Why AI Needs a New Network – Six Five On The Road at the Cisco Partner Summit

How are industry leaders like Cisco and NVIDIA advancing enterprise networking for the AI era?

‍

From the Cisco Partner Summit 2025, host Matt Kimball, (VP and Principal Analyst at Moor Insights & Strategy), is joined by Cisco's Will Eatherton, SVP Infrastructure Engineering, for a conversation to explore how Cisco and NVIDIA are joining forces to power next-gen AI infrastructure. They unpack innovations like Cisco’s Nexus Hyperfabric AI platform, the joint Cloud Reference Architecture, and how these technologies are enabling enterprises to scale AI workloads with greater performance, efficiency, and agility.

‍

Key Takeaways Include:

‍

🔹Drivers of AI Networking Expansion: An exploration of why rapid AI workload growth demands new network architectures and what’s motivating the Cisco × NVIDIA partnership.

‍

🔹Cloud Reference Architecture Advantages: Insights into what their new joint solution empowers customers to do, from streamlining deployment to enhancing scalability for AI clusters.

‍

🔹Nexus Hyperfabric AI Unveiled: A look at what makes this platform unique for building AI-ready networks and its impact on ease of cluster scale and performance.

‍

🔹Future of Cisco × NVIDIA Collaboration: Will Eatherton shares thoughts on the evolving partnership and its implications for the next generation of enterprise AI infrastructure.

‍

Learn more at Cisco.com.

‍

Watch the full video at sixfivemedia.com, and be sure to subscribe to our YouTube channel, so you never miss an episode.

Or listen to the audio here:

Disclaimer: Six Five On The Road is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

‍

Matt Kimball: Welcome to another session of Six Five On The Road. My name is Matt Kimball, Vice President and Principal analyst with Moor Insights and Strategy. And today I'm here talking with Will Eatherton, Senior Vice president of Infrastructure engineering at Cisco.

‍

Matt Kimball: So AI, the AI era is upon us, and more specifically the AI inference era is upon us. And with inference comes a lot of traffic. Now going back to this, as most folks think about AI, they think of one company, Nvidia, with good reason. 

However, if I was going to draw a parallel to Nvidia in networking, it would be Cisco. Cisco. And interestingly, Will, there's a strong partnership between Cisco and Nvidia and there's been a lot of news coming out of the two companies recently. Right?

‍

Will Eatherton: Absolutely.

‍

Matt Kimball: So talk to me. You know, Cisco and Nvidia just announced a few announcements, but one really big announcement around AI networking for people that are watching at home and maybe are a little bit less familiar, kind of. Can you walk us through that and talk about that a little bit?

‍

Will Eatherton: Sure. So first I'll just mention the partnership. You know, a year, year and a half ago, many folks would have never expected that Nvidia and Cisco would have come together. The view would have been we were going to be highly competitive and overlapping and that there wasn't a partnership to be had. But what we've seen and what we had announced earlier this year and then followed through with announcements, including in the past week, has been that, you know, there's many types of Ethernet and that there's the speed that these deployments need to happen across enterprise, sovereign, cloud, NEO Cloud requires more feet on the ground and that's something that Cisco and Nvidia can do together. So what we announced last week were a couple things. The big one was a new switch, which we all love switches and networking. And this was based on the Nvidia Spectrum ASIC, but it's built into a Cisco switch with Cisco software on top of it. And that's really fundamental for our customers.

‍

Matt Kimball: Nvidia living within Cisco in perfect harmony with software.

‍

Will Eatherton: Yes. And the key point there is that customers can use all the software management tools that they've had experience with at Cisco and an ecosystem that has tens of thousands of CCIE’s and all of this training, but they can do it with the proven technology stack for AI that Nvidia's provided. And so that was one key announcement last week and we were very happy. We had Jensen drop by our booth and sign the switch and we have pre orders for that. We just announced that the orderability is turning off.

‍

Matt Kimball: So just to be sure, for the audience at home too, we're talking Cisco switch, Cisco Silicon running and working with Nvidia Silicon to drive and feed those GPUs as fast as possible. Right. So keep them fed and optimal performance nonstop.

‍

Will Eatherton: Yes. And there's different combinations. So that new switch, the N9100, that's the Nvidia Silicon and the Cisco software. And the key thing is that it's NCP reference architecture, which is the Nvidia Cloud partner. And that means that you can be part of that Nvidia program while you're using the Cisco software on that chip in the backend of the GPU to gpu. But with that said, there's also combinations with Cisco Silicon, Cisco software running for both the backend and the front end networks. And so the end goal here is a set of options that the customers can choose based on what their goals are. But all of them combine the Nvidia and Cisco ecosystem.

‍

Matt Kimball: So help me out with this. So you talk about the, you know, the cloud reference architecture, what does that do for a customer that they maybe couldn't do before or they had to do more inefficiently before? What am I getting at as a.

Will Eatherton: So if I start with simple terms before, if you wanted to be in the Nvidia ecosystem, you wanted to have that reference architecture approval for Nvidia. There was one stack that had all of the components and software and it had to be done exactly one way. The new thing with the Cisco cloud reference architecture is that it was based on the design tenets of NCP from Nvidia. It was blast reviewed by Nvidia, but it has Cisco switches with Cisco Silicon in there running, essentially a connection between our switches and the Nvidia NICs that are in the server. So server, switch, server can now run as one unit and it's again Cisco software ecosystem in that case with that reference architecture. So it's a choice, it's still going, you know, thousands, tens of thousands of GPUs, but it's alternate ways to deploy that architecture.

‍

Matt Kimball: So optimal performance, I mean, you're really kind of, you're bringing the best of both worlds into one single platform for consumers. You know, you say, you know, reference or a reference cloud architecture. Is this something that only applies to hyperscalers and NEO Cloud, or is this something you're seeing enterprises pick up as well?

‍

Will Eatherton: Well, right, so the Nvidia cloud partner reference architecture, which is the Nvidia thing, that one's very much more NEO cloud focused that was intended for that and it's more of that multi-tenant. The Cisco cloud reference architecture is generally just about a scale point from thousands to tens of thousands of GPUs. And we absolutely have engagements with high end enterprises. For instance, quants have a lot of their own models, so they're candidates for this. We see life sciences as a growing area, so they begin. And so those kinds of ecosystems where it may be completely in house, there's no GPU being sold as a service. That absolutely is a target for the cloud reference architecture.

‍

Matt Kimball: So kind of shifting a little bit. So this thing called Nexus hybrid fabric, hyper fabric AI is coming, right? What's unique about this? Give us a 10,000 foot level. What makes it unique, what kind of benefits do you get and how does it change? If I'm looking at building out that AI ready network, how does it help me?

‍

Will Eatherton: So I'll start off and say that we have a long history with how to manage infrastructure. I like to say that we've done it every wrong way that is possible and a few right ways. About 18 months ago we looked at our data center strategy and we said if we were a startup and we were going to disrupt both Cisco and Arista, what would we do? That was fundamentally new. And we came to an approach that was first of all SaaS delivered, but it was really around starting at the design process. I want to build a cluster with 1000 GPUs and I want to use vast data as an example and what kind of storage capacity I do want to have. So I do a whole design and then essentially I have an intent that's put in there. And at that point it would essentially compile that into an architecture and a bom you can then order and then when it comes everything from plug this cable into this box, oh, you got the wrong port, move it to another port. It's really the Meraki approach for data centers. And so we've been in beta. We have a number of customers in the early version that have gone to production and we're very excited that again we have a long history with our Nexus portfolio. But this is a disruption approach. So it just makes it so much easier for enterprises in particular to adopt AI.

‍

Matt Kimball: I really like that because as I talk to enterprise IT folks and as I speak to even larger Neo clouds, like kind of customers, one of the things I hear is complexity, complexity, complexity. It's a huge barrier to adoption for AI in general. And there are a lot of elements to that. Right. But a big element to that is networking. So the more frictionless you can make that deployment of that infrastructure that goes out. And by the way, every company needs to upgrade their network to support these workloads. The more, the more, the more frictionless you can make that obviously the greater value. Right. But that whole time to first token kind of metric that everybody seems to be trying to measure goes down drastically. So I love what you're doing there and I have to think that the feedback you've gotten to date has been.

‍

Will Eatherton: Pretty spectacular, it's pretty exciting and actually interesting enough. Nvidia, as we've gone through and demoed with them, they've been very excited because in their standpoint they want to democratize and see as many GPUs out there as possible. And they do worry about the tail of the enterprise. And so this approach can help a lot more companies and maybe it's only 64 or 120 GPUs to start, but it's approach that again is not only the network, but it also incorporates the servers and storage with initial partner vest.

‍

Matt Kimball: Yeah, I love that, I love that. So you know, we're talking about, you know, some of the work you've done with Nvidia and kind of all the great work you're doing as you kind of look forward to, maybe six months out, maybe a year out, maybe two years out. Where do you see this relationship going? I mean it's quite frankly it's very interesting that you see, you know, you started out this conversation with a year ago, who would have thought, right? And I would agree with that. And I'm just wondering how much further you can take this because you are the two relative giants in the AI game for you know, for neoclouds, for, for enterprise and even for commercial enterprise. Where does this go next?

‍

Will Eatherton: So first of all when we put together the agreement, it's intended to be a five year base and then we'll see where we're at. So it is multi-generational. So some of the, I guess simple things are as new silicon comes out, you know, 100T, 200T, we'll be revving our switches as the algorithms involve on how to get traffic end to end. That's something that we will be, you know, sharing back and forth and ensuring that the software updates handle that we've talked about there could be silicon aspects of shared IP in some cases optics and they've been pushing co packaged optics very hard. We also have development in that area there's some opportunities for that. So I'd say those are sort of the base technology. I would say at the higher layers security is a big addition into the cluster that we've been working on. We have announcements that will be coming out next year and more of those areas and I'd say the last area that I'm interested in and right now I'd say we're more at an engineering level is the multi data center. And so this is something we've worked with. Well our history is routing and WAN and interconnection but we've worked with hyperscalers for instance in multi data centers. I think as we start to see folks not just think about that from management but I might actually want to stretch jobs which is where the whole scale across topic comes up. I think we're still early, there's still R and D on how to really make that work. We're doing work on that. Nvidia's put out some work on that. I think that's an area for collaboration and potentially leveraging. There's many types of Ethernet but some of the Ethernet switches have a more deep buffer and have been for that sort of WAN areas that that could be applied to. So again I think there's. One of the things I'm very excited about is there's so many areas of R and D and it's much more than here's a new silicon, let's build a box of it.

‍

Matt Kimball: I love that it's really. It's kind of the two plus two equals five equations. Right. I don't like to say one plus one equals three. That's too easy.

‍

Will Eatherton: Sounds good.

‍

Matt Kimball: A little bit more advanced. One last question for you. It's kind of a bonus question. Silicon. Not a lot of folks know this and not enough folks know this. I should say Cisco is not just a switch company, not just a networking and security company. You're also a silicon company. Right. Can you talk a little bit about what you're doing in that space?

‍

Will Eatherton: Sure. So you're right that not enough folks know that. You know Silicon Cisco's had a long history of going back 25 years probably of doing silicon. In the last several years we've gone pretty hard on everything from the high end web scale multi-tens of terabytes to things that would be used in service provider networks and enterprise. I think yeah. An example here would be hyperscale where they are looking for I would say a top couple vendors and I think today it's safe to say that, you know, what you see most out there is Cisco, Nvidia and Broadcom are the top three. Which that's one reason why many folks a year, year and a half ago had said, you know, we'll never partner because it's naturally competitive. But I think again where Nvidia has been focused is on, you know, mainly the GPU to gpu. And while we have some overlap there, we've just seen so much more benefit of having the partnership and working through the occasional coopetition kinds of topics.

‍

Matt Kimball: Yeah, fantastic. There's so much to what you're doing and to what you said from a kind of the full stack from the silicon to the hardware to the software and even to the security aspect that you know, it's just. And when you bring two giants together like a Cisco and an Nvidia, it seems like not an unstoppable force. Perhaps it is, but more importantly, truly that easy button for the Neo clouds, but more importantly that enterprise company that two years from now is going to be starting on their inference journey.

‍

Will Eatherton: Well, I think the whole industry really wants to see that this gets democratized out to enterprises and I think that is going to be a key area that we're poised to find focus on. In the meantime, while that's still early, Neoclad continues to be a big focus.

‍

Matt Kimball: So Will, thank you for taking the time to chat about the partnership between Cisco and Nvidia and how you're driving, you know, kind of driving away the friction for AI adoption at the largest of companies and readying the market for that next wave of adoption from enterprise. Truly a compelling story. And for those of you watching, thank you for taking the time to watch this episode of Six Five and Building the AI network of the Future.

‍