Hello everybody. I am honored today to be here with my good friend Tom Gillis, whom I've known for many, many years.
Yeah. Too many years, right?
Yeah. We worked together and we're partners in crime at VMware doing all kinds of things around network and security. He ran that business for us there. Did a great job.
He was actually kind of, I think, before that at Cisco and now come back to Cisco. You're the GM of security and infrastructure. Is that right?
Great. And that includes all the security products and now all your AI and UCS servers, all the AI infrastructure. Yeah. Networking, compute, the whole stack.
Wow. That's fantastic. Well, he's a seasoned veteran. Cisco is a very important partner of us. They're also an investor in us as a company and we've had a many, many multi-year partnership. And I'd like just to explore the new dimensions now of what we are doing in AI because that's obviously the most important thing. And what we've kind of struck into now because we back up the world and protect the world's data, that a lot of this data now has become a participant to data and AI conversations. And we've been able to build with Nvidia this product called Gaia that could run on prem. Yes. And one of the first server vendors we wanted to go to was you. Yes. So maybe you can tell us a little bit about the AI servers. You call them AI pods. How that ties to UCS and what all you're doing around that AI infrastructure.
Well, first of all, S, we've enjoyed tremendous success in the AI infrastructure space. AI is just taking off. And this is maybe not obvious, but the networking is actually the hard part. And here's why. If you're familiar with traditional workloads, they're using a CPU. And a CPU is a sequential processor. So it's executing bit by bit by bit. A GPU, which is what powers these AI apps, these are parallel processors. And each GPU has lots and lots of processing elements on it. And furthermore, an AI app will span multiple GPUs. So a lot of these very powerful AI applications, they could be using 10,000, 100,000 little tiny processing elements, right? So it's these very, very parallel processors and it's the network that has to string all that stuff together. So GPU to GPU communication is a real challenge. Now this is Cisco's bread and butter is networking. Bread and butter. We found a way to really make networking great in the context of these AI servers.
Bang. And the thing that we do is we make this kind of special networking that we in the industry we call it a backend network. It's GPU talking to GPU. We make it the same as what your front end network, which is all the stuff that powers your data center today. And an AI pod just is an extension of what the former UCS server was with now a lot more GPU capabilities on it.
How should I think it's got the GPUs. It's got special net interface cards, network interface cards, and then it's got this integrated Cisco networking to do the backend networking so that when your GPUs need to talk to each other, it's going to go smooth as molasses.
So, as you know, Cohesity, we secure and protect the world's data and we run on premise and many of our large customers, we run on UCS servers, correct? So adjacent to this workload that's backing up and protecting data we can now run a RAG, retrieval augmented generation, capability with Gaia, the product we built with Nvidia, on this AI pod, which is fantastic. Now you also built around this AI defense and some other security capabilities. Talk a little bit about that.
Yeah, so Nvidia has this notion what they call an AI factory, which is the whole kit and caboodle you need to run an AI application. It's more than just the GPUs, the networking and all the security stuff. And for AI applications, security really needs to evolve because if you think about a traditional application, you click on a link and a bunch of stuff happens and it's very predictable. With an AI based application, it's now like a chat interface and you can ask it the same question twice and get slightly different answers. So static rules don't really work there. And so with AI defense, we actually, brace yourself, we use AI to understand and supervise the interaction between the user and your AI app. So AI protecting AI is really the only way that you can understand is this app being used correctly or is someone trying to trick it or manipulate it into sharing something it shouldn't be sharing.
Well, we're very excited about that and a lot more to talk about in a future session because we're very in kind of our notion of resilience everywhere. Also concerned about agents that go rogue. Yes. And you know agents being able to rewind to a previous state. So we'll talk about that in a future session.
We got much to talk about because identities are going to be human and nonhuman and that could be a good part.
