4. Building AI-Ready Data Centers with Jeetu Patel, Tom Gillis, Kevin Wollenweber, and Will Eatherton

Jeetu, Tom Gillis, Kevin Wollenweber, and Will Eatherton discuss why the "AI revolution" is fundamentally a networking challenge and how Cisco has evolved its portfolio to serve as backbone for the next generation of enterprise intelligence. By aligning our networking, compute, and security innovations, we are providing our customers with the foundation they need to move from experimentation to production at scale.

Transcript

SPEAKER_01 0:00

Hello again and welcome to Cisco's Inside Product Series. We have some very exciting guests today. So we've got Tom Gillis who runs our data center and security business. We've got Will Etherton who runs our engineering for the data center business. And then we've got Kevin Willenrubber who runs all of our data center product management business, which is everything from hyperscaler to neo clouds to enterprise to service providers and on and on. So welcome, gentlemen. Tom, I'm going to start with you. Thank you. What is happening in the industry as it pertains to AI and infrastructure? Give us the macro picture first.

SPEAKER_03 0:36

Yeah, sure. You know, everyone knows the power of AI, it's amazing. The nature of the applications are fundamentally different than applications that we've lived with for decades. And they're really the way to think of it is they're parallel. So that means that an application doesn't have to fit in a box. And in fact, in you know, both Kevin and Will's world, a lot of these high-end applications, we're dealing with a single AI application that can span more than just one box, more than just one rack, more than one row. GJ, these things run across a whole data center, right? And multiple data centers. So in this world, the network becomes the backplane of these applications. And it's it's kind of akin to what the PCI bus was in the world of a server and a box. The PCI bus is the guts of the system. Now all of a sudden the networks are playing that role. That puts very special requirements on the network. And I think that's one of the reasons why.

SPEAKER_01 1:30

No matter how good the GPU is, if it's not networked, dude, the network has got to be better, right?

SPEAKER_03 1:35

And and uh and so that's why there's so much emphasis on the network. Now we we tend to think of this in terms of what we call a front-end network, which is the network that has lived in a data center for what decades? But when a GPU is talking to another GPU, that's what we call a back-end network. And I think both, you know, Kevin and Will will provide perspective on back-end network, front-end network. How do we turn this so make it easy for customers to be able to deploy these AI applications and not get bogged down in some of this vinucia? And that's I think the Cisco advantage.

SPEAKER_01 2:03

So I'm gonna go to Kevin first is walk us through um architecturally what changes. And um what specifically do we need to make sure that we're uh we're keeping in mind as this architecture shifts in this agentic world?

SPEAKER_00 2:19

Yeah, yeah. I mean, first of all, I think Tom Tom laid it out really well, but a lot of what we build when we build out AI infrastructure are the same piece parts and components to what we've been building in traditional data centers. You have compute and you have network and you have storage, but now you have multiple networks and you have different um capacities and need to access storage. And so you can think of taking the components we had, adding a ton of complexity, and then having to have it all orchestrated and managed and operating like like a single system. And so there's two main areas of investment. One, we have to really focus on the piece parts and the components, and I'll talk about that in a second. But then we also have to focus on orchestration and management and making it simple and easy to use. Because you don't want customers focused on, you know, how does this GPU talk to that GPU and how do I optimize the fabric to make it communicate well? You want them to focus on running AI workloads and getting value out of it.

SPEAKER_01 3:08

Okay, so um so when you go into a customer, how should the customer be thinking? Because most customers are using hyperscalers for their AI workloads. Uh and then now what you're starting to see is these workloads, especially in inferencing, are getting pretty distributed. Some of them are getting local. You might have a desk side computer that's running agents, you might have your own enterprise data center that you're doing stuff. What is the playbook on what an enterprise should be doing to go out and keep the token costs low, the outputs high, and the complexity as low as possible so that they can actually drive this at the velocity they need?

SPEAKER_00 3:44

Yeah, well, it's kind of crazy. Although we've been talking about AI spend for what feels like decades, it's only been the last few years that we've seen this massive increase in spend. And most of that spend is with the hyperscalers, with the neo clouds, and with the model builders training the models themselves. So now we have these amazingly powerful models, and what's evolving is this this world of inference. So people actually using the models. And when you say things like agentic workflows, it's really leveraging these models that have been trained on a certain set of functions and using them to perform tasks. And so that architecture actually looks fundamentally different. And yeah, we've been doing all the training in these massive data centers in cloud and in neo clouds, but a lot more of this for inference is going to move either on-prem in the customer. You talked about having little things on on people's desktops to be able to run agents, and that makes the network even more critical in how they manage and how they operate these workflows.

SPEAKER_01 4:36

So, Will, think about thinking about the back-end network and the front-end network. Um, you are architecting these things for customers and for um our own systems, and you're starting to also see that there's a lot of kind of semi-custom work that gets done in extremely large-scale data centers. Talk us through what your learnings are and what what should customers take away about Cisco as it pertains to the kind of things that you're doing, and why should they be as excited as you and I are?

SPEAKER_02 5:06

So um, first of all, just to hit that we we really are seeing three distinct customer engagements uh today. So, hyperscale for AI, we've been engaged for multiple years now. Um, and in that context, it's a very focused, sort of high performance networking uh where we're going in with our you know Silicon One-based systems, a range of our software, um, and we're been working together with those hyperscalers to really tune and get the highest performance across that. And that's again has been several years. What we're seeing in the last 12 months is uh obviously more engagement both with NeoCloud uh class of customers as well as enterprises.

SPEAKER_01 5:41

And define NeoCloud.

SPEAKER_02 5:42

So NeoCloud would be you know folks who essentially their their business is selling a narrow set of as a service, uh, whether it's bare metal or it's more virtualized. Um, and so they're a smaller sliver of what has been the IaaS uh cloud providers.

SPEAKER_01 5:56

And companies like Core Weave are examples of NeoCloud.

SPEAKER_02 5:58

Trevor Burrus, Jr. Yes. And so and sometimes it's you know pretty coarse as far as like they carve up sets of servers that they will then uh release or allocate for a year, and sometimes and a lot more the higher value is when they can be more um more flexible in how they allocate uh uh in that. So in all of these, you know, one of the key things Kevin mentioned is the the number of potential different networks can be daunting. And so you can have you know what's been our traditional front end really you know stems out of what's been a traditional data centering with a lot of feature and functionality on how do you connect servers to servers, servers to internet, servers to storage. Um that that's the front end. Um the back end is you know been Infiniban in the early days. It's pretty clear in the last 12 months that that is now Ethernet. Um everyone, including NVIDIA, is pushing that. Uh, but we also have management networks, sometimes distinct storage networks. And so, you know, having a strategy around how you're gonna operate across these is is a key aspect and something we uh engage with with customers. Also, what we're now seeing is a new trend, which is as inference is becoming a bigger topic and it's not as simple as the NVIDIA, you know, they've had these reference architectures and superpods, if you've you know seen from GTC. That has been a fair repeatable pattern for the last couple of years, but now in the next 12 months, we're seeing more with the inference accelerators. That is complicating these architectures, and I think the key commonality is going to be Ethernet, is what we're seeing, uh connecting accelerators to GPUs. Uh, also storage. So uh storage to accelerators, storage of GPUs is requiring higher bandwidth. So those front-end networks um are requiring increased bandwidth, is what we're seeing with customers. So you know, thinking about you know, planning these architectures, having flexibility, and the operational simplicity is a key topic. And I'd say that's common across both, even though there's some differences, it's common across enterprises and and NeoCloud.

SPEAKER_03 7:51

G2, you know, one of the things I think is interesting, you listen to these two guys talk, they spend a lot of time with these top-of-market customers that are doing like crazy things. And the learnings that we Cisco have from that experience, we're folding into products to make this stuff usable for the enterprise. Right. And isn't that like our opportunity?

SPEAKER_01 8:09

So actually, let's talk about that, Tom. So take a step back and tell me why a customer would be interested in getting something from Cisco versus a competitor. Why are we different? What do we do that is a structural advantage in the market?

SPEAKER_03 8:25

Yeah.

SPEAKER_01 8:25

And um and how should a customer parse the value that we provide versus someone else provides?

SPEAKER_03 8:31

Yeah, yeah. These guys can probably articulate this better, but I think it comes down to really simply customers don't want different little bits and pieces of network that have to be managed differently. Historically, we've seen that. Remember when like telephones were like totally different networks? Ultimately, Ethernet subsumes it, and customers want a single set of management, a single set of hardware that can run all this stuff. And I think that's our opportunity is these back-end networks, front-end networks, with the programmability of Silicon One, we can address all of these needs with your network. It's not a different diddly bob, and I think that's going to really, really matter in the near future.

SPEAKER_02 9:07

So I have an example uh from the last six months in that. So we were working with uh uh Quant uh beginning of this year in the Cisco Life Amsterdam time. One of the feedback we got was that InfiniBan, you know, InfiniBan, again, very respected technology that everyone's clear now won't be the winner. Um, but uh they did have a feature uh with this partition ID that allowed jobs to have distinct allocation from a tenancy, so you could have both separation of resources as well as some security benefits of that isolation. And the feedback we were uh brainstormed with them, it'd be great if Ethernet had that. You know, and obviously there, you know, there should be nothing that InfiniBand has that Ethernet doesn't have. So we we went and uh architected that um and then we were actually able with Silicon 1 P4 programmability to add that feature in in our Silicon 1 software layer, you know, since February. We then added that into our Nexus OS layer. We then added in a top level for configuration and visibility into Nexus dashboard, and actually at Cisco Live uh recently we were able to show that that feature, and that is actually uh we've had interest both from NeoClouds where they want to have tenancy where the first layer of tendency is my customer, second layer of tendency could be you know jobs, job ID that they're working through. But there's also enterprises that want to have you know some layer of separation. And so this is a very specific example, but I think it really helps illustrate that that you know full Cisco stack um and also the agility that we can move at when we have all those pieces together versus hey, we have to go to another company and work with them on the silicon you know side, uh, or if it's a fixed chip and it takes two years to spend that. So um I like that example.

SPEAKER_01 10:44

So I I I just want to make sure that I I I put some emphasis on what you just said because it's extremely important. One of the big reasons that at least I've seen customers wanting to do a great deal of work with us is we are one of the only um vertically integrated platforms that is a co-designed full stack. And what I mean by that is silicon and optics are the foundation, systems hardware, systems software, platforms for security and observability and data, applications that we build, models that we build, and then uh agents that are built on top of that. That whole thing as a co-design full stack. And what you just articulated was the fact that you were able to deliver that was because of the fact that we actually built every single component of it. So we're almost like the Apple of the infrastructure world, where we build the hardware, we build the software, we build the chip, um, and all of them just kind of are designed to work with each other. And the way that they work well is so magical. And because we work with the largest of the largest customers, which are the hyperscalers and neo clouds and sovereign clouds and service providers, we're able to take all the learnings from there to your point, Tom, and then make sure that you can take it all the way to the enterprise. Um so that's at the highest level. Now let's go into the details. Um what were in your mind, Kevin, because you're the product guy that's setting the direction for our data center, the key priorities that you wanted to go drive on um our entire data center business? And let's specifically talk about enterprise here.

SPEAKER_00 12:27

Yeah.

SPEAKER_01 12:27

So what are we doing in the enterprise? Firstly, what do you see as trends in the enterprise? And then what specifically did you say you wanted to go out and prioritize? How was that delivered and what's next?

SPEAKER_00 12:38

Yeah. I think I think since we're kind of started at the top and moving down, I want to talk about that orchestration and management function and the ability to get up and running fast. So if you think about the biggest concerns that a lot of enterprises have around moving to more and more of these AI technologies and bringing them on-prem, is unfortunately right now, they're still expensive. And so nobody wants to make a decision to bring a bunch of hardware and technology in and then have it sit idle and not be able to run it. And so I think there's some amazing innovation that that we'll talk about in silicon itself and what we're doing there. But in general, what customers want to be able to do is bring the technologies to their premise, get them up and running, and actually be able to show a real ROI for running AI workloads. They're not deploying infrastructure for the sake of deploying infrastructure, they're deploying it so that they can actually bring some of those workloads on-prem, uh, deploy some of those inference and agentic uh workflows that we talked about, and do it at a cost structure that's much, much lower than what they're able to do today in cloud. And so we've been prioritizing partnerships with with uh some of our large partners like like NVIDIA that have access to many and many of these these uh AI technologies, but then bringing our simplification of the network. We've been building networks for over 40 years now. The network started with with uh two professors that were trying to get computers to talk together, and this is just uh extrapolation of that problem space.

SPEAKER_01 13:57

And so um so so give me a set of priorities that you actually set up to say these are the things that I So it seems like one of them was okay, simplify the management.

SPEAKER_00 14:07

Yeah.

SPEAKER_01 14:08

And you've done the simplification of the management in a multitude of different ways because you've said Nexus dashboard and ACI got converged.

SPEAKER_00 14:15

Well, think about it that way. Like take the same technologies they're already building those front ends with today and extend them to AI so that if they want to start to run AI workloads, it's not bringing in another networking technology, another operational tool, and another set of experts that have to go figure out how to make that work. They can just extend the existing network infrastructure they have, and so their experts that run the network today can run that AI infrastructure. Trevor Burrus, Jr.

SPEAKER_01 14:36

And then you also made sure that that was fully integrated into Cisco Cloud Control, which became our unified management plan. And Will, you were one of the spiritual leaders internally of making that happen. Walk us through the calculus on that. Why did you do that? What was what was so magical and important about it?

SPEAKER_02 14:52

So I think the the key, and you know, we've been working, as you mentioned, to bring the Nexus portfolio together. When we go cross-domain, some of the big gaps Cisco has historically had, it's not just about a UI interface, it is about things like I'm uh I want automation to have a single gateway point. So being able to bring together that common API gateway uh and start to normalize the the cross controller. And then the other part was you know a consistent way to provide an AI uh front end with new interface models with AI Canvas and not having to build that separately into each product. Um and so with the Cisco Cloud Control architecture that we put together in the last year, it it you know, the the first pass, uh easy-ish, is to get the um the SaaS products to be able to then be more seamless with uh Cisco Cloud Control. The other piece that we've been working on, which engineering-wise has been more work, but is a big win, is when we also now start to have the on-prem controllers be able to be cloud tethered and still have that Cisco Cloud Control front end. And so things like Nexus dashboard being able to have, and many customers have more than one of these. They may have you know global deployments. We have customers that might have dozens, uh, dozens and dozens of controllers, being able to have that then pulled together and have everything from common inventory topology, uh, being able to put together uh the beginnings of a unified EVPN fabric that'll work across our catalyst area, our uh data center area in a cohesive fashion, building upon each of those sub-products. Um, so these are some of the key key aspects, and it definitely is uh the the customer reception um has been you know amazing, and so now we can you're working on our roadmap for the next uh six or nine months to continue to fill out these features.

SPEAKER_01 16:42

So the this was a very important contribution that you made because you were one of the big kind of drivers of hey, I need to have this BGP EVPN fabric that was like a a data center and a campus branch network should just be able to get configured um in one click. And we showed that demo, that was a live demo on stage at Cisco Live where they were able to do that. Um what is the benefit for the customer and what's the ROI for the customer when that happens? Anyone of you can uh talk about it. Automation.

SPEAKER_03 17:14

You hit the nail on the head there. Like customers just want to solve a problem. Yeah. I want to make sure that my telephones that are out in the branch.

SPEAKER_01 17:19

But why is it important for my data center and my campus branch network to be?

SPEAKER_03 17:24

I've got telephones that are out in the branch, and I want those telephones to talk to the call manager. I do not want those telephones to be logging into the source code repository or the finance system. The network is the way to do that. You want to set up a segment that goes from the data center, from the application out to the telephones, and we make that super, super easy.

SPEAKER_02 17:42

Yeah, so another so a customer may not necessarily want to create a single fabric uh in the definition of a fabric, but uh to the point uh Tom's making, um, being able to have uh have them connect through what we call border gateways and then be able to have policy that extends from an application running in the data center to a laptop uh or an endpoint that's in the campus, that becomes you know, we we we stitch it together in a way that that now is is feasible, and we can do everything from better telemetry and reporting uh to uh better ways to insert policy on top of that. And so this is something that you know we've been asked from customers for a couple years, um, and frankly, it's taken us too long. Um, but now that we're here, we're gonna move as quickly as possible and filling out that that functionality.

SPEAKER_03 18:26

And I'll in a post-Methos world, this kind of segmentation is what that is, right? This kind of segmentation is more foundational than ever because we have to be working on the assumption that the attackers are in. There's just so many vulnerabilities. They're finding ways to get in. Let's make it hard for them to move around.

SPEAKER_01 18:43

Yeah.

SPEAKER_03 18:43

EVPN and this.

SPEAKER_01 18:44

You're trying to prevent lateral movement. Correct.

SPEAKER_03 18:46

And that there's no better way to do that than in the network and putting segments in place that are like almost like common sense, we make that easy to deploy.

SPEAKER_01 18:53

So uh if you were to let's switch to mythos, because mythos is a pretty important phenomenon of what's happening right now. Yeah. And specifically in the data center, it gets to be, you know, scary. Superably scary when something goes uh go go goes awry. So um you've got methos, um mythos has identified a bunch of vulnerabilities. Your data center infrastructure might have some of those vulnerabilities. Yeah. How do we help customers? Yeah. Out. What should customers do?

SPEAKER_03 19:24

Yeah. So so I'm gonna argue never miss a good crisis to drive change. So let's not sugarcoat it like this mythos thing is a crisis for our customers. Because the the way that we as an industry have been working is that we would create a data center design, approve that data design, harden it, and then you don't change it for like leave it there as long as you possibly can. Exactly, for like ever, right? So like, you know, every literally every couple years, you're like, okay, I'm gonna update this thing only when you have to. And now in this postmeetos world, it's not unreasonable. Exactly. Right? Now, these concepts are not radical here. Like in the cloud, that's how it works. That's how Kubernetes apps works. Lots of little tiny changes instead of one big, giant, gut-wrenching change. So if we can all get there, I think we're all gonna be better off. But but we have to adapt the way we've been thinking about managing infrastructure. And unfortunately, Cisco, we're on our toes here, not our heels, because we've been thinking about this for a while. How do we manage vulnerabilities in critical, inline, high-performance systems like a switch, a router, you know, a firewall. And so uh you know, one piece of the puzzle, and I want to point out this is only one piece of the puzzle. We have this capability we call Live Protect. And so Live Protect, you can think of it as a finger in the dike. It's an emergency control that can be applied on a system. And the intent is that it is very, very targeted, very, very specific, and lightweight. So as customers get comfortable with these capabilities, you'll be able to put them on a system without a reboot for a system. And our design intention is we want to go there, you don't even need a change control window. Like these things are just gonna automatically go in. But Will, I'm gonna ask you this question. It's a kind of a fake question. Is Live Protect a patch? Uh so no, it's not a patch.

SPEAKER_02 21:12

Uh it is uh answering very seriously. Um it is a uh it is a control, uh compensating control, um, and it is something that at the next patch window, then you'll want to replace with a full patch. Okay. One thing I want to add to what you said, Tom, is you you know, things like Live Protect, um, I'm excited about as a technology that I think is gonna, you know, we are gonna have uh the world has changed. Things will be different for the next several years. But in the short term, I do want to uh you really um convey that we have confidence in our core network OSs, iOS XE, iOS XR, NXOS. We've been doing a lot of hardening of these NASs for several years. Um we've also been working to improve the agility, our ability to address so um you know, while we definitely are seeing a you know, uh a a without getting any specific numbers, we are seeing function. We're seeing a step function. Um I've been very happy across the engineering team to see the rate um and it does involve engineers working weekends and such, but the rate of of working through those um and that we will be able to, you know, really sort of this first wave in the upcoming, you know, patch cycles, uh, we do believe we'll be able to put out quite a bit uh to help address that and get ahead of you know attackers in that standpoint.

SPEAKER_01 22:28

I actually think this is the first time that you actually have a real shock at um adversaries not always having the advantage like they have in the past. Because they've had to they've they have to be right once. The defender has to be right every single time. This is the first time that you'll see kind of a flip where um if like I I I think I give a lot of credit to Anthropic on the fact that they did Project Last Wing, they gave companies time to go out and make sure that they got hardened. It's material. It's material right and and what that'll do if we can continue to keep going down this path, I think the defenders might actually have a very durable sustained advantage over the adversary. Yeah. And who knows, maybe like this becomes a very different world especially as you write code in memory safe languages. I've actually well this is um something that you and I have talked about but I was um I was with uh CNBC recently and one of the things that um they asked me was hey so what is your vision of what AI is going to be able to do in the next couple of years and I've basically told them that hey look there's going to be no such thing as legacy code and a legacy company because we will have eradicated all legacy code from Cisco within less than 24 months and I think that's a very very realistic goal to go out and get to I love your optimism and I and I do agree with you in the kind of medium term.

SPEAKER_03 23:52

In the near term it's going to be rough sledding right because this isn't a Cisco problem. This is a software problem this is a software problem yeah and every customer is going to be inundated with every single vendor for every single device that has software the thermostat in your conference room is like oh it needs an emergency update right so so it's gonna be it's gonna be crazy. And and I think you know I'm proud of the fact that Cisco is I think leading the industry here and we're trying to bring a mindset shift which is CI CD, cloud principles, lots of little changes, not this once you know every two year giant gut-wrenching change. So LiveProtect is a bridge between more frequent updates but it is not the complete answer, right? It's just a temporary kind of like emergency measure the temp the right answer is more frequent updates. And what I think I'm really excited about is the work that I've seen on team are doing with digital twinning Anuraga's our leader who runs the workplace technologies business. And DJ and team who's uh runs the AI we have an AI agent that is a CCIE, right? And so if we can use this to help streamline and automate the automation and the upgrade process, take this in in totality better quality software, right? A good foundation that we're starting from Live Protect bridges the gaps and then AI powered more frequent updates that's how we get ahead of the attackers.

SPEAKER_01 25:13

Yep.

SPEAKER_03 25:13

Infrastructure that isn't running two-year-old vulnerabilities, right? But it's moving more like your iPhone. When's the last time you updated your iPhone I don't even know.

SPEAKER_01 25:21

I don't even know because it's like it just doesn't work. Yeah so Kevin what are you most excited about on the things that you've built and what are you most excited about are things that you're going to build.

SPEAKER_00 25:33

Yeah I mean we we actually haven't talked as much about the the deep infrastructure components but you know I've been doing this for for decades at this point building high-end systems at Cisco all hardware based, hardware-based forwarding it because Cisco is the only job we've had right I've had and uh we're gonna keep it that way for parts of uh four decades so I'm I'm looking to get to to five decades soon. So um but you're only 25 how is that even more so so if you think about these infrastructure components we used to build a new chip every four years and and it would take us 10 years to roll it out across these companies. If I look at the pace of change that we're driving with AI, you know we're releasing new silicon every 18 months. We're building chips now that are denser than entire rooms full of systems that we had even five years ago. And so what excites me is the engineering innovation and the pace of change and the fact that we own all these different components. So we're not waiting for somebody to deliver us a piece of silicon we're building it. We're not waiting for somebody to figure out how we connect these devices together we build the optics and you know by owning all these technologies we have the ability to put them together in ways that you know people that are piece parting and and buying from bins and trying to pull them together can't do. And then on top of that I can grab security technologies and and bring those in and bring networking and security together in ways that you know you can't if you're just just bidding from a store. And so I really I'm excited of what we've done. I'm excited at the growth that we've seen but uh the next couple years is going to be an immense amount of growth and an amazing pace of change and we feel like we have a lot of the tools to drop it.

SPEAKER_03 27:06

Let me tease a topic here G2 you made the analogy of because Cisco owns silicon the software the system all the way up the stack has that sort of Apple feel. And boy I loved remember the SE30 right that was back in the in the day it was just a beautiful machine because of that that level of integration. So I'm gonna argue that our switches you can you can see it and feel it touch it today. Use a Nexus switch it's because of the programmability and the optics and all of that but when you when you take the things that you talked about and you talked about I'm gonna argue the next kind of unit of measure here it's the whole rack. I think that in the timeframe you Kevin you talked about like customers are going to buy data center by the rack. You don't piece them together out of boxes anymore. There's gonna be liquid cooling they're gonna have these direct to optics and so that talk about secure AI factory for a second. That's the answer. Right so our version of of the rack is a secure AI factory. It is the compute the network the storage and importantly the network services and what's the number one network service in my opinion firewall. Firewalls are a pain in the neck they're cumbersome they're hard to use we just make it magical and easy and built it in there and then load balancing and Cisco has all these pieces.

SPEAKER_01 28:09

And things like the security for agents and AI defense and observability and that entire stack. And so how um how real is that and when does um how do customers go about buying it?

SPEAKER_00 28:26

First of all it it's real it's deployable today. We have customers that are buying these technologies and and building out data centers today. Going back to the ease of use thing what we've tried to do is build this these concepts of of pods and put things together in in groupings that allow people to deploy for training workloads or inference workloads and give them a blueprint for exactly how to roll that out. And so it's it's something that we see deployed today. We're gonna see even more of as more and more of this infrastructure moves on-prem. Yeah Splunk pods yeah yeah exactly we we to people were deploying a lot of Splunk and they needed AI and so we put that together into a blueprint and it's off and running.

SPEAKER_01 28:59

So it's so here here's a question I'd have for the three of you what are questions that you would wish customers would be pondering more and asking more about and where do you think Cisco is most misunderstood?

SPEAKER_02 29:14

So one that I've seen in discussing with customers is that they they view that the the template that Nvidia's had now for a few years, which has been a good you know the superpod, you know it's been a sort of just rinse and repeat uh motion that that um that that is the going to be the de facto moving forward and and they look to existing solutions um that that have been out in the market for a while um often the the question that I it's the time to ask um is how are things like the new inference architectures going to change that? And what we're seeing is you know there's a range of new inference accelerators not only NVIDIA with Grok, uh Cerebrus, a range of other companies and the how you put together the GPU with that inference means that you need to have flexibility in the architecture. It means that even where the the final unit might be a rack level to what's Tom saying, but it might not be for instance a full NVL72 with with the the scale up going to the the hundreds of GPUs in one shot, Ethernet starts to play more of a role in that, potentially in the handoff between the accelerators and the GPUs it's also playing a bigger role where storage is becoming a bigger topic. You have these very large contexts and putting those into storage at high performance on that front end network we talked about earlier can be a big deal. So going in and thinking about how you have you know leverage if you have you know Nexus or other architectures day how can you leverage your your network as you expand to these how is this going to change? And maybe the pattern is not going to be as simple as this this existing superpod just repeated for the next forward couple of years.

SPEAKER_01 30:54

Tom what do you wish customers asked more about and what do you think we have most misunderstood about I love this question.

SPEAKER_03 31:00

I wish they would ask the question where does security go? Because in the conventional thinking security goes in a box and that box lives at the edge of the network. It's usually called a firewall firewall, IPS, fancy, fancy box. In the Cisco model, this when we're delivering an integrated solution and we're thinking holistically about the data center, you know where security goes? Everywhere everywhere.

SPEAKER_01 31:22

Right? Fused into the fabric.

SPEAKER_03 31:23

Fused into the fabric. And so examples of this is we have smart switches which can allow each individual port to be a little tiny baby firewall. Super fast layer four firewall that can do really great segmentation. Another example isovalent we can see every single east-west transaction every process that initiates a connection every process that terminates a connection that that data exists today but it's a thousand times more than you could look at but with the integration of isovalent into Splunk all of a sudden we can see that without burning up some gigantic congestion bill. And I believe that that level of east-west inspection is going to allow us to run the infrastructure more efficiently find problems more easily but more importantly find the bad guys right in that post mythos world we're like the bad guys are in we have to identify lateral movement and looking at the process level is the way to do it.

SPEAKER_01 32:12

So And what do you think we are most misunderstood about as a company?

SPEAKER_03 32:19

We're the networking company and I think what's exciting about this world is that we're taking security and we're transforming it by fusing it into the network. We're taking compute and we're transforming it.

SPEAKER_01 32:28

It's actually it's interesting because you cannot be a networking company if you're not a security company except that we're the only one who is a secure networking company in the world. Everyone else actually doesn't have that kind of part of the stack. It's our clear advantage.

SPEAKER_00 32:40

Yeah yeah yeah I don't know for me it's connectivity like the the exciting thing about AI is like like we'll describe the architectures are changing and the network is becoming that critical component in gluing everything together and and we are the network company and so it's an exciting time to be building networks and and fusing security and yeah I'm just I'm having I'm having fun.

SPEAKER_03 33:01

Any questions that I didn't ask that you'd want our audience to make sure that you you addressed well I I want to start with one if you look at the vast majority of AI workloads today where are they running in the cloud? And if you look at like kind of the life cycle adoption from a kind of a business standpoint, we're in that kind of early adopter phase G2, you you were right at the tip of the spear you were the first guy with like AI coding everybody go, right? Literally like you're flogging us like go, go, go, go, go.

SPEAKER_00 33:32

And then we got the bill it was like whoa whoa stop almost but you're taking that in industry there's there's a move towards optimization.

SPEAKER_03 33:43

Right? And so so you know and I think that that the right answer is going to be very much of a hybrid model where there's some tasks you want to put on the fancy latest and greatest model, but there's other tasks that are super valuable tasks like implementing dark mode. Do you need the frontier model for that?

SPEAKER_01 33:59

No.

SPEAKER_03 33:59

No. And so so understanding tokenics, understanding what things cost and then having the ability to kind of blend an on-prem open weight model with these frontier models to me that's the only answer. And Cisco's uniquely positioned I believe to deliver and you have to have inferencing everywhere.

SPEAKER_01 34:15

You can it can't just be in the data center it has to be that's super exciting. That that changes everything when the inference is at the edge correct uh in addition to the data center. Yep gentlemen thank you again and um I I have to say that what the teams are doing right now is um is a proud moment for Cisco and um thank you for all the leadership that you've all shown and I think we are just getting warmed up we're just getting started there's so much more that we're gonna do and um the pipeline that we have from the chips to the agents um is nothing short of spectacular so a lot of people are like well can Cisco keep this up and like we're only gonna accelerate it. Thanks again