AI Proving Ground Podcast
AI deployment and adoption is complex — this podcast makes it actionable. Join top experts, IT leaders and innovators as we explore AI’s toughest challenges, uncover real-world case studies, and reveal practical insights that drive AI ROI. From strategy to execution, we break down what works (and what doesn’t) in enterprise AI. New episodes every week.
AI Proving Ground Podcast
What Cisco Live 2025 Revealed About the Future of Enterprise AI
At Cisco Live 2025, the networking giant rolled out a sweeping agenda to make AI not just powerful, but practical — and secure. In this episode, we caught up with leaders from Cisco, NVIDIA and WWT to talk about what this year's announcements actually mean for enterprise teams tasked with building scalable, secure, AI-ready infrastructure. From the rise of the Cisco Secure AI Factory with NVIDIA to the reality of agentic workflows and persistent inference traffic, this episode unpacks the architectural shifts reshaping the modern data center.
Learn more about this week's guests:
Kevin Wollenweber is Cisco's Senior Vice President and General Manager of Data Center, Internet, and Cloud Infrastructure. In this role, he leads product strategy to enhance Cisco's infrastructure solutions for the data center, high-performance routing, and mobile networks. His leadership is pivotal in driving growth and developing cutting-edge solutions to meet the dynamic needs of businesses worldwide.
Kevin's top pick: About Cisco and WWT
Chris Marriott is the VP/GM of Enterprise Platforms at NVIDIA, where he has spent 14 years advancing enterprise solutions. With a background in engineering, including 10 years in ASIC development, Chris combines technical expertise with strategic insight to address the evolving tech landscape.
Chris's top pick: About NVIDIA and WWT
Neil Anderson has over 30 years of experience in Software Development, Wireless, Security, Networking, Data Center, Cloud and AI technologies. At WWT Neil is an VP in our Global Solutions and Architecture team, with responsibility for over $16B in WWT's solutions portfolio across AI, Networking, Cloud, Data Center, and Automation. Neil advises many large organizations on their global architecture and IT strategy across Global Enterprise, Global Service Provider, and Public Sector. Neil is also on the advisory board of several high-tech companies and startups.
Neil's top pick: Building for Success: A CTO's Guide to Generative AI
The AI Proving Ground Podcast leverages the deep AI technical and business expertise from within World Wide Technology's one-of-a-kind AI Proving Ground, which provides unrivaled access to the world's leading AI technologies. This unique lab environment accelerates your ability to learn about, test, train and implement AI solutions.
Learn more about WWT's AI Proving Ground.
The AI Proving Ground is a composable lab environment that features the latest high-performance infrastructure and reference architectures from the world's leading AI companies, such as NVIDIA, Cisco, Dell, F5, AMD, Intel and others.
Developed within our Advanced Technology Center (ATC), this one-of-a-kind lab environment empowers IT teams to evaluate and test AI infrastructure, software and solutions for efficacy, scalability and flexibility — all under one roof. The AI Proving Ground provides visibility into data flows across the entire development pipeline, enabling more informed decision-making while safeguarding production environments.
This year's Cisco Live, held last week in San Diego, was called the most consequential Cisco Live in the past decade, marking a pivotal moment in steering enterprise infrastructure toward agentic AI and secure-ready networks. While Chuck Robbins and Gitu Patel spotlighted Cisco's commitment to AI-ready data centers and a unified, security-first approach to networking, reuters highlighted Cisco's deepening role in the AI boom and analysts called Cisco a hidden, sovereign AI play which may sound exclusive to governments, but it's the same infrastructure enterprises need to safely scale AI adoption. In this episode, we'll talk with Cisco Senior Vice President and General Manager of Data Center and Provider Connectivity, kevin Wollenweber, nvidia Vice President and GM of Enterprise Platform Solutions, chris Marriott, and Neil Anderson, my colleague here at WWT, who leads our cloud infrastructure and AI solutions teams. Kevin, chris and Neil will cut through the headlines from Cisco Live and there were a lot of them to reveal what really matters Operational simplicity, scalable infrastructure and AI-native network and security stacks built for a future run by autonomous agents, because networks are fast becoming the connective tissue of enterprise AI and Cisco is laying the tracks.
Speaker 1:This is the AI Proving Ground podcast from Worldwide Technology everything AI all in one place. Let's dive in Well. Neil, chris, kevin, thank you so much for joining the AI Proving Ground podcast today. I know your schedules are absolutely slammed out there at Cisco Live, so thank you for joining yeah thanks for joining us, yeah, thanks for having us, absolutely.
Speaker 1:Great to be here, kevin. It's your show here. Why don't you kick us off? Tell us you know, give us an update on what are some of the announcements that are taking place there at Cisco Live and you know what's piqued your interest so far there on the show floor.
Speaker 3:Yeah, I think the most interesting thing for me has been if I kind of contrast this to last year. I think last year there were a lot of ideas about either things we wanted to do in the AI space or kind of announcements of potential futures, and what we're really starting to see now is actual real customers, real use cases, real adoption and just the momentum that we're seeing behind. Everything we're doing in that AI space is exciting to see, and so, you know, we announced new switches to enable us to drive higher bandwidths as we build out some of these AI fabrics. We announced a bunch of stuff with NVIDIA. We're kind of taking the reference architectures that they're building for AI factories and evolving those into secure AI factories with Cisco, and so I don't know, it's just it's an exciting time to be here.
Speaker 1:Yeah, and Neil, I'll get to you here in a second. But, chris, you know, what do you see from NVIDIA's perspective? What's getting you excited in terms of what's advancing the agenda from an AI perspective?
Speaker 2:Yeah, yeah, absolutely, I think. You know, to Chuck's point, I think, to Chuck's point, I think it might be the most important Cisco Live yet, right, and I think the undertones of the entire show, through everything from compute and software and networking, obviously was skewed heavily towards security, right, and I think, in the new age of AI, that underpinning of every enterprise is trying to figure out how they go deploy AI and how they deploy all these new workloads and you know, one of the things that could prevent that adoption is security. So I think all the security related announcements that we saw at the show was fantastic. So that was my takeaway for sure.
Speaker 1:Yeah, neil, I mean certainly lots of buzz going around in Cisco Live about AI, cyber. You know certainly infrastructure plays a major part there, but you know what's actually resonating with you in terms of what it means for our clients' ability to push forward their AI journeys? What are you hearing that you know that feels most relevant or urgent for our listeners to kind of understand or know about right now, kind of understand or know about right now.
Speaker 4:Yeah, first of all, I would say like I have never seen the speed of announcements that Cisco has been doing. I've been working with Cisco or for Cisco for 25 years. Never seen anything like what we saw this morning. Just the speed and the number of announcements. It's just absolutely incredible. And these are. You know, g2 made a point of saying this is not stuff, that's futures. This is now right, which I think is urgent for our customers. And I do agree with the security piece. That can really stall an AI project in a real hurry If customers are not comfortable with the security governance that they have over what they're trying to achieve. And I think and the third thing I would say, brian, is the seamlessness with which Cisco and NVIDIA are working together. It almost feels like one company, sometimes like they're just they can finish each other's sentences, and so that level of collaboration is just helping both companies go faster in the market for our customers.
Speaker 3:Well, and actually, if I could tag onto that, I think, one interesting thing the pace is there and we're talking about a lot of new technologies, but I think things like this are important because we're announcing so many things this year. I don't think we've had this many things to announce in any Cisco Live that I can remember, and in doing that I want to make sure that the message is understood and people kind of understand the why behind what we're doing and how we're actually executing, especially as we look at, you know, partners like NVIDIA and partners like WWT as well.
Speaker 1:We'll dive deeper there, kevin. You know what is. What is that kind of singular message that you know that it all boils down to?
Speaker 3:Yeah, it's. It's about a lot more than just a singular piece of technology. You know, what we're really trying to do is enable the pace of innovation that we're seeing in AI to be consumed by a larger set of customers. We've all collectively been working with the big hyperscalers and the model builders and the people that are doing a lot of the training of these models, but the real kind of future that we see is inference and usage of the models and, as that moves deeper into sovereign networks and into enterprises, security, ease of use, being able to turn these technologies up quickly and make sure that they're confident and comfortable with the safety and security of them is of the utmost importance to the enterprises that we talk to.
Speaker 4:Yeah, I'd build on that, Kevin, that you know AI where you run. Ai is not a product. It really takes an architecture and an ecosystem to pull that together. It's a rather complicated stack and very robust capabilities in that stack, but I see Cisco and NVIDIA partnering to try to make that a lot simpler for customers to understand that complicated stack. The NVIDIA software on top of that is a bonus, because now customers can actually accelerate what they're doing and get to a use case that's delivering outcomes for their business a lot faster.
Speaker 1:And so I've been impressed with the way that the architecture is coming together. Yeah Well, chris, just a few months ago we had met Neil, kevin and I we met with one of your colleagues, kevin Deerling, and that was while we were at NVIDIA GTC, which, of course, did not disappoint. So many great announcements, whether it's Rubin, dynamo, blackwell, ultra, but I am curious, between then and now, what roadblocks still exist or stand between that vision that we saw and heard about at NVIDIA GTC and what the reality is within the enterprise setting. How do you see Cisco Life building on what we heard there?
Speaker 2:Yeah, I would say a few challenges, right. So I think, as we start to move towards accelerated compute with all these new GPU platforms like obviously, even with Blackwell, but going into Rubin, for enterprise, I think the open question is like the return on investment for AI and the investment and how it's going to give like business outcomes. It's, it's obvious, right, and there are so many places that enterprise can go. Do that. You know. The one thing is they've got to figure out and you know Neil was teaching me about this just yesterday, right, you know Neil was teaching me about this just yesterday, right, they've got to go figure out the top one or two use cases and stay very focused about, like the ROI, that that use case is going to deliver number one, but then number two when they actually go to bring in this infrastructure.
Speaker 2:You know, a lot of on-prem enterprise data centers aren't necessarily architected for the power or the scale of compute, and so I think we see, you know, over the next five years I think it was something like a 50x increase between NCP, csp and enterprise of infrastructure and data centers that are going to be required, and so I think part of it is a data center build out plan. Part of it is the improvements in the enterprise data centers, then the third part of it are potentially like co-location centers. Then the third part of it are potentially like co-location centers where you know partners or channel partners or anybody can go land that equipment, stand it up and be kind of fully still managed, tied into the enterprise's data, and so I think that that that's a solvable barrier, but it's obviously something that you know the between all of us. We have to have a good plan and get in place before we get there.
Speaker 3:What I love about some of this stuff you were talking about the ROI and kind of the use cases that the customers are going to pick up what I love about this is when the use case clicks, it just like light bulbs go off. So you know, when DJ did his demo of AI Canvas and he just kind of showed a real example of how we can take these agentic workflows and have a queryable assistant and actually do troubleshooting workflows across networks, you look at it you're like, oh, that's amazing, and it's this ability to build composite applications from different sets of data that we never could really do before. And that's just one really small example. But to me, when it clicks, the light bulbs go off and it's an easy sell.
Speaker 4:Yeah, that use case, gavin, I'm really excited about because I just met with DJ and Anand a little bit ago and I'm super excited about that use case because we've been delivering that as a bespoke, you know use case for our clients. The idea of a NOC agent, a NOC co-pilot, with the two of you you know your company's underneath the hood but we've been building that out for some of our clients and it's kind of challenging because we have to start almost at zero with every one of those right and they kind of work as well as the customer has data to power them. But I think this idea of then putting the Cisco developed deep network model underneath that man, that's going to be really cool. That's going to make our the not copilot kind of applications that we're delivering, kevin, for customers. It's going to make it so much better today than it is today.
Speaker 1:Yeah, neil, can you clarify a little bit about why the use case drives so much considerations with the infrastructure and why it's important to kind of lead with that use case first, so that you know what you're building for?
Speaker 4:Yeah, I mean use cases. I put them into two buckets. There's horizontal use cases that everybody has the same problem, I don't care what vertical you're in. And then vertical use cases I don't care what vertical you're in. And then vertical use cases and what we have found with several of our very large clients is this idea of a NOC co-pilot something to make their NOC agents that much smarter to be able to resolve problems much faster, like that.
Speaker 4:Just you know, when we talk about that with customers it really gets attention. Everybody can identify that, everybody who has an IT network operations center or even a security operations center, which they're also excited about for the future what this could do for them. It's just a use case that really really resonates with our customers. But what's lacking there a little bit has been this we're building those use cases on top of general models, the publicly general available models. They only have so much network smarts in them, right, and then we can ingest customer ticket and resolution data which makes them smarter. But I'm excited about, like, the idea of building that on a purpose-built model from Cisco that has decades of network experience in it already. It's going to make the things that we're building on that so much more accurate for those not co-pilots, so we're really excited about that one.
Speaker 1:Yeah, I do also want to get to the fact that perhaps there's or not, perhaps there is a lot of work to be done in the data center. We hear a lot about AI ready data centers. There this week at Cisco Live, I also hear, or we hear, a lot about AI factories. I do feel like they can bleed into each other a little bit. So I was hoping you know, Chris or Kevin, can you take a swing at telling me you the relationship between an AI factory and an AI-ready data center.
Speaker 2:Yeah, I mean, maybe we both can, but I can give you our perspective, because I think the term has been, let's just say, used frequently, maybe overloaded. All of our partners have AI factories. Nvidia has an AI factory. What is an AI factory? I get asked quite a bit and so we're starting to, you know, at least talk about these things a little bit, and so I would say what we build out with, or what all of us build out with our partners in, like CSPs and in NeoClouds and sovereign AI dentist centers, that aspect of it.
Speaker 2:I think we now consider more AI infrastructure and some of our CSP partners and everybody will bring their software stack and their value add and we'll build those things even in the cloud and offer kind of AI factories for even on-prem customers in some cases. But we really believe that the enterprise is going to become like the AI factory and the idea being an enterprise can take, you know an open model, a cutting edge model, bring it into the enterprise, fine tune it with that business. You know critical data for that particular industry, their business, their use case, and then now they have IP built into that model and so now you have this flywheel of train, train, train a model, get it deployed into inference, you get new data, you still fine tune it every night or every other night, and you become this AI factory. And so I think there are, you know, plans even for large AI factories that are really you're going to have AI factories in every kind of vertical.
Speaker 2:So in manufacturing, in retail, all these places where you need tokens. Some of these places will need their own AI factories to generate those tokens locally. Some will use them from different areas, and I think that's really where Dynamo also can play a big part. And so Dynamo is NVIDIA's open source tool for essentially disaggregating and taking the inference workloads and steering them to available GPUs. And so you take the appropriate GPUs for the pre-fill stage, where you process the prompts, and you take other GPUs for the decode phase and it takes like a very spiky workload, potentially at inference, and you're able to batch those processes in so to really take advantage of the usable compute. But I think that's the distinction. In Sovereign you have AI infrastructure. Enterprises are going to be AI factories, and I love what Cisco has done with AI defense, with the secure AI factory, and I think that's going to resonate heavily in industry.
Speaker 3:Yeah, I think that was exactly the point. When we started talking about AI factories was, as we see, adoption from enterprise, and one of the big things that they need to figure out how to do is how do I integrate that into my existing data center and ecosystem in a simpler way and realistically? How do I get visibility and understand safety? Security that sits both around the models and around that AI factory itself, and that's kind of where this concept of secure AI factory came from is. We know that our enterprise customers are going to want to build these AI factories. We know that we have some pretty interesting technologies from a security perspective inside of Cisco and we can take things like AI defense and build a blueprint or an architecture that allows customers to deploy that simply, easily and securely in their networks.
Speaker 4:Yeah, and I think the other thing I would add into there you know sort of comparing the AI factory to the AI ready data center is yes, of course there's tremendous energy in building AI factories, but customers also have an existing data center that they need to run and they need to manage that. This idea of Cisco being able to bring that together for customers and make it simple to run not only your AI workloads but your traditional workloads and have kind of a seamless backend to be able to manage that, that's what I think of the AI-ready data center. That's what really compels me and I think compels clients.
Speaker 1:Yeah, neil, stick with you here. Whether AI factory or AI-ready data center, it all hinges on a tightly integrated IT infrastructure security operations. How can enterprises or how are we advising clients to efficiently scale that infrastructure to accommodate the demands you know that are just growing and growing for this new wave of AI innovation?
Speaker 4:Yeah, I think this is where it's really important to understand. You know, obviously there's hyperscale customers that are building AI factories and they're building you know tremendous net new models and can you know sort of constructing those. There's also these neocloud providers that are building tremendous data center investment around being able to offer you know kind of an AI, you know factory, as a service and those are tremendously important. But when you get to an enterprise data center where our clients are asking us, okay, that's great, but because of the nature of my data, I need it very secure. It's my intellectual property. I want to run this in my own data center.
Speaker 4:That's a very different type of AI stack, I think, than either the hyperscalers or the neocloud providers. Customers want to leverage their existing skill sets. I've got expertise in Cisco networking in my data center. I've got expertise in Ethernet in my data center. I want to use a storage vendor that I'm familiar with and I already have expertise in. How do I bring that together? And that's where I think the partnership between NVIDIA and Cisco is super important, because Cisco is trusted already in those data centers. They're already supplying that and so I think the ability to then bring the innovation with NVIDIA to those same clients. It's absolutely critical for worldwide's clients.
Speaker 3:And over time, I actually expect to see this, this expand. So today we're talking about specific use cases and, you know, wanting to deploy small ai pods or a small ai factory inside of their data center. But over time, you know actually nvidia talks about this a lot as well ai is not going to be a thing. Ai is going to be a part of everything that we build, and so you need to be able to modernize data center, get that ecosystem ready to consume these technologies and over time, I think it will become the most broadly deployed application across everything that they build in a traditional data center.
Speaker 2:Yeah, for sure. And like building on top of that, I think, the like ease of deployment and like the simplicity of like monitoring and managing and all those pieces of the AF factory. I think that's really, as you know, I learn and learn more about like Cisco's offering with like HyperFabric as well, to be able to give you kind of a single pane of glass almost between all of that infrastructure and the same tools tied in, even for Cisco networking that is already running in all of these enterprises. I think that really is, you know, it removes a barrier from IT when they're looking at either new infrastructure, new workloads, those kinds of things.
Speaker 3:Right, that's actually something that I don't think a lot of people recognize is, if you think about a traditional data center operator, most of the components that we're bringing in AI and AI infrastructure and secure AI factories are the same components. You know, you've got compute, you've got networking, you've got storage, but they operate together as a system, and so you know you have multiple networks. You've got a front end network and a back end network, and so, even though the components are the same, the way you operate it is different, and what we're trying to do is just find ways to simplify that. Let them get infrastructure deployed and focus on the cool stuff that comes with all these AI applications and not. You know how do I tune back in rocky parameters to build an efficient fabric, and I think we've really found something there.
Speaker 1:Yeah, certainly love to see the tight, you know the blooming partnership between you know, not just Cisco and NVIDIA, but WWT as well. Chris, you had mentioned Dynamo and Hyper Fabric At GTC. I thought it was interesting. Jensen build Dynamo as control software, as the operating system, the AI data factory. So in regards to Dynamo and Hyperfabric, where does Nvidia's Dynamo stop and Cisco's Hyperfabric start, or is it all just intertwined at all times?
Speaker 2:Oh, yeah, yeah, no, it's a great question, right? I think the way to think about it is and again, my understanding as well is like you're going to take, you know, once you have Cisco infrastructure at the base, you're going to build up that infrastructure with Hyper Fabric to be able to deploy, monitor, manage, you know, configure all of that infrastructure and really it's going to be like dynamo running on top of that to deliver kind of the inference workloads to available gpu is kind of like inference orchestrator. So they're very much intertwined, uh, between the two lives. And and yeah, denimos, as we've seen, like when you, when you put it, especially with blackwell, when we've moved now to fp4, but even with hop, the speed ups generation over generation. Once you add Dynamo into the mix for that AI factory, it's pretty stunning. Just what like that simple piece of open software. So I think, combined with the two, it's going to be excellent.
Speaker 3:Yeah, I was about to say.
Speaker 3:What's really interesting to me is we've been building out a bunch of these fabrics and starting to see customers deploy this, but what I struggle with, or what I think a lot of enterprise struggle with, is, as they start to roll these out, scheduling of workloads and how they manage those across GPUs.
Speaker 3:So I actually think that integrating something like Dynamo on top to be able to efficiently schedule across these clusters they're building is one of the biggest gaps we had in some of these enterprise deployments. And we have, I would call them, small clusters relative to the size of like the hyperscalers and what they're building. But we've got, you know, 1000 or so GPUs and clusters inside of Cisco and we end up having to carve out resources for teams and give them those clusters and we don't know how well utilized those clusters are. And so having a much more dynamic approach, especially in inference, where you're not running these you know days, weeks, months, long type type of jobs I think that's the critical missing piece, and so it's going to be really interesting to see how we integrate that together with stuff like Hyperfabric.
Speaker 1:Yeah, we're talking about integration here. That feels like your wheelhouse. What do you see from the integrations? What considerations do our clients or listeners need to think about as they're doing that?
Speaker 4:Yeah, I mean it does take a complicated stack to pull this off and there's a lot of different moving parts there. Things like hyperfabric are going to simplify that. Dynamo is going to make it more efficient. I do agree with both these gentlemen that those two working in concert are going to be super important, right, and I think it's going to bring a lot of value to clients. But you know and Kevin, I think you hit on this that while it could be the same components, you're talking about network, compute storage and accelerators and orchestration at the end of the day, how you operate, that is very different You're talking about. You know we have built some super pods with NVIDIA and you know you're talking 8,000 cables in one of these. And it is not your grandfather's old mobile, right? It is a little bit of a different beast to build that. We know how to build it because we've been working with NVIDIA for quite a while. But that tight integration between you know compute storage, network and being able to bring that to life at the speeds we're talking about here is super incredible.
Speaker 4:The other thing I like about what Cisco is bringing to the table not only AI defense, but also HyperShield this idea of being able to spread enforcement points out for scale, I think is hugely important. You know the idea of coming back, and you know bringing all your traffic back to some central point is what I call a choke point firewall strategy. That's not going to scale here. You need to rethink the way that security is deployed in the network, and that's what I'm a big fan of Hybershield, because I think it's the right architectural direction. Forget that Cisco's product name for a minute. It's the right architecture for the security that we're talking about at AI speeds. I don't know of another way to architect that and so I think that those kind of integration points are going to be huge for our clients. Brian.
Speaker 3:Yeah, I mean, I just love the idea of like reforming the tools and using the AI tools to protect the AI infrastructure. To me, that's just an amazing concept. But, exactly to your point, we're used to managing users with access to resources in a data center and now that you're getting into these agentic workflows, you're we're talking about 1000s, hundreds of 1000s, millions, billions of agents at some point, you know, flying around all over the place. We can't, we can't actually scale the support and the and the security behind that in traditional ways, and so we've got to think about modernization of the tools that we're going to use to protect these massively growing applications like agentic workflows.
Speaker 4:Yeah, and I think, Kevin, I was talking last night with a couple of Cisco fellows. They may be bringing an idea to your desk that we were thinking about last night around this idea of agent-to-agent communication and how do you secure those workflows at scale. So you'll probably be hearing about that pretty soon.
Speaker 3:Oh, I love that. I mean, look at in the last couple of months, the number of agentic communication protocols and technologies that we've seen enter the market. And so I think this is that next frontier is, as agentic workflows grow, making sure we have standardized ways for agents to communicate and then, honestly, standardized ways to make sure that they don't misbehave, because think about how expensive these GPU resources are, and if agents are running unchecked and they're taking resources on or doing malicious things with resources, that actually becomes another type of DOS attack or another attack vector in the network. And so, as amazed as I am and excited as I am about agentic workflows and AI in general, we also have to make sure that all of the security and monitoring and other technologies are keeping pace, because, you know, we're creating potentially new areas for attack surfaces.
Speaker 2:Yeah, and I think your point about, like agentic controls because once we start having these agents, you're almost going to have to be able to control each agent's access to information in the cluster as well right, like you know, we've all heard of when customers first get started with AI and they had just pointed the entire company's organization of data at a chatbot and suddenly you can you know, you can query like HR files and things like this, and so that access control for agent is going to be, yeah, very critical to uh to secure like enterprise data as well that's how I explain this, this ai defense up, when I'm trying to really really dumb it down and simplify it is just think about these models, as you're the world's smartest 10 year old kid who has access to every piece of information in the world, and it was always told you know, don't take candy from strangers.
Speaker 3:And it's got its rules that it's going to follow, but as soon as you get it outside of those rules, it has access to everything and it can tell you everything about every piece of data that it has. And so I think there's a lot of investment that has to continue to drive in this space and continue to evolve as we evolve these types of workflows.
Speaker 4:Yeah, this idea of you know models don't have any concept of RBAC right. They don't have access control kind of built into them. The model knows what the model knows and if you have access to the model, you have access to all the data. So the need to build something on top of that to actually control that access a bit more and I think we're also gonna see, we're already seeing it this idea of a massive supermodel with it knows all kinds of things, I think you're going to see a shrinkage of like much more purpose-built models so that, if for nothing else, you can kind of control the access to the different data sources a little bit more granularly.
Speaker 3:Yeah, and the funny thing is, yes, that might actually mean less GPU resource for that particular model, but it means that we can actually run a ton more models and we can use AI for things in the cost effectiveness of AI that barrier for the ROI that we were talking about earlier as that bar goes down, we can actually use it for everything, and that's why I think it's just become we're not in two years, we're not going to be talking about AI and how we, you know, put this little pod in over here to run AI. It's just going to be ubiquitous and be part of everything that we do.
Speaker 2:Yeah, and I think that was you know. I think the initial reaction from everyone when DeepSeek, you know, first dropped, is suddenly like, oh my God, test time scaling. Suddenly, you know between infrastructure, gpus, everything it's going to take, you know you can run this on your cell phones and it's going to crash the market. We're actually, you know, between going from Hopper to Blackwell to Rubin. We have like orders of magnitude improvement in the number of tokens we can process. But I think once you reduce the cost of tokens and the cost of AI, that's really where the hockey stick and everything starts to take off, because then you don't have to really, oh, like you know, consider how much you know. I should only apply it here, I should only apply it there. We want the cost of AI to drive down to the floor so everybody can use it in every application.
Speaker 4:It's classic GVON's paradox, right yeah?
Speaker 3:Well, look at what we just launched with your RTX Pros in MGX. You know PCA-based platforms, and so now we have a more cost-effective power-efficient and sort of scale-out for inference, because you're just going to be running these workloads on single GPUs and we can bring that to a much larger section of the market.
Speaker 4:Yeah, I love what G2 showed on stage this morning too. Kevin, around you know, like the, if you think about chatbots, it was kind of this bursty traffic right in and out of the GPUs to do inference. But when you think about agentic AI, it is this persistent demand for tokens and persistent results that's coming out of the models persistent inference. I think we've already seen that in the agentic AI that we've built at Worldwide. We've seen exactly that slide was like holy cow, like I'm stealing that, because that's exactly how I need to explain to customers because we've seen it every day.
Speaker 3:Well, I would love to partner with you, and Chris you guys as well on just exactly how we're seeing this rollout of networks, because that's anecdotally what I've been saying, and I do think that the networks that we build have to change in terms of how we operate them. I think we're going to see a lot more consistent traffic like that. It'd be great to get a real view of this, because I think, for those that are sort of dipping their toe in the water or wanting to move towards it when we talk about AI-ready data centers, that's a lot about what I mean when I say AI-ready data centers. Let's prepare for this wave that's coming, even if it's not something you're deploying today, and I think that's a perfect proof point of it.
Speaker 2:Yeah, yeah, totally agree the fact that for agentic workflows now you have agents all over the cluster, potentially in different data centers, where you have to connect them like super high speeds with low latencies. And then, even if you take something like Dynamo, where you are creating the KB cache, large KB cache as far as the, the prompt calculation, Now you've got to send that KB cache to the other computers to the decode. You're shipping a vast amount of data across the network and not to forget the huge storage array of data that you have there, all that compute going and hitting and creating track of traffic over as well. So the network is going to be critical.
Speaker 1:Yeah Well, obviously you know a ton to go over here, Neil. You know so many moving parts. How are enterprise AI teams validating that their AI infrastructure choices are going to hold up under real world conditions, knowing that there's just so much movement going on all at once?
Speaker 4:Yeah, and what I tell customers is look, you do not have to go this alone. We have a tremendous lab investment at Worldwide that we call our AI Proving Ground Lab, where we have these architectures, including Cisco Secure AI Factory with NVIDIA, built with HyperShield, with AI Defense from Cisco. These are already built. You can come into that lab and we can start exploring the art of the possible together. What model do I want to use as a basis? How is this going to scale? How many GPUs do I need? Can I use a storage partner that I already have experience in? There's all sorts of startups trying to tackle deepfake detection. Which one works? We're doing all those kind of studies today with customers in the AI proving ground. They do not have to go it alone. We can help accelerate that journey. You get your hands on and let's actually get to the outcomes faster together and get you on your way.
Speaker 1:That's our motto with the Proving Ground. Yeah and Chris or Kevin, not to put you on the spot to make a pitch for the Proving Ground, but I am curious what value do you see that type of composable lab environment offering the industry in general?
Speaker 2:Yeah, I can start. I mean like honestly, like what WWT does with the AI proving ground becomes kind of like the tip of the spear for enterprise customers and so the fact that they invest early in new architecture and have, like the Cisco's secure AI factory, stood up and be able to like bring customers in and test their actual customer data on those things, it's like gold, because without that it's you know you get, you fall back into the typical six to nine month buying cycle that enterprises have, not to mention the time ahead, just to evaluate AI, to figure out sizing all of those kind of things, to evaluate AI to figure out sizing all of those kind of things. So what WWT does in terms of that lab is just.
Speaker 3:It is invaluable. In my opinion, yeah, and for me, that's exactly what the AI proving ground is. If you think about you, go back to building networks and other things. We've been doing proof of concepts for years, but unfortunately, the cost of building out a large AI factory to do testing and validation it's not something a lot of our customers can actually do. And so being able to do it in the AI proving ground on real hardware, get real results and then make that business case to help with the ROI problem that we talked about earlier is one of the only ways we're going to get to do this, because not everybody can go out and build a large AI factories just to say, all right, let's put this in the corner and try some things and hope it works. And so the ability to both partner with WWT, cisco, nvidia, go in and actually test with real models, real GPUs and real data and then identify what those use cases are that want to go and drive. To me, this is the only way we're going to get this done.
Speaker 1:Yeah, kevin, stick with you here for just a moment. Beyond integration, beyond validation, I think a lot of times we'll hear from clients that they're looking to future proof, at least to the best extent that they can, and that seems like an incredibly hard thing to do these days, considering how fast things are changing. From Cisco's point of view, we know it takes an ecosystem, but how can we make this all work together so that clients aren't locked into a single solution for an unbearable amount of time?
Speaker 3:Well, I think Neil hit it on the head when he talked about it earlier. That's why a lot of enterprises are looking at how do I integrate this in with technologies that sort of look and feel and operate like the technologies I'm deploying today. They have large fabrics today, they have compute deployed. Today, they have storage partners and obviously if there's a differentiation in a storage partner or an architecture, we'll suggest that. But what they're really looking to do is extend their existing architectures as much as possible. Let their operators that know certain tools, certain technologies. If they have Cisco Fabrics deployed, if they can just extend that out and add some of these AI technologies in, it really makes the expansion easier. It makes the business case easier and allows them to move a lot faster.
Speaker 1:Well, you know, just in terms. You know we're running short on time here and appreciate the time that the three of you have given us here on today's show Any closing thoughts? I know Cisco Live isn't done yet. Us here on today's show Any closing thoughts? I know Cisco Live isn't done yet, but what are the key takeaways from the time you have spent so far on where our listeners should be focusing for the next three, six, 12 months?
Speaker 4:I would say something that we tell clients is you cannot afford to be on the sidelines of this. You've got to get started because it's going to take learning, it's going to take investment. It's going to take a little bit to get that flywheel effect going. You know what I tell clients all the time is you know that first use case is going to be a bit painful. It's going to be all new right You've got to get. You've got to figure out the security, the governance model, the investment to get going. But what we like to do at WorldRide is come in and help them build a flywheel. So the third, fourth, fifth use cases are much faster and actually a lot cheaper to get going and before you know it you've got a couple dozen use cases that are in production. That's really our goal with customers. But what I think about is like you cannot afford to be on the sidelines waiting for something to happen. It is here and you need to get going yeah, chris, any parting thoughts here?
Speaker 2:yeah, well, I think I think there, there, there's uh, maybe two or three things. Right, I think one thing that we we left off, that I I think is going to resonate in every enterprise and deliver intelligence out of existing data that's in your enterprise, I think is a super, super valuable point that a lot of enterprises are going to pick up on. And the second, I think, is I've heard it anecdotally a few times at the show is like, in addition to enterprises trying to figure out, well, what's my use case, should I get started? Maybe you know, to Neil's point, not also not being worried about failing if it's not the first use case that you try.
Speaker 2:But one of the pieces that I've heard is that, well, we're not sure if we want to deploy on this generation, because you know it may. Maybe we wait until the next generation because, like you know, I'll say right, our roadmap moves very fast in the high end, right In the enterprise range. I think we are a little bit slower and enterprises should not be scared of deploying with what technology is there today, because they can take that same use case and if they say, bring in Blackwell or bring in Blackwell Ultra in a year or whatever it's going to be, it's simply a performance upgrade for part of those workflows and that infrastructure will live for a long time in your data center. So I think, yeah, don't be afraid to get started to fail, and then don't let the architecture and the pace of innovation slow you down from getting started.
Speaker 1:Yeah, and Kevin, what are we going to be talking about? What are you seeing now and what do you think we're going to be talking about, kind of this time next year?
Speaker 3:the models themselves are also improving, and so he was talking about, you know, don't let's not put artificial harness around things or try to fix things within the model, outside of the model, because the models get smarter and better every single day. And so if something is, you know, 80% effective today, tomorrow it's 90%, then 95. And then you know it does everything we need in the future. And so Neil's point about move fast and, chris, your point about Neil's point about move fast and, chris, your point about let's just get started with what we have, I think resonates with everything we're seeing.
Speaker 3:I think next year, if I look at the pace of innovation, I think next year you know, we talked about things like AI defense and we talked about hyperfabric. We talked about this AI canvas. I think next year we're going to have so many whether they're agentic workloads or just new uses of AI real proof points. I think we're going to see a lot more real customers. We definitely brought a few customers with us in the journey this time and you're seeing them on stage, you're seeing them talk about it, but I think every customer we work with is going to have proof points and use cases that show real value for them next year, so I'm excited for that.
Speaker 1:Yeah, no, absolutely. It's going to be an exciting 12 months until we connect again. And on that note, I'll let the three of you go. Thank you guys so much for taking the time out of the busy schedule. Hopefully you get some rest. I know you're probably not out there it's a busy week but thanks again for sharing the insights and all the knowledge you have.
Speaker 4:Thanks, pleasure to be here. Thank you, brian.
Speaker 1:Okay. As the dust settles on Cisco Live 2025, one thing is clear the age of theoretical AI it's over and the enterprise deployment era is here. From today's conversation, three key lessons stand out. First, ai use cases must lead infrastructure decisions. Whether it's a NOC co-pilot or an agent-driven troubleshooting assistant, the most successful AI initiatives begin not with a tech selection but with a clear, outcome-focused use case. Second, ai infrastructure isn't just about GPUs. It's about integration. The Cisco, nvidia and WWT trifecta is pushing beyond hardware to offer composable, secure and manageable AI-ready data centers. And third, security must scale with AI adoption. As agentic workflows proliferate and inference demand becomes persistent, enterprise needs security models that can handle billions of machine-to-machine interactions. In the end, the message from Cisco Live could not be clearer Don't get stuck at the starting line. Start now. Start small and build with partners who understand the complexity of making AI real and safe at enterprise scale.
Speaker 1:If you liked this episode of the AI Proving Ground podcast, please consider sharing with friends and colleagues and leave a rating or a review. And don't forget to subscribe on your favorite podcast platform or watch on WWTcom. This episode was co-produced by Naz Baker, cara Kuhn, mallory Schaffran and Stephanie Hammond, and a special thanks to the teams at Cisco and NVIDIA, as well as Tori McLeod, sarah Chiodini, rod Flores and Diane Devery. Here at WWT, our audio and video engineer is John Knobloch. My name is Brian Felt. We will see you next time.
