In this AI Factories – Data Centers of the Future segment, Gilad Shainer, senior vice president of marketing at NVIDIA, joins theCUBE’s John Furrier at the New York Stock Exchange to explain why networking is becoming the “operating system” of AI-scale infrastructure. Shainer breaks down what interconnects really are and why low-latency, jitter-free fabrics such as InfiniBand and Spectrum-X are essential to turning thousands of GPUs across servers, racks and data centers into a single AI supercomputer. He details how extreme co-design of hardware and software – spanning GPUs, networking, frameworks and telemetry – redefines the data center as one coherent compute engine for next-gen AI factories.
The conversation also explores how density, energy efficiency and reliability shape the new AI factory blueprint, from liquid-cooled, copper-linked rack-scale GPUs to optical scale-out powered by co-packaged optics that can cut network energy use and boost uptime. Shainer highlights how BlueField-4 data processing units sit at the heart of storage, security and edge deployments, and how Spectrum-X Ethernet is enabling synchronized, large-scale AI clusters for partners such as OCI, Microsoft, Meta and Cisco. It’s a deep dive into why AI factories are emerging as the new unit of value in enterprise infrastructure and why networking will be central to the next era of distributed, AI-powered systems.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. If you don’t think you received an email check your
spam folder.
Sign in to AI Factories - Data Centers of the Future.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for AI Factories - Data Centers of the Future
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for AI Factories - Data Centers of the Future.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. If you don’t think you received an email check your
spam folder.
Sign in to AI Factories - Data Centers of the Future.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to theCUBE + NYSE Wired: AI Factories - Data Centers of the Future
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Gilad Shainer, NVIDIA
In this interview from theCUBE + NYSE Wired: AI Factories – Data Centers of the Future event, Glean co-founder and CEO Arvind Jain joins theCUBE’s John Furrier to unpack what’s really working in enterprise AI today and what comes next. Jain explains why knowledge access remains the first successful AI use case at scale and how Glean’s enterprise search brings AI into everyday work. He details the past year’s lessons with AI agents – from the need for guardrails, security, evaluation and monitoring to democratizing agent building so business owners (not just data scientists) can create production-grade agents.
The conversation dives into Glean’s vision of the enterprise brain powered by an enterprise graph, highlighting the importance of deep context, human workflows and behavior to reduce “noise” and drive outcomes. Jain outlines core building blocks – hundreds of enterprise integrations and a growing actions library – that let agents securely read company knowledge and take actions across systems (e.g., CRM updates, HR tasks, calendar checks). He discusses how organizations are standing up AI Centers of Excellence, prioritizing “top 10–20” agents across functions like engineering, support and sales, and why a horizontal AI data platform that unifies structured and unstructured data – accessed conversationally and stitched together via standards like MCP – sets the foundation for AI factory-scale operations. Looking ahead, Jain says Glean’s upgraded assistant is evolving from reactive tool to proactive companion that anticipates tasks and accelerates productivity.
In this AI Factories – Data Centers of the Future segment, Gilad Shainer, senior vice president of marketing at NVIDIA, joins theCUBE’s John Furrier at the New York Stock Exchange to explain why networking is becoming the “operating system” of AI-scale infrastructure. Shainer breaks down what interconnects really are and why low-latency, jitter-free fabrics such as InfiniBand and Spectrum-X are essential to turning thousands of GPUs across servers, racks and data centers into a single AI supercomputer. He details how extreme co-design of hardware and software...Read more
exploreKeep Exploring
What is the significance of Nvidia in the context of AI technologies and the stock market?add
What are the computing requirements and infrastructure necessary for running AI workloads effectively?add
What are the considerations for connecting GPU ASICs in a data center, particularly regarding media and cooling methods?add
What role does BlueField-4 play in AI supercomputers and data processing?add
What is the importance of networking in the design of AI supercomputers and edge computing systems?add
What are the potential applications of AI in a retail environment that enhance customer experience based on real-time data and personal recognition?add
>> Hello, John Furrier with theCUBE here at our East Coast studios of the New York Stock Exchange. Of course, we have our Palo Alto studio connecting tech and Wall Street, Silicon Valley and money. This is our AI factory series. We're featuring the leaders who are making it happen. Of course, Nvidia, the leader in AI and AI factories and AI technologies has been making the market. They are the most active stock here on Wall Street. Of course, in Silicon Valley, all the innovators are leaning into Nvidia. We get a lot out here. He's the Senior vice President of Networking at Nvidia, CUBE Alumni. Welcome back. Welcome to our first time in our studio here on the East Coast. This is our East Coast access point. It's our Metro PoP, point of presence, networking Silicon Valley and Wall Street. Thanks for coming on.
Gilad Shainer
>> Happy to be here.
John Furrier
>> So I want to get into some of the cool things. Nvidia dominates the boards here in terms of the financial markets, but as people look at the future, there's a lot of roadmap, a lot of things happening in the market that people are learning more and more about. I think over the past three years at GTC, you're starting to see very transparent disclosure from Nvidia, hey, this is our roadmap. Jensen's very clear what he's doing, supply chain and the ecosystem we're developing, but networking is the hottest area. I've been screaming at the top of my lungs on the CUBE. Networking is the operating system. Jensen said it. And Dynamo, KV cache, whatever you want to call, and networking is the key to success. You are key to that. InfiniBand connecting Blackwells. Now you've got scale out, scale across, seeing a lot more kind of, I won't say convergence, but the collapsing of Spectrum-X. You got connections and interconnects. The key to this factory model and supercomputing and accelerated computing is these interconnects.
Gilad Shainer
>> Right.
John Furrier
>> Describe what is an interconnect because a lot of people here, certainly on Wall Street, like I don't understand all this interconnect stuff. Why is it magical? Why is it such an important? Why are people talking about interconnects?
Gilad Shainer
>> Yeah. So I think it's very simple. Okay? It's really very simple. When you build a data center, a data center consists of compute engines and something that will connect those compute engines together. Now, in traditional data centers or before AI, most of the workloads were just running inside one server, inside one CPU, and therefore the most important part in the data center was the compute engine itself, and you just needed a capability to access it. So if user wants to run a workload on the cloud, you need the way to access the server, but then the workload just stay within a single CPU, for example, or within a single server. Now when we look on AI, those are distributed computing workloads, which means it's a workload that cannot be executed just on a single CPU, just on a single GPU, just on a single node. That workload need to run across multiple GPUs, multiple servers, and we're not talking about two or three. We're talking about thousands and 10 of tens of thousands and hundreds of thousands. So now the compute engine, it's not just a GPU ASIC, but the compute engine is the entire data center. And therefore you need to have a way to take those GPU ASICs and connect them together so they can exchange information between themselves and to form a single unit of computing out of those ASICs, out of those GPU ASICs, and therefore you need to bring a interconnect infrastructure that makes that connections. Now, that networking or that interconnect infrastructure, it's not just I need to move a simple data from one side to another side because you need those engines to behave like a single unit and therefore you need to build something that has very low latency, something that has ability to move massive amount of data between those compute engines. Everything needs to be synchronized because if the two of us are working together to finish a task, if you're going to do your job first and I'm going to come second, the jobs will take a long time to finish. In order to make it more efficient, we need to work together. We need to be very synchronized in the way that we work. The same goes here. So that connectivity cannot have jitter, which means one GPU cannot get data before another GPU get the data or one cannot be after the other because that will delay everything. And this is what networking does. Networking connects compute engines, connects compute ASICs to form a single supercomputer that needs to scale across hundreds of thousands of GPU units.
John Furrier
>> Yeah. What I like about that description is it sets the table for why this wave has happened, this accelerated distributed computing computing paradigm because the single node was the old model. Hey, you put a server in there, load some software, do some stuff on it, whatever. Now you're basically saying, okay, let's take all those computes, make it one single thing like an entity, like a data center. Now they're talking to each other as if it's one unit. Okay. Great. Makes a lot of sense. You network it together. But talk about the role of software because what all makes this work is now the software because it's not as simple. There's like many tasks happening. We could be working together, but we have clones of each other. There's other things coming in contending for resource, and so there's all kinds of complexity that evolves from this new architecture. So okay, put them all together. That makes sense. It's a system architecture. But now you've got to overlay software. Talk about that piece of it because I think that is what's coming out of this wave.
Gilad Shainer
>> So at the end of the day, you build a data center to run software. That's the purpose of it. But I don't look on that as separate elements, meaning there is a separate software, there is a separate hardware. When you build a supercomputer, need to build it in a co-design way.
John Furrier
>> Yeah.
Gilad Shainer
>> Okay? Co-design means that what you need to run on the software side, the workloads that need to be executed, the problems that you need to be solved, needs to map into the hardware, the infrastructure that you built. And it goes all the way from the workload, the framework, the libraries, the devices that need to execute the connectivity, the management, the telemetry that enables you to optimize everything. It's one big co-design of a data center. At the end of the day, of course you measure that in the outcome. You measure that in tokens per second. You manage that in the productivity of the workers that are running on top of it, but it's not separate element here, separate, it's one co-design that actually enables those AI supercomputers.
John Furrier
>> Two weeks ago when we met in D.C. for GTC, I guess it was kind of a mini pop-up GTC. It was really huge. Jensen used the word extreme co-design. With the word extreme makes it sound a little more sexy. But talk about the co-design piece because I think you guys pioneered that one with the way you're thinking about the interconnects, the multiple resources together, compute. And then the co-design piece, you guys pioneered that. And the other trend that you guys also pioneered was density. You know, in the old days, density, whoa, density is too many parts. You get a lot of heat, all kinds of things could blow up. You got the co-design. You got density. Density is actually a feature, not a bug in this because the more you get closer, talk about that concept of co-design, extreme co-design, and the density opportunity challenge that you guys solved.
Gilad Shainer
>> Yeah. So co-design, it's the way to build a supercomputer. When you build a supercomputer, everything needs to work as single element, right? When you build a race car, it's not that I can pick a wheel from that company, I can find another piece from here, and I'm trying to stitch stuff together. That will not give you a race car. A race car needs to be a element that is so optimized in everything that it does in order to be able to win the race. That's a supercomputer is. A supercomputer is not a collection of elements. It's a co-design that covers everything from the software to the hardware to build a single unit of computing. Now, density, or the way that you can refer to it as computing density for example, data center density is one of the means to build something that is very, very optimized. When you're building an AI supercomputer, it's an expensive unit of computing, and you want to make it the most efficient that you can. It needs to be the most optimized that you can. And when you deep dive into the infrastructure, you have those GPU ASICs, and you have the infrastructure that connects them together. And actually it's not just a single infrastructure. There are several infrastructures. There is a scale-up infrastructure that connects those GPU ASICs and made them like a rack scale GPU. There's a scale-out infrastructure that connects those racks together to form that supercomputer. There is the scale-across because I would like to connect multiple data centers to make multiple data center a single unit of computing. I have a north-south network. I have a storage network. And I need to have a good storage processor that enables the storage access, especially for inferencing. There is multiple infrastructures in place, and you want each of them to be fully optimized. Now when we look on a scale-up infrastructure that takes those GPU ASICs and form a rack scale GPU, for example, there is a huge amount of data that needs to go between those GPUs, and therefore there needs to be a media that actually connects those. And the most cost-effective media is copper. Now copper has limit of distance, and if you can pack, if you can increase the density of those GPU ASICs and use copper, use the most efficient way to connect those. And that's why we're focusing on increasing the density in a rack. That's why we brought liquid cooling. And liquid cooling enables you to put more GPU ASICs in a rack and use copper and actually build a most efficient data center. Now when you go and connect those racks, then you cannot use copper anymore because of distance, this is where you use optical connections. And in order to increase the efficiency there, if for scale-up, we focusing on density and using copper, on a scale-out, we brought co-packaged optics and we built both Quantum-X InfiniBand with co-packaged optics or Spectrum-X Ethernet Photonic with co-packaged optics because that reduces the amount of energy you need to invest in moving data between those racks and building the most efficient infrastructure for scale-out.
John Furrier
>> That's a masterclass. I really appreciate the commentary. So now we're back to interconnects. Copper, very efficient, good energy savings there. Then an interconnect is know InfiniBand, Spectrum-X, Quantum-X. You got that kind of high throughput key connections across to other racks. Racks are highly optimized. They do their thing like you talked about multiple compute engines and fabrics around that, optimize like a race car and they connect across. Now you mentioned co-package optics. There's two things that people are talking a lot about now. Jensen put it on the slide at GTC. He had kind of that, the AI stack, I don't know what he called it, but it was like an AI stack. At the bottom, the physical layer, he had energy. Okay? Okay, energy's now like the physical layer. Energy and then latency becomes effective because scale up or scale across, you're going to want to do this in the fastest way possible. So working together, we've got to have almost real time speed of thought kind of latency 'cause that's the third scaling law. Talk about that dynamic between energy, savings, latency because you guys are getting that right. You obviously see it as a physical layer and that's bounded by energy. Everything's bounded by energy. Talk about that, those two pieces, because co-package optics, optical connections, in my mind I think, oh, this connecting thing's got a little connector. That could fail. Okay, too many in a rack. Anyone who's done networking cabling knows, okay, yeah, I put it in. Yeah, you do your best job but it could fail. But that does that interest latency? There's engineering involved. Explain the concept of the energy and the latency when you start thinking about these interconnects like co-package optics, like Spectrum-X and others.
Gilad Shainer
>> Yeah. So first there is good amount of engineering that goes into building those supercomputers, and I'm happy to work among very smart people at Nvidia that built in those engineering pieces in every place in that supercomputer. It's a privilege to me to work with those great people, really smart people. You completely right. One of the limiting factors of the capacity of a data center, it's energy. Energy will set how much compute you can build, you can bring into that data center. Energy will mandate that data center scale. Now in order to maximize what we have, in order to maximize the energy that I have in a single place, I want to make sure that everything is most energy efficient. That's why copper for NVLink. That's why copper for scale-up.
John Furrier
>> Copper is great.
Gilad Shainer
>> Copper consumes zero power. Now when we look on scale out in the traditional data centers, it didn't need to use much of scale out and therefore there was not much fiber. There was not much optical connections and the energy that was spent in optical connection was not that critical. When we are talking about scale out on an AI supercomputer, there is a good amount of data that needs to go between those racks on scale out and therefore there is a good amount of fiber optic connectivity, which mean that we're spending a lot of energy there. And you spend the energy because the distance needs to cover by light. So you're moving data over light, photons, but then you need to translate it to electrons in order to connect into the switch, in order to connect into the superNIC and so forth. And this is where you spend energy. Now we want to minimize that because today that energy can almost get to a 10% of compute capacity. So getting 10% more compute in an AI supercomputer computer, it's really important. And the most efficient way or the way to minimize energy consumption is to take those optical engines, those engines that on one side connects to electricity, on the other side goes to light and drive the signal over distance to take those optical engines and put them as close as you can to the switches. And if you do that, then you can minimize the energy consumption. Thing that I need to talk with you or to convert something and you kind of sitting far away, I need to spend lot of energy to reach to you. But if you sit close to me, I can spend the least amount of energy. That's why co-packaged optics moving the optical engine to be very close to the switch, for example, and I need to spend much less power and then I can optimize the scale out infrastructure.
John Furrier
>> What does co-packaged optics mean? Because in the extreme co-design, what I hear you saying is that you're making optics integral to the system like a race car. What does co-packaged optics mean? Is it just embedded into the chips? Describe what it is. What's the-
Gilad Shainer
>> So let's assume you have a GPU that connects to the NIC. NIC connects to the switch and then the switches connect to more switches between them. The way that a switch connects to another switch is that you have a box, a switch box, that box has ports which is interconnect ports and into that port you actually plug what we call the transceiver. And transceiver is an entity that has an optical engine. Sometime it has a DSP and has laser sources that actually create those light, and that transceiver will communicate to another transceiver that is plugged to another switch at the end. Now that transceiver sit outside of the box, and it includes a lot of laser sources. It may include a DSP.
John Furrier
>> It's like a translator.
Gilad Shainer
>> And it's an optical engine.
John Furrier
>> Yeah.
Gilad Shainer
>> And that engine is to drive the signal all the way into the switch itself and spend a lot of energy in doing that because there is a lot of single transitions until it gets into the switch basic itself. Co-packaged optics means that we take the optical engine that sits outside of the box and put it in the same package as the switch. So now the light goes all the way to the switch package and then from that point it's very close-
John Furrier
>> You cut a lot of the-
Gilad Shainer
>> And then you can cut energy, you can save almost 3.5X...
John Furrier
>> On energy....
Gilad Shainer
>> on energy consumption.
John Furrier
>> What about latency?
Gilad Shainer
>> Of course you because there is this component it needs to go through, and you also increase the resilience of the data center, which is very important. You increase the uptime of the workloads. You can increase the non-interrupted run time by five X using co-packaged optics because there is less components. Things are much more reliable, think are much more resilient. It sits inside of the package. It's an entire data center, becomes much more effective, and we're actually building a better AI data centers with both Spectrum-X Ethernet Photonics and with Quantum-X InfiniBand Photonics.
John Furrier
>> Everyone loves talking about F1. You mentioned racing. Of course all the tech companies have their logos on these cars. It's like that. You don't want to have a race car with all these off-the-shelf components. Essentially what you're saying is the way optical was before, you have transceiver. That was the state-of-the-art connection point, interconnect. Now you've embedded it into the switch, so you reduce a lot of that in between.
Gilad Shainer
>> Right. You make energy much closer.
John Furrier
>> Energy savings, much closer, energy savings reliability and also speed and latency. So you don't want your car to fail some little switch breaks. Moving data around, okay, that's great. Now as you look at the build-out, as we saw in GTC, the scaling loss, thinking is the new third power law scaling law. Jensen presented Nvidia's vision there. As you look at supercomputing coming up, you got a lot of content around, okay, CapEx build-out. Everyone's talking about that. Now the next wave is edge. I kind of teased out the conversations that there's going to be Microfactory soon because distributed computing works in a beautiful way. Get the centers down, factories first, hyper-converged edge, maybe a Microfactory in between talking to some satellites, whatever, a lot of networking. Networking is the fundamental future of all AI.
Gilad Shainer
>> Right.
John Furrier
>> And you're seeing that with the telecom announcements within Nokia. So I guess my question to you, as SVP of Networking, the aperture has changed. So you've got to scale up, scale out, scale across, co-package optics. Now you've got a whole nother sphere or realm of networking. What's your vision around some of these things that are emerging? Because the efficiencies with co-packaged optics could be co-packaged factory connections. If I have an edge device, why wouldn't I want to bring training and inference at the edge? I mean the Thor development I've been following is highly compelling. That's going to talk to the factory. What's your thinking around that?
Gilad Shainer
>> Well, and that's another example of co-design. And co-design, it's not just inside the data centers. It's connecting data centers, is go to the edge, as you mentioned. And how do I get to the edge?
John Furrier
>> Networking.
Gilad Shainer
>> It's networking again. Right? And they need to make it very efficient. We also introduced, by the way, BlueField-4. BlueField-4 is a data processing unit, and it's very key to AI supercomputers. BlueField-4 sits in several areas. One, it sits as the core of the storage infrastructure because it runs the storage stack, and in inferencing it's really important to be able to get access to storage and KV cache and a lot of area around storage platform. And that device runs the entire storage infrastructure on it. It sits on the compute server. It sits on the storage side. It's a great storage processor. It's also a security processor because it enables secure access by separating the infrastructure domain from the application domain. And it's a great device for the edge because you need to have more computing at the edge where you place GPUs at the edge and actually can have the BlueField-4 data processing unit to be the compute, the secured access, and actually also the storage processor as well.
John Furrier
>> Yeah. I mean, it's funny, the old days, not to update myself, but storage, compute, networking, they all had their boxes. They'd connect them with some ethernet. You do some things with it. You go to the storage, make a query route, route the packets from point A to point B, process them. Now networking's actually in the storage fabric too. So you got to talk the storage and memory, so HBM memory. So it is very clear to me that that has collapsed in with networking software as fundamental. You mentioned edge. I mean, my view is that that's going to hyperconverge with every spectrum and ethernet and essentially maybe a AI factory box. Jensen held it in his hand. He said AI factory in my hand, did a meme on that. And it's like why not sister that up against a wireless access point and run full multi-protocol, seamless access with AI embedded into the networking there. 'Cause I mean wireless is networking but they don't have AI. So I see that coming very, very quickly. That makes networking even more critical because that's really scaling across to the edge.
Gilad Shainer
>> Yeah. So networking, as you said, networking is key. And networking is essentially the heart of the data center. And when you design networking for AI supercomputer, networking for AI the edge, networking for AI for 5G, 6G, it's not that you just do off the shelf element or take off the shelf ethernet and just connect that. You need to carefully design something that enables you 100% synchronization across that entire supercomputer, across multiple supercomputers, across the supercomputer at the edge. And that's the core element when we designed the Spectrum-X Ethernet, the core element was to make sure there is no jitter, which means there is no skew of synchronizations between those compute engines. You build a full balanced system. You build a supercomputer. Everything needs to work fully synchronized as a single unit because if you have, for example, 100,000 GPU data center and only one GPU is delayed, only one GPU is delayed, all the rest are waiting, doing nothing, and waiting for that single GPU to end. This is why jitter is so critical. This is why off the shelf ethernet was not an option for building AI supercomputers, and this is why we created Spectrum-X
John Furrier
>> Orchestration, resource management, it sounds like an operating system to me.
Gilad Shainer
>> Yeah.
John Furrier
>> So, I mean, I think what I love about what you're doing is you're cracking the networking code in a whole nother way. And once OS is, I put that in air quotes, OS as a coordination layer, apps come. I mean, just think about the end user benefits of having AI anywhere from the edge. I could be at a retail outlet, okay, you could do facial recognition today. They can do inference. But imagine if I walk into a retail outlet and they say, oh, there's John. He's with Gilad. Oh. They're hanging out today. He's at Nvidia. He's the CUBE host. Instantly goes back, my agent gets my model, brings it to the edge. Now I'm having a completely different AI enabled experience that never existed before. And I could even go, hey, look at my Ring doorbell. How does that happen? So imagine that kind of intelligence from networking where you can span topologies and handle all that with AI under the covers.
Gilad Shainer
>> Yeah. Every data center is going to be accelerated. Every element is going to be accelerated and we need to make sure that we're bringing the co-design technology into every place. That's why we are working in a large ecosystem. So the technology that we built is being used by many of our partner and customers to enable AI technology, AI capabilities in every place. We built Spectrum-X Ethernet infrastructure that is being been used by OCI, that is being used by Microsoft, and they're running their operating system, SONiC, on top of that. It's been used by Meta, and they're running FBOSS operating system on top of that. So it's not just building something that is very synchronous. Co-design, it has a full flexibility that everyone can run its own operating system on top of that. The partnership with Cisco, it's a great partnership because we bring the AI capabilities of Spectrum-X Ethernet into the Cisco switches, and then Cisco enables their enterprise management across that entire infrastructure and enable enterprise customers that need that enterprise simplicity to continue having that enterprise simplicity with Cisco.
John Furrier
>> You guys are bringing core-
Gilad Shainer
>> But now everything is inside.
John Furrier
>> You guys are bringing core AI to other people, and co-design is a really pioneering movement and I think bringing that out in the open with the ecosystem, obviously in the supply chain, clearly that's been working for Nvidia, but Cisco has AI too. They have some AI action, but you're bringing Nvidia AI, factory AI to them to add value to their AI. And I just think it's going to be a great run. I think co-design's genius. The networking is, I think the fund, we're watching it very closely. I think that operating system concept of what networking will continue to do, but morph into a lot of AI inside the Nvidia system to create value. Thanks for coming on. Appreciate your time.
Gilad Shainer
>> Thank you for having me.
John Furrier
>> Great to have you, always, geeking out here with Nvidia. Networking is the future of all OS transactions with AI enabled infrastructure, and AI infrastructure's the hottest area. Beyond the data center AI factories, you're going to have a lot of other AI factory-like systems all talking to each other. They can't be waiting. They're going to be processing. The softwares on top, integrated it in. Co-design is the feature. We're doing our part to co-design content here in theCUBE. I'm John Furrier, your host. Thanks for watching.
John Furrier