experience we want every one of you to have. If you are considering a change from your current networking provider, you know who they are, I encourage you to start with a single site or even a single floor. Experience what a self-driving network can do in your environment. You will be amazed at how simple the experience is. And I will ask you to not miss Ravi Shahrabi's general session later today, plus our four networking spotlights throughout the week to see how our self-driving capabilities are coming to life across every single domain. We have talked about the networks that make AI possible and how HP is delivering the next generation of self-driving networks that are self-healing, self-protecting, and self-optimizing. Now, let's turn to the next major shift in the AI era, the rise of the agentic enterprise. AI is no longer just a tool for finding answers. It is a critical part of how work gets done. Agents now reason across data, applications, models, and workflows to help you make decisions, automate processes, and are increasingly taking action on your behalf. Soon, IT will be responsible for thousands of agents that are part of your enterprise workforce, operating across every function. But today, much of that innovation is still happening in local clients, in the hands of developers, and small teams, often outside formal IT oversight. That speed of adoption is exciting, but also creates a real challenge, the shadow cost of an agentic workforce that now must be managed at scale we have never seen before. Agentic AI demands a new set of enterprise requirements. Agents need to be secure and governed with clear guardrails for what they can do, what systems they can act on, and most importantly, what data they can access. They need to be trained with trusted enterprise data because the agents are only as good as the data and context behind them. And they also need infrastructure that can scale as demand grows without runaway costs. When we introduced our HP Private Cloud AI 2 years ago, we gave enterprises a turnkey AI factory that simplified AI adoption and provided more control. It brings AI to your data, not the other way around. So, today we're enhancing Private Cloud AI for the next generation of agentic workloads, helping you govern agents, ground them in trusted data, and scale your inference initiatives. Let's unpack what is new starting with agentic governance. You can now register agents built in any framework and wrap them with security controls that protect API calls, identity, and encryption with zero code changes required. A new three-tier identity model verifies the user, governs the agent, and enables human approval for sensitive actions. Today, we're also announcing new capabilities for secure agentic operations with NVIDIA Open Shell and Nemo Cloud. Open Shell provides a modern acting runtime for advanced private AI agents with policy enforcement built into how agents run. Each agent operates in its own isolated environment with guardrails for what data it can access, what systems it can interact with, and what actions it can take. And with Nemo Cloud, you get an open-source reference stack and blueprints to govern agentic workflows, helping you move faster while maintaining the control and accountability enterprise AI requires. As agents operate across production environments, they also introduce a new class of operational risks. And that's why we are bringing Zerto to your agentic enterprise. If an agent makes a mistake, Zerto helps you quickly roll back to a clean state, reducing downtime and helping you protect your business. Governance in your agentic enterprise is paramount, but governance alone is not enough. Your AI agents are only as smart as the data you use to train them. Traditionally, that data required custom preparation for every use case and months of building the right AI data pipelines. But not anymore. Private Cloud AI helps make that data you already have ready for agentic AI. With a governed data layer and integration with the NVIDIA AI data platform, you get a unified way to access, prepare, and manage enterprise data across your existing environments. Now, with the latest Storage MPX 1000 as the storage layer for Private Cloud AI, you can build on a high-performance data foundation designed for modern AI. The MPX 1000 adds real-time metadata enrichment and native MCP support, so your agents and applications can retrieve the right data and context faster across structured and unstructured data. That means less custom integration work and 7 to 12 times faster time to value compared to what you normally do, which is yourself building the whole environment. Once you have governed agents and trained them with the right data, it is time to scale across both agentic AI and your broader enterprise inference workloads. Private Cloud AI can now serve larger models across multiple systems with multi-node inference, so capacity grows with demand. A new unified gateway simplifies access to frontier and open-source models. This gives your team one unified API for model access with centralized credentials, budgets, and policies. We're also expanding Private Cloud AI with new configurations that scale up to 256 GPUs, including the new ProLiant DL394 with NVIDIA Beta CPUs designed specifically for inferencing. And for long context workloads, we're also adding capabilities that reduce the need to recompute context over and over. This delivers significant cost benefits to first token and massive performance gains in compute capacity. With Private Cloud AI, you now have the foundation to build your agentic enterprise with confidence. And what makes it stronger is the ecosystem we have built around it. We continue to expand the HPE Unleash AI program, our curated ecosystem of validated partners, blueprints, and orchestration frameworks for Private Cloud AI. With more than 60 partners and hundreds of use cases, Unleash AI helps you find trusted solutions for scaling AI across your enterprise. From securing agents and models with partners like CrowdStrike and Fortanix to expanding where you can deploy Azure Digital Reality and Equinix Private Cloud AI, the Unleash AI ecosystem helps you move from AI ambition to real-world impact faster. Take examples of that. For St. Jude Children's Research Hospital, that means bringing AI closer to doctors and researchers, accelerating life-saving discoveries while protecting highly sensitive medical data. Or for Blue Star Operations, the business behind the Dallas Cowboys, it means reducing lower value work streams and advancing strategic decision-making across football and business operations. And for the Ryder Cup, it means being able to power real-time event intelligence from crowd management and concessions to volunteer assistance and operational planning. In fact, the Ryder Cup organization is now leveraging Private Cloud AI to turn the next event into a massive success by really using a digital twin approach so they can help architect the 2027 tournament experience. But look, these are just a few examples of how we are giving you a faster, more structured path to AI adoption that will transform how you run your business. And there is more to come. Tomorrow, Fidelma Russo will share additional news in her CTO general session and go deeper into how our latest cloud and AI innovations help you build a new operating model for your agentic enterprise. AI today is about moving faster from ambition to outcome, accelerating time to talking, reducing execution risk, and ensuring your environments are ready to perform from day one. Our AI factory solutions are designed to do exactly that with validated architectures, agentic operations, and enterprise grade support. They also meet you where you are, designed for your unique operating models, governance needs, and scale. For enterprises, I shared how Private Cloud AI is a secure and governed pre-packaged AI factory for your agentic enterprise. For model builders, service providers, and neo clouds, our AI factory at scale is built for large multi-tenant AI environments. And for governments, regulated industries, and sovereign entities, our AI factory for sovereigns enables you to deploy AI aligned to your local data, security, and compliance requirements. Across our AI factory portfolio, our deep collaboration with NVIDIA helps you build on the latest accelerated computing platforms like NVIDIA Vera and Vera Rubin. NVIDIA Vera CPUs in our latest ProLiant servers are powering agentic workloads across enterprises. In supercomputing, NVIDIA Vera and Vera Rubin architectures are advancing our Cray portfolio for both HPC and AI. And in AI factory at scale, NVIDIA Vera Rubin and VL72 is driving the next frontier of rack scale solutions. Compared to NVIDIA Blackwell, Vera Rubin and VL72 delivers AI training with 1/4 of the GPUs and AI inference at 1/10 of the cost per million tokens. So, think about that, the massive gains you can get to get to that token faster. So, whether you are building for the enterprise or training frontier models, HPE gives you a path to build and scale on the latest NVIDIA accelerators. As AI scales across more users, more data, and critical operations, trust must be built into that foundation. That is why we are making confidential computing standard across the full HPE AI portfolio, helping protect sensitive data, models, and workloads while they are in use. With NVIDIA confidential computing, AI workloads run in trusted execution environments that add a hardware-protected layer of security across the stack. And for organizations operating in the most sensitive environments, we are taking that trust foundation even further. Our sovereign AI factories now include defense-grade security hardening, federal compliance readiness, validated encryption standards, and global data protection requirements all built in. So, if you're in defense, government, or financial services, this is the sovereign AI architecture you have been waiting for. Architecting for AI requires looking ahead, anticipating and designing for the constraints that will shape the future. But, there is one challenge we all need to overcome, not just for our industry, but for our society and our planet, and that is power. Every model, every workload, every agent depends on power. Because at its core, an AI factory is doing one thing, turning electrons into tokens. The US is on track to have a 19 gigawatt power gap by 2028. That's roughly enough electricity to power 16 million homes. And data centers are expected to account for nearly half of the US electricity demand through 2030. One customer, Siemens Energy, is tackling this challenge head-on, helping build the energy infrastructure the AI era requires. They are doing it by applying AI to their own business, with HP helping deliver the AI foundation across networking, storage, and compute. Let's take a look.

I want to thank the Siemens Energy team for working on such an important challenge. We're proud to support your ambitions. Initiatives like this underscore a bigger point. As AI scales, the future will not be defined by compute alone. It will be defined by how efficiently we can power it, cool it, and connect it. That is why research becomes so critical. For six decades, HP Labs has helped shape the future of enterprise computing. Your next generation infrastructure will need to operate with much greater intelligence, efficiency, and transparency. Today, our researchers are applying AI to improve AI systems themselves, making them more scalable and more sustainable. This is where HP has a unique advantage. We engineer the complete architecture from compute servers, networking, storage, software, and security. And we apply that system expertise across the full stack. With innovations like GreenLake Intelligence, we are developing predictive self-driving intelligence that can learn world patterns and place data where it needs to be before an application asks for it. Across your broader data center environments, we're using AI to improve resource management, identifying idle patterns, and reduce energy and water consumption without compromising performance. We're also advancing the next frontier of computing through our work in quantum. With initiatives like the Quantum Scaling Alliance and our work in distributed quantum simulation, HP is helping bring quantum out of the lab into the real world. You can see that future taking shape right here at Discover, where a quantum chandelier sits alongside an original HPE Cray 1. It is a great reminder of how far high-performance computing has come and where it is heading next. As networking, HPC, AI, and quantum converge, progress will depend on how effectively we bring these technologies together at scale. Yesterday, we took another major step forward with the announcement of an expanded industry collaboration to advance hybrid quantum. Together with these leading companies, we are building a full-stack hybrid quantum platform that extends our world-class HPC and AI infrastructure and moves quantum closer to real-time and real-world deployment. Quantum advancements like this are accelerating the path to faster, more efficient solutions for the world's most complex scientific and industrial challenges. These are the challenges that inspire our HP Labs. Ultimately, it all comes back to one simple mission, to advance the way people live and work. It is a guiding force behind our innovation, our people, and our long-term strategy. Today, we have covered how architecting for AI starts with your network, and how we can help you transform into an intelligent enterprise. You have seen the powerful outcomes that result from people with bold ambitions matched with the right technology and the right partners. Because none of this happens alone. Our partners help HP bring these ideas to life with the expertise, reach, and execution customers depend on every single day. They help us extend our impact, bringing innovation closer to the customers and communities we serve. Many of our partners have generously sponsored this week. Discover is only made possible because of you, like us, believing in the vision and the power of pursuing it together. So, I want to thank all our sponsors and all our partners. We appreciate you very, very much. We take tremendous pride in knowing that our innovations are a catalyst for your success, driving outcomes that propel new opportunities for you and your customers. As you experience all Discover has to offer this week, keep these things top of mind. First, architect deliberately. The choices you make today will define your success tomorrow. Second, start with a network. Make your network the core foundation of your AI and cloud solutions. And finally, choose HPE as your partner to bring the full stack to help you build your AI future with confidence. This week is an invitation to think bigger, to move faster, to architect the future you want to lead and for the world. And remember, you don't have to build this future alone. We are here to provide the intelligent foundation so you can move boldly, live with purpose, and unlock your ambition. Thank you very much. I hope to see you on the floor. Enjoy the rest of the week.