General Session by Rami Rahim-The next generation of networking: From vision to self-driving reality

It's time to stop running today's business on yesterday's network. It's time for a purpose-built platform built for one thing: delivering an exceptional user experience. This is the self-driving network from HPE. Not legacy tech playing catchup, not bolt-on retrofitting. It's intelligence powered by unmatched user experience data. So it doesn't just see what's happening; it knows how to self-optimize, self-protect, and self-heal. And because it's grounded in the right data, it only continues to get smarter. And let's be honest, not all AI is created equal. Our autonomous agents work together across the entire network. An AI-native infrastructure with security built in from the start. AI that makes the network run better and transforms what the network makes possible. Whatever your business is trying to achieve next, from video calls to AI workloads, the network is here to get you there. HPE's self-driving network. Unlock ambition.

Thank you for joining us at HPE Discover's networking general session. Please welcome to the stage Executive Vice President, President and General Manager of HPE Networking, Rami Rahim.

Thank you. Absolutely. Not only is it the namesake land-grant university, a small city within Columbus, the capital city of Ohio, we have 66,000 students and 8,500 faculty. We're the fourth largest university in the country. And networks matter a lot to us, as you've already alluded to. On our Columbus campus alone, we have 22,000 HPE wireless access points, indoor and outdoor across numerous acres on campus. We have 15 colleges, a hospital, a comprehensive cancer center, and maybe a few of those folks out there know that we also have a football team. So, one that's unique to the hoot and holler that we just heard: we have a stadium, the Horseshoe. Not the Horseshoe Casino — although similar to the Horseshoe Casino, the house always wins at Ohio State. At the Horseshoe, we can host 100,000 participants viewing. That creates a unique networking challenge and fan experience, ticket check-ins, everything being electronic. Last fall we hosted a team from down south and they were ranked number one when they came to Columbus, but maybe they left a little lower. That was the largest broadcast in NCAA history of any sporting event. So there's a lot on the line during an event there; there can be another 100,000-plus people outside the stadium. It's not uncommon for a home game day for us to have 200,000-plus people around that. With the partnership with HPE, as we speak, engineers are working together, collaborating on refreshing our Wi-Fi network there. And that's critically important to us, not just to ensure that fan experience is there. 200 yards away, we have a seven-billion-dollar hospital system, comprehensive cancer center, 2,000 beds in that facility. And that's going on next to 200,000 people enjoying a game. We're putting in, I believe, the single largest implementation in the country: 2,000-plus wireless access points, HPE, Juniper, Mist. Super excited about that.

Oh, absolutely. AI ops is something that we absolutely are excited to have in place. We have 100,000 managed devices where expectations for us on bring-your-own devices is substantial — adds tens of thousands more. Throw in a game like that, we need AI ops that's provided to crunch all of that data in real time. And we've seen this in effect move us from scenarios where problem resolution can take hours — sometimes we just can't afford that with the various operations going on at the institution. So it's literally moving us from absorbing that threat intelligence, applying the AI ops, and giving us answers to resolution. We're seeing things being resolved in minutes versus hours.

Thank you, Rami.

As you said, we are innovating in both platforms to deliver a consistent self-driving experience. This is possible due to two things: microservices and a common agentic AI framework. With microservices, we are able to develop self-driving innovations once and deploy them on both the HPE Aruba Central and HPE Mist platforms, much like a single app experience is delivered on both iPhone and Android. So no matter what agentic AI ops platform you are on, you are going to be enjoying the benefits of a self-driving network. And with a common agentic AI framework, we are able to accelerate self-driving capabilities at a much faster pace on both platforms with trusted actions that deliver measurable value. Our agentic framework is unique in the industry because of these key foundational pillars. First, we use real live experience data — every user, every minute — which HPE has uniquely validated against real customer support cases and enriched with digital twins to maximize the efficacy of our AI-driven insights. Second, we have an API-first approach, which means all the data is available via our APIs. This makes our MCP server and tools extremely powerful, delivering agentic automation at scale. Third, a powerful set of AI agents and skills analyze all the data sets and apply reason — from HPE Marvis Minis, which serve as digital twins of user experience, to agents analyzing packet captures, logs, knowledge-based articles, security vulnerabilities, and all the support data and signals we get. They all work together to proactively diagnose and autonomously root-cause problems impacting users without inundating operators with data. And fourth, models take the analysis and curate it to understand and analyze post-connection issues. For example, Marvis has a Large Experience Model, or LEM, that identifies the cause of bad Zoom and Teams calls and predicts future problems to prevent them when possible. We bring all of this together into the agentic framework, resulting in a system that is continuously observing, reasoning, and analyzing with autonomous actions that optimize user experiences. This is a self-driving network.

Of course. We all know that today's networks need to cater to high-density requirements while also meeting the performance expectations of every user and application. But most networks are not designed to meet intermittent surges in peak traffic. Think of an all-hands in an office building, a crowded classroom during a popular university lecture, or even this very room right now where thousands of you are eager to see the best networking solutions ever. Operators try and accommodate these situations with a combination of static boundary parameters and on-demand changes, but often that just isn't enough. Let me show you a self-driving network powered by real agentic AI and how that handles this problem. If we look at Marvis, we see that all the users in this office building are currently happy. But was this the case all of last week? Nope, it wasn't. So what happened? How did unhappy users become happy? Let's take a look. Last week, Marvis detected that over 6% of user minutes were bad, which may not sound like a lot, but it impacted hundreds of people. Marvis has a self-driving action for dynamically fixing capacity issues. This action was enabled and was able to autonomously fix the problem by enabling dual-band 5 GHz. This reduced the peak utilization from 90% to 54%, enabling a better experience for users that were unhappy. But how did Marvis know what to fix and when to fix it? The unhappy minutes in the past week triggered various models and agents in the HPE agentic framework to reason and analyze and root-cause the problem. For example, one model analyzed the service-level experiences. Marvis Minis agents were activated to test the network using digital twins and other skills went into effect. Based on the reasoning and analysis, Marvis determined that the wired and WAN was not the problem; the wireless network was. But what in wireless wasn't working? Coverage was fine. Roaming was fine. Wireless capacity was bogged down on a few APs that were functioning at 90% peak utilization. This is when Marvis went into full self-driving mode. It changed the RF parameters and validated the service-level expectation of the users to ensure they had 100% satisfaction. Rami, this wasn't manual tuning. This wasn't trial and error. This was a network optimizing itself to deliver the best user experience. And this is available right now in HPE Mist today. Like right now. Today. Right now.

Absolutely.

Well, let's find out. We heard from our Mist customers that Marvis actions — which automatically identify network issues and proactively resolve them — is mission-critical to daily operations. So that's why HPE Marvis, our industry-leading AI engine, is coming to HPE Aruba Central. It has all the simplicity and all the impact that customers have come to expect from Marvis. This is experience-first AI in action, and it is a perfect example of how we are able to develop self-driving innovations once and deploy them on both platforms — Aruba Central and Mist — seamlessly, thanks to our microservices foundation and the common agentic framework. That's how we are bringing Marvis to HPE Aruba Central — right into the global NOC view. Behind the scenes, Marvis does all the heavy lifting: correlating logs, alerts, signals across the entire stack. Your morning cup of coffee view will now showcase end-user impacting issues across wired, wireless, and SD-WAN with recommended actions. Missing VLAN, MTU mismatches, negotiation failures — they're all coming to Central. In addition, what we call the Marvis Trust List is also coming to HPE Aruba Central. These are actions that you can choose to be fully autonomous. When enabled, Marvis not only finds the root cause, it fixes it for you. Imagine a security camera connected to a wired port in a bad state — no camera feed. That's a real problem. With Marvis in self-driving mode, that port is recovered automatically and expeditiously, and the camera is now working again. That is the power of bringing Marvis into HPE Aruba Central: not just more visibility with highly accurate, actionable recommendations, but a better end-user experience thanks to self-driving capabilities.

Yes, Rami, we have made quite a bit of progress. A few months ago, we made a commitment on common hardware by announcing the first dual-platform access point, the 723H, which is now generally available. Within the first year of the Juniper acquisition, not only have we cross-pollinated AI models and agentic frameworks, we have delivered an access point that works with both the Mist and Aruba Central platforms. And on that same theme today, I am super excited to announce that we are doing something very similar with switching. Our world-class HPE Networking CX portfolio, which previously was supported by Aruba Central, will also very soon be supported by Mist for day-zero, day-one, and day-two operations. Let me give you a sneak peek. It starts with a simple scan of a QR code of the CX switch from within the Mist installation app. Once onboarded, the devices show up in the inventory and they are ready to be configured with templates that enable you to centrally define and apply consistent configurations at scale, like pushing VLANs and port profiles to hundreds of switches. With telemetry coming into the Mist platform every minute for every wired client, combined with high-efficacy AI/ML models from Marvis, we can measure pre-connection and post-connection wired SLEs. The successful connect SLE measures pre-connection experiences, and the bandwidth congestion and throughput SLEs give us insights into post-connection experiences. With the SLEs, IT administrators can quickly identify when an experience is bad, including interface anomalies and other issues. But we don't stop there. All the goodness we get from Marvis actions will also be available for CX switches. This includes proactive recommendations for missing VLANs, MTU mismatches, bad cables, and more. And best of all, a self-driving trust list will also be available, starting with the ability to autonomously fix stuck ports to remediate wired clients in a bad state. This enables real-time closed-loop self-healing, reducing mean time to repair, eliminating manual troubleshooting, and delivering great user experiences at scale. This is huge because we all know the network being up is not the same as users having a great experience. So for our CX customers, no matter what agentic AI ops platform you choose, you get simpler operations, actionable and proactive recommendations, and a real self-driving network.

That's right, Rami. The best evidence is the new Gartner Magic Quadrant for wired and wireless access published just four weeks ago. It shows HPE as a leader. And even more so, we are furthest to the right for vision and highest in the ability to execute. In my humble opinion, it is clear who the overall leader is.

Thank you, Rami. Thank you, everybody.

Sure. My name is Marlon Drummond. I run — I'm Senior Director. I run Engineering and Automation, and by extension AI. Fifth largest in North America. Just to sort of lay the context of the difficulty from a threat perspective: about 4,000 endpoints from the SD-WAN and the Edge Connect. But if you took the entire scope of the threat landscape, it's probably double or triple between the cloud, trading, every business line. So it is quite extensive and it's extremely difficult for a legacy bank — and I don't mean that in the sense that truly legacy — but it's very, very difficult to do.

Yeah, it is job number one, I think, for the most part. If you're a bank and you're regulated like we are in 29 countries, you have all of the environmental from a regulatory perspective — you have to deal with every single country. It is a very difficult process. So security for us is job number one. We actually don't have any other job other than protecting our client data. And not only that, it's a blueprint to how we operate. We feel that it's our competitive edge. If you have 13 million clients, you're in lots of different countries. That data in itself forms the basis for decisions. It's our competitive edge. That you protect with everything in our fiber.

Absolutely. We think about this a lot and this is primarily to keep the services up. I know you talk about self-healing, but for the most part in general you troubleshoot at the network layer. That's the first — that's the only place that you can get some immutable evidence to be able to identify what's going on. So we've always lived on that mantra. For the most part how we detect and manage, troubleshoot, protect is all at the network layer. Being able to identify when somebody is knocking at the door, that's very important. And the only way you're going to really see that is whether it be through lateral movement in the network and so on and so forth. So you want to make sure that that side is fully covered. Part of me setting the stage for the 4,000 — at least what we can identify and speak about — clients is the Edge Connect, and that platform using DPI engine helps us sort of create a persona or personality for a user. Anything outside of that becomes an anomaly. And that's something that allows us to now be able to go look and hey, something's going on, but it's part of the intel that we need to be able to shore up our threat intel.

So, as you know, I know it's public knowledge: we are AI-first and AI-everything right now. We've pledged to our invested shareholders that we will generate a billion dollars in revenue just using AI. So we've embarked on this incredible effort — the AI-ification, if you will, of the entire bank. That's a significant effort. The operational side of it I think is sort of table stakes. The operations side is near and dear because that's where we actually can make a difference. We're extracting telemetry, building the harnesses and creating a framework to be able to take that data. If you took the Edge Connect for instance, that's one place where we also collect data to be able to put into whether it be through the SIEM or through some sort of MITRE framework to be able to mitigate or create a threat response. But the whole idea is that this will be AI everywhere, collecting data, mining it, normalizing it, vectorizing it, and using that data to be able to help us solve the ops problems.

Absolutely. And I have to tell you, I hear it a lot from a lot of people. We're sort of that cradle-to-grave kind of bank. We want to make sure that you come in early and stay forever. Thank you so much for joining us. Appreciate it.

Hey, Rami. Good afternoon, Las Vegas. I'm excited to be here.

Well, Rami, it starts with a zero trust approach to network security. This assumes that no user or thing is trusted by default and requires continuous verification of every access request. Successfully implementing a zero trust approach requires five core elements: visibility into all connected users and things, policy-driven orchestration, ubiquitous policy enforcement, real-time detection, and automated AI-driven responses. You do have to have the right mix of protected layers in the network to deliver autonomous protection for the strongest defense possible. And HPE has a full security portfolio which includes firewalls with industry-leading efficacy and performance; NAC with comprehensive access control with consistent enforcement across every type of device or user on the network; SSE with agent or agentless deployment options, supporting a broad set of applications with intelligent routing; and last and certainly not least, SD-WAN with integrated application performance and security, which is optimized for any environment.

That's right. I am very excited to say that we have now combined our Edge Connect SD-WAN and our SSE into a unified SASE orchestrator with one console, consistent zero trust policy, and AI-driven operations for simpler, faster, and more secure connectivity.

Let's check it out. The new SASE Orchestrator is a powerful solution combining both networking and security under one umbrella. Here you see the dashboard page. This is where an administrator would get a quick snapshot of everything that's going on within the environment for the last hour to up to seven days. But this isn't the special sauce. Let's go ahead and check out the business intent overlays. To me, that's really where Edge Connect shines. This is a great example where Edge Connect has been self-driving for many, many years already. It autonomously delivers the best quality of experience while also handling the worst kind of links for any type of application. Now, let's talk a little bit about security. We'll navigate over here and check out the global firewall policy rules. What's great about this feature is the fact that these firewall rules are written here but automatically distributed across the entire SD-WAN branch fabric. Next up, let's take a look at the SSE policy rules. These are the new capabilities that are part of the unified orchestrator. Now, being in Vegas, blackjack is on my mind, but visiting a gambling site on my company laptop, probably not so much. Let me show you how quick and easy it is to deploy a web filtering policy for our ZTNA users. I'm going to type block gambling, come down to the destination and look for gambling, make sure it is set to block, and save it. Now, it's going to push this out. As you can see, successfully added that policy rule. Fairly easy — even a product manager can do it. I'm going to go ahead and apply it. And all at once, we're now pushing out that policy for all of our users. Making these changes is all done directly from that one place. That's what's really exciting about the fact that what used to be done in two different products is now possible in this single SASE orchestrator. Super simple. Now, there's one more thing I want to take a look at, which is the new SASE Copilot, where a user can ask a variety of different questions. We also have some pre-populated ones. What's great about it is it allows them to resolve issues and minimize risk to their environment very quickly. And I want to tell this audience that this is coming out just later this year.

Well, Rami, we are all over it. Our AI-aware firewall lets you safely embrace AI by giving you governance over how AI is used, with real-time visibility, number-one-rated efficacy and threat detection, and simplified security operations. That means that you can protect sensitive data without slowing innovation. So, here we have the Security Director Cloud, which manages our SRX hardware and virtual firewalls. Up at the top, we have a new capability — the Security Director Copilot — which lets our administrator ask different types of questions. For this exercise, I'm going to type: show the latest threats to my environment. The copilot is going to spend a little bit of time thinking, crunching data. What's happening behind the scenes is that we're pulling information from all the SRXs that are in this particular topology. We're also leveraging threat intelligence from our HPE Threat Labs. As you can see here, we get a variety of different pieces of information: usernames, source IPs, destinations. But we really get some real detailed information in terms of things like what the threat is, industries that are being targeted, and what countries. And at the bottom, we also have a great set of recommendations. It's not just about showing the information; it's also being able to help supercharge our administrators. Now for this next segment, let me set some context. Looking at some of the policies we have here, I'm looking specifically around some AI governance rules we've put in place. There are three types of rules: sanctioned AI apps, which are allowed by the organization; unsanctioned AI apps, which is a fancy way of saying we're going to block it; and tolerated AI apps, which are AI apps that have some guardrails. Tolerated is a little more nuanced — it's about having the right types of guardrails in place. So here we have upload protection rules, prompt protection rules, and keyword protection rules. Now let's see this in action. We've switched to the end-user view. Happen to be here on hpe.com, and I'm going to go ahead and try to go to ChatGPT. Immediately I get a block message — this is one of the apps that we've actually flagged as not being allowed. It's an unsanctioned app. Now let's try a tolerated app. I'm going to go over to Gemini and remember those guardrails I spoke about. I'm going to attempt to upload a corporate file, a company report. As we can see, Gemini is struggling a little bit. It's attempting to pull this file, but what's happening behind the scenes is our firewall is going and blocking this. Lo and behold, the upload fails. Now let's try a different example. I'm going to grab some text here that has some unique words like restricted, secret, encrypted. I'm going to say, let's summarize this information, paste that, and submit. Lo and behold, Gemini struggles. It can't process this information because our firewall was doing what it needed to do. But as I mentioned before, this is a tolerated application with specific guardrails. So let's see what happens if I say summarize HPE Discover and submit. Already we see a little bit of a change in behavior here. It's searching the web, it's going to figure this out. And yes, it is our flagship annual conference. As you can see how you would use this technology in a real-world case.

Absolutely. To me, this is what makes this solution so robust, so powerful. It proactively detects threats, simplifies operations with guided insights, and enforces granular real-time controls to safely govern AI application usage. You know what I really like about our AI firewall is that it gives customers the ability to see, govern, and protect how AI is being used across their entire organizations without slowing down their businesses.

Thank you, Rami. Thank you, everyone.

Oh, absolutely. I've been with Sentara Health for 29 years, and during that time I've witnessed and been a part of many major technology transformations. We are one of the largest nonprofit integrated health systems in the US — Mid-Atlantic and Southeast. Over 14 billion in operating revenue, 35,000 employees, 12 hospitals in Virginia and northeastern North Carolina. We have over 200 connected sites with 400-plus total points of care. And we have a health plan division that serves almost a million insureds in Virginia and Florida. It all keeps us busy.

We move massive amounts of data across the organization, and it directly impacts patient care. That's why our network has to be resilient, secure, and always available. Because when data is delayed, care is delayed. Our clinicians rely on real-time data for decision-making at the bedside, and that requires instant, reliable delivery everywhere. Areas like radiology, where images from some of our systems are always pulled fresh with no caching — latency simply is not an option.

Wow, there are a lot of opportunities for AI. One example where we've seen real success is using ambient AI to capture patient conversations. It generates the clinical notes and even helps identify potential care gaps. It allows clinicians to spend less time focused on documenting and more time focused on the patient. But for those kinds of capabilities to work, our network has to deliver that data in real time, reliably, securely, and without latency, so the clinicians can trust it at the point of care.

Well, I could spend a lot of time talking about our initiatives. Healthcare's got tons of them. But to enable them, our infrastructure has to be ready to grow with us. We need to be able to expand our operations and embrace new technologies while maintaining performance at scale. And HPE is a trusted partner that has given us the solutions we need now and into the future. We love the AI capabilities that this gives us. You've heard about some of them today already, but we would love to see that extended across all the networking domains.

Excellent. I can't wait.

Thank you, Rami. Appreciate it.

It is great to be here at HPE Discover, and I'm so excited to show how we ensure our customers' networks deliver the experience that we promise at scale without drowning in complexity. Because you don't need to be a routing expert to manage your routers. With Mist and Marvis AI engine, we make routing operations simpler, easier, and more intuitive. Many of you guys have the horror stories when it comes to launching new applications: the escalations, the sleepless nights, the weekends in the situation room. There's got to be a better way. There is. And it starts with HPE AI-native routing. Let me give you an example. So, it's Friday afternoon and our network operations manager is sitting down at his desk. He's ready to go home. He works for a leading healthcare provider. And on Monday, they're going to do a big launch. They're going to put virtual AI agents deployed into every hospital, office, and clinic and their whole network. It's going to manage everything from patient care to hospital triage. If the network falls short, patient care will suffer and the staff is the one that picks up the slack. So, how can we help our network operation guy with this? The network is ready and everything's connected to Mist. There's just one problem: until Monday there are no users. So let's see what we can do with experience twins. Everything looks green, which is great, but what does all this mean? Effectively, Marvis turned the routers into digital twins generating synthetic application traffic into the network just like a real user would, detecting degradations in real time without a single truck roll or single user being impacted. That's the power of Marvis. So, if we revisit our engineer, he's sitting on his couch, trying to have a good day, and lo and behold, notification: the experience has deteriorated. The latency has increased from 80 milliseconds to over 200. Let's go to Marvis and see if we can fix this quickly before it ruins our weekend or worse, the launch. That red does not look good. But what does it mean? We've got a bunch of failed tests across two separate KPIs. So there isn't an obvious root cause. Let's ask Marvis: why has the latency suddenly increased? And Marvis has a clear read on what happened. It analyzes the network in real time and it gives us an answer in plain English: the latency has increased because the traffic has shifted to a less optimal path, and it gives us the steps to validate and fix the problem. It looks like some configuration changes have removed the preferred route, forcing the traffic onto a backup path. Marvis even has a recommendation to fix it. Well, it's a relief that the routers kept the network up, but let's see what we can do about these SLEs. If we check out the latency view, we can see these failed tests have increased gradually over time. And if we want even more details, we can see the granular spikes in latency. Overall, this aligns perfectly with what Marvis told us. So let's go ahead and look at what's going on with these excessive hops. Since Friday's baseline, we can see two additional network hops. This new route is definitely the issue. So we're going to go back to Marvis actions and see what we can do to fix this problem. And there we have it — the missing prefix. Marvis knows exactly how to fix the issue and how to restore the prefix. So if we click here and look: I approve this recommended configuration change is a good idea. But you guys know what they say: trust but verify. So let's check the actions on those experience twins one more time just to be sure. And everything's back to normal. The latency's restored, the network's recovered, the application's performing as intended. And if I'm ready, I can even take one step closer to self-driving by letting Marvis do this automatically next time. And with that, we're back to enjoying our Saturday and watching the game with no one but us the wiser about the disaster that never happened.

Thanks, Rami. Thanks, Vegas.

Sure. Thanks, Rami. It's great to be here. I'm Ben Croy. I lead global networking for The Walt Disney Company. My team owns network architecture, engineering, and operations with a strong focus on media and production infrastructure. One of our largest production networks to support studios like Marvel, Pixar, Lucasfilm. At any one time, we might have over 200 concurrent productions globally, and a major film could easily generate a petabyte of content. This data has to move quickly, securely between partners, creatives, and departments globally.

For an animated feature like Zootopia 2, we may create the content in a single Burbank location and then regionalize it around the world to prepare day-and-date release in 35 or more languages. So the scale and the global movement of content is constant. And that's what our studio production network has to support.

For one thing, film is made digitally end to end now, driving a lot of change. A few things stand out. We've gone from terabytes to petabytes. Visual effects workloads have grown significantly with over 80% of the movies now relying on some type of VFX. Virtual production environments emerged as well on films like Mandalorian and Grogu. This is much more sophisticated than the green-screen VFX processes of the past. And there's no physical fallback anymore. It's all data end to end.

Well, for us it comes down to delivery and predictability. For our large feature productions, we release globally at around the same time in every country. If we miss that window, there's real financial impact. So, the network must be reliable, scalable, and able to move very large data sets without becoming the bottleneck. We leverage HPE Mist platform for our campus and Apstra for our data center fabrics. We found that these solutions give us more consistency across environments and a platform that allows us to scale.

Well, it's foundational. The network is a critical part of production, but it's ideally invisible. Our goal is simple: we want our filmmakers to focus on story, character, amazing visuals, and sound, not to worry if the network will do what it needs to do.

Two things. Speed and simplicity. That's it. Speed and simplicity.

Thank you very much.

Yes, thanks Rami. And you said it. What's really exciting is that we're bringing the power of AI-native self-driving operations into the data center. Our solutions continuously collect rich real-time telemetry across the network from our switches, routers, and firewalls. That telemetry flows directly into the Mist platform where Marvis AI turns it into deep visibility, operational insight, and automated actions. And what makes us truly unique is that we bring design, deployment, assurance, and operations together into a single lifecycle experience. The result is continuous visibility, automated operations, and self-driving optimizations across the entire data center network.

Let's do it. With HPE Networking, you get the proven self-driving capabilities you already know and trust applied to the data center, too. For example, the Marvis dashboard that Sunnolini showed for wired and wireless actions also surfaces key anomalies for the data center network, and it clearly explains how to fix them. In addition, service-level expectations, or SLEs, which are a key part of the Mist platform for optimizing user experiences, are combined with our knowledge graph to continuously measure network health. We take this a step further with application awareness. We map application flows directly onto the data center network, allowing operators to instantly see the real impact of a switch or link failure. This moves application troubleshooting from reactive guesswork to intent-based decision-making. And to minimize downtime altogether, Marvis AI-powered intelligence enables proactive maintenance with Marvis Minis running network tests that continuously validate network intent and services. Further, predictive analytics monitors system and optics behavior, tracking 30-plus metrics such as voltage, current, temperature, CRC errors, and more. Machine learning models analyze these metrics to predict optics failures before they occur. And now we go even further with Agentic AI. The Marvis AI assistant uses reasoning models and intelligent agents to solve problems the way a seasoned network engineer would. It correlates data from multiple sources — contextual data, switch telemetry, application flows, historical support cases — and then it performs expert-level reasoning, leveraging skills to rapidly identify the root cause and recommend next steps. What once took hours or days can now happen in minutes. And by demonstrating the logic, process, and reasoning used, the agents give operators insight into the solutions being recommended, building trust in the network and allowing it to act autonomously.

Agreed. And we have been working hard to deliver a truly integrated data center infrastructure solution across HPE Networking, compute, storage, and hybrid cloud to enable faster deployments and streamlined operations. We integrated our management capabilities with OpsRAMP, our hybrid cloud observability platform, and Compute Ops Management, our platform for managing and automating server infrastructure. These integrations deliver comprehensive observability, predictive assurance, and proactive issue resolution across server, storage, and now the networking domain. Plus, we have now integrated with Morpheus, our management platform for virtualization and containers. We did this because many times when a virtual machine gets provisioned, the virtual network could take hours or days to provision in the physical fabric. Meanwhile, the app team and the business are waiting. By integrating data center operations with Morpheus, the network and server teams no longer work in silos.

Yes. Let me show you. Here we are in Morpheus, a single pane of glass for managing your entire virtual infrastructure: VMs, clouds, clusters, workloads — all here. First, we connect Morpheus to our managed data center fabric. That's the bridge between the virtual world and the physical network. Now we create two virtual networks, VNET 10 and VNET 11. We provision two VMs, a VM on VNET 10 and another VM on VNET 11. Both are live, hosting critical workloads. Now watch what happens — without anyone touching the network, VNET 10 and VNET 11 were automatically created and pushed to the physical fabric. No ticket, no waiting, no manual configuration, no human error. And the proof: both VMs communicating perfectly across the HPE managed fabric. But here's where it gets interesting. What happens when a VM moves? Let's start a continuous ping between the two VMs. Now we migrate one of the VMs from server 9 to server 15. Watch the packet loss counter. Zero. Not a single dropped packet. The network and security policies follow the VM automatically, invisibly, instantly. We've automated networking and removed silos. This means that you get faster deployment and zero-ouch networking without manual errors. And when you move workloads for resiliency or server utilization reasons, the network and security policies follow automatically.

That's right, Rami. Right here at Discover, we announced that HPE is expanding its AI data center solution to include our QFX switches managed by Apstra Data Center Director. This creates a full-stack pre-integrated solution spanning compute, networking, storage, software, and services, which accelerates AI data center deployments with assured interoperability, scale, and performance.

Absolutely. We continue to innovate in our data center. In addition to recent enhancements to the MX and PTX routing lines, we keep adding new QFX platforms for scale-out and scale-up networking. We recently introduced the industry's first Ethernet-based scale-up solution, the QFX5252, purpose-built for AMD Helios systems. And you can see this product in the demo area right over there. Trust me, you can't miss it. It's massive. It's like the size of a fridge. In addition, we launched the QFX5250, which was the first 100%-liquid-cooled switch using the Tomahawk 6 chipset. Just as we were the first OEM vendor to ship 800 gig, we did it again with 1.6T connectivity. And I'm excited to say this product is shipping now.

Yes.

Thank you, Rami.