theCUBE + NYSE Wired present the East Coast AI Leaders Executive Series | Shekar Ayyar, Arrcus

Clips
More from theCUBE + NYSE Wired present the East Coast AI Leaders Executive Series

Shekar Ayyar

CEO

Arrcus

play_circle_outline Arrcus: Revolutionizing Networking for Growth Startups in AI and a Comparison to the 1986-1996 Era

play_circle_outline Arrcus role in transforming networking with common abstraction layer

play_circle_outline Revolutionizing Efficiency: Ethernet Architecture's Impact on Distributed Computing and the Intersection of Compute and Networking

play_circle_outline Arrcus as underlay and overlay network, importance of partnership with chip players

play_circle_outline Customer benefits of tapping into Arrcus fabric, programming functions on top of the platform

play_circle_outline Unlocking Customer Success: Programmability Driving Speed in Infrastructure and App Development

Info
Transcript

Shekar Ayyar, Arrcus

Shekar Ayyar

CEO Arrcus

TheCUBE is at NYSE location, CUBE East for Media Week, focusing on growth startups in AI. Shekar, CEO of Arrcus, a networking startup, discusses the current super cycle in networking and the need for innovation in networking infrastructure to support AI growth. Arrcus provides a platform with an abstraction layer for different network functions and workloads. The evolving landscape of AI applications demands efficient networking solutions, connecting data centers, edges, and endpoints. Arrcus collaborates with chip players like Broadcom and NVIDIA to offer a ... Read more

explore Keep Exploring

What company are you involved in, and what super cycle in networking are you referring to? add

What unique value proposition does Arrcus offer that is similar to what VMware did with compute abstraction? add

What are the challenges associated with creating efficiency within a rack in a data center and in between racks and data centers? add

What is the significance of being both an underlay and an overlay network in the context of networking challenges and the GEN AI wave? add

What is necessary in order to cover the entire spectrum of architecture, from simple switching to complex routing? add

What is the foremost use case for programmability in today's rapidly changing technological landscape? add

bolt Powered by CUBE AI

Shekar Ayyar, Arrcus

search

>> Hello and welcome to theCUBE here in the NYSC location, CUBE East. I'm John Furrier, your host of theCUBE. We are here for Media Week all week here, breaking down all the hottest growth startups in AI. The AI leaders on the East Coast, we did a session in California in our Palo Alto studios. We're kicking off the week here. Shekar is here, CEO of Arrcus, CUBE alumni. Last time he was on theCUBE was at VMware Explorer. I called it VMworld back then. Shekar, great to see you.

Shekar Ayyar

>> Indeed, great to see you again.

>> Your founding partner over there at Arrcus is here as well. Great to see you.

Shekar Ayyar

>> Thank you so much. Thank you for having me.

>> So you're involved in Arrcus, which is a startup that's been baking out for a while, obviously, in the networking area, which we're going to go deep in, a deep dive into. I want to get into the super cycle we're in. Some people don't like the word. I actually like the word because last time we had a super cycle in networking, you go back to 1986, that 10 year period from 1986 to 1996, really probably was one of the biggest transitions in the history of computing and networking. We went from proprietary to open systems and the rest is history. Cisco was formed in '84. They were two years in. They hit that wave perfectly and then that spawned a massive revolution of wealth creation, entrepreneurship, technology innovation, all grounded in open systems that was precursor to the web. The web hit the scene, and then, again, the goodness of open continues to thrive. We are in a similar super cycle right now, and so for the next 10 years we look out. Just put a mark in time. Just say the ChatGPT moment. Go back two years, let's say two years in or now, whatever, for the next 10 years.

Shekar Ayyar

>> Indeed.

>> And that's why people were saying it feels like the '90s. So you talk to anyone around NVIDIA, Arrcus, this is huge, and you've been through the movie before.

Shekar Ayyar

>> I have, indeed. Yeah, yeah.

>> Tell us what your reaction is to this super cycle.

Shekar Ayyar

>> I think absolutely, as you characterized it, internet web drove the last generation. There was clearly a shift in compute. We went from servers to virtualization to cloud. We absolutely had a networking boom when Sun Computers and all these guys came out and took advantage of better networking. I think AI is obviously now the next great buzz, but also the reality of how things are going to get involved calls for a real refresh in terms of how the underlying plumbing and networking is working. So I see this as a moment in time where we're all going to look back about 10 years from now and say, "Wow, that was the time when, in fact, network took the next giant leap."

>> You had an event last night, private event here at the NYSC. You had lot of investors. You had a lot of potential customers there. Maybe even some customers. We really weren't checking on that, but for the most part, you're one of those companies where it's like you are in the right area. The old expression, the world spins to your doorstep. Explain Arrcus. You guys were founded pre-AI moment, but you guys were really doing in some really hardcore work, ex-Cisco pedigree, a lot of background in networking. You were at Vmware. You ran all that. I think you did the deal, if I remember correctly.

Shekar Ayyar

>> Yes.

>> That created NSX as we know it. So you were on that side of that telco side. Now you've got the networking all coming together. This is where the action is. Talk about Arrcus, why it was founded, and what you guys are doing right now.

Shekar Ayyar

>> So if I just flash back a bit to the VMware moment, if you look at what VMware did really well, we created an abstraction on compute. It allowed for people to bring different kinds of compute workloads on top of different hardware infrastructures and then make it a level playing field for everyone to go and become applications on top of infrastructure. What Arrcus has been doing, and this is the reason why I loved the company when I came on board as CEO from VMware, is it had been building already a platform with an abstraction layer that essentially allowed for different network functions and workloads to be dropped on different form factors and forms of hardware underneath. So very similar to the VMware value proposition, but brought now to networking. And if you look at how networking has evolved over the '80s, '90s, 2000s and 2010s, it has really lagged how compute has evolved. So you have a very efficient compute infrastructure, everything from on-prem to cloud. We're now into high performance computing, GPUs, et cetera. But if you look at what's happening in network, pretty much every CTO, CIO still buys a box, and the box comes from some large incumbent vendor and they don't quite have the flexibility and ability to go in and stitch these things together with a common abstraction. So Arrcus started out as a company that said, "Look, we're going to change the world of networking, we're going to build that common abstraction, and we're going to make it available really to every customer for every application and every use case, which is a very large hurdle, a gigantic, monumental task to take on. And we've done that, we've done that successfully, and of course today we're at this moment where we are looking to now grow.

>> There's a lot of conversations that have a pretty large wide aperture that we cover on theCUBE, on the DevSecOps side. As you see that on the software side, open source developers are booming, but the revolution is going on at the compute networking layer, and that's why NVIDIA's stock prices at an all-time high. It's clear that the compute and the GPUs, XPs, use whatever processing units are out there, it's going through a mass change, changeover. We call it clustered systems. It isn't the server box anymore, it's the servers in the chips and around the chips. It's what's around it. It's the ethernet. And so this really forces your hand at Arrcus on the industry in general because everything's getting faster, smaller, cheaper. Sounds familiar?

Shekar Ayyar

>> Yes, exactly.

>> The networking has to be faster and distributed.

Shekar Ayyar

>> Indeed.

>> This is the real architectural thing that will probably spawn more investment. So take us through why that's important right now. Obviously, NVIDIA's an investor in your company, so that's a pretty big deal. Why is it important that this flash point of compute and networking and this fabric clustered power is coming, and why is networking so important?

Shekar Ayyar

>> Yeah, so I'll break this up into two parts. One is, how do you create efficiency within a rack in a data center, and then how do you create efficiency in between racks and in between data centers? Both problems are actually super critical and important now. If you think about how in particular infrastructure is deployed, data centers are created, the historical view was to build these racks of general purpose compute and then use them effectively for all kinds of different generic applications that were delivered. Cloud became a big injection point and inflection point into how these data centers grew large. Then along came the edge. We talked about edge applications, IoT, and what happened at the periphery of both the service providers as well as the enterprises, but couldn't quite nail a killer use case for the edge. I mean, lots of things happened and people came up with technologies. Now, if you think about it, AI essentially becomes a perfect inflection point for all these things to come together. You can no longer build just racks and racks of high-performance compute infrastructure and look to only network the close proximity clusters and say that our job is done. You really have to take that from there to the top of rack. You've got to connect between racks, between data centers. And so when we look at what we do at Arrcus, in particular our ethernet architecture, our contribution to everything from IETF to UEC, et cetera, we see this as a perfect fabric for extending what happens in the core of the AI substrate to a distributed architecture. This also then takes it from core to edge, from data center to the endpoint, from the carrier core to the aggregation points and cell site routers. So think about the entire world of network, high-performance computing now as spreading out and requiring the perfect high-efficiency fabric, which is where our partnership and investment from NVIDIA, et cetera, comes in as well, because that then brings in the best in compute and high-performance infrastructure to a leader in innovative networking, as we are doing at Arrcus.

>> NVIDIA, obviously, is the leader. Their stock price is massive. Their success has been massive. Jensen Wong said at his last GTC conference, the democratization of super computing is here.

Shekar Ayyar

>> Yes,

>> You mentioned HPC, high-performance computing. That was also a very narrow business for the highest customers. Boeing wants to design a new wing. They'd have hours of compute time, almost like sign in and use it and then someone has to wait. Now, when you have this computer revolution coming with AI, in every era that I've seen in my career over the past 30 plus years, whenever you bring more compute and networking to the table, people take advantage of it. And we're seeing that with AI and the use cases come out and clearly from that new use cases and probably other workloads. So what is the benefit that's going to come out of this? Because, again, we're in a super cycle next 10 years. We'll probably be there by the next five. You guys are in the middle of it. You guys hope to be that growth. What are the use cases? What are those workloads? What's demanding that power? What's going to take advantage of it? I mean, we had some people here at the NYSE and during climate week doing stuff with the climate change and the biosphere instrumenting the computing power.

Shekar Ayyar

>> I mean, first of all, if you look at just the applications of AI, as you have seen probably from numerous reports that come out, including from folks like McKinsey and Goldman Sachs and so on, pretty much every vertical is looking to get the advantage of AI. Some that are getting dis-intermediated pretty quickly and heavily and others probably a bit slower. But if you think about everything that's happening in financial services, in autonomous driving, in industrial automation, in oil rigs, et cetera, the entire gamut of what these industries are going through, they're all relying on how you can take AI in general and then GEN AI in particular and then bring that to the advantage of their use cases. Then if you dissect what's happening specifically in the AI world, a lot of the action is actually moving from training to inferencing because, of course, there are giant training models, LLMs, that are already in place and people are improving them or on them. But now the realization is, how do you take these large models, how do you then dissect these or complement that with something that's more vertical specific? And then furthermore, geographically, these endpoints and the distribution points and the inference points are all now going to be located in various different places. They could be at the edges. They could be globally distributed. So this then calls for efficiency in the connective tissue and the architecture and the fabric that brings everything together. So how are you going to ensure that your highly-priced GPU compute resources that you now have brought essentially by paying a lot of money to whoever's supplying those to be used for just the high-performance needs? How can you then offload something from there to be computed by, let's say, a smart thing? So this is where our work with NVIDIA bringing on top of NVIDIA's Bluefield DPUs comes in, because we're now jointly presenting an architecture to customers that shows how they can not only leverage their DPU stats, but also take stuff off of it and then do that more efficiently. Then you have the question of, how do you take something and then move it all the way to either a carrier edge or a Colo edge or an enterprise edge? And that edge can then become a micro-data center because it has the combination of compute network, and then you also then need to interactively then connect to the large model to go in and say, "How am I going to train and basically pass results back and forth at the same time?" So you can now see the importance of why the pipes and plumbing and whatever wires that connect these things together are incredibly or sometimes even more important than the compute stacks on which the GPUs sit on. And this is the fortuitous moment that we find ourselves because we've been innovating, as you said, even before the AI boom came in.

>> Yeah. What's interesting is that there's always been these networking challenges, routing tables, coherency, flapping, convergence, all these things that just make packets work.

Shekar Ayyar

>> Exactly.

>> Now you complicate that even further when you go to the GEN AI wave, where you move up the stack where you have data harmonization. Harmony is the gift of GEN AI when it's working because you have distributed computing. You mentioned edge, cloud on-premise edge. Okay, it's awesome. Now, before we get into some of the pedigree on your team and some of the advisors and investors, you also have some top customers. I want to get into those use cases, if you don't mind. But first I want to get a little bit technical because my first impression of Arrcus when I was looking at you guys was, "Wow, you guys look great." I thought you were an overlay network only, it's a technical term, but you're not. You're an underlay and overlay. Kior mentioned that, he's the founder and CTO, last night. Why is that important? I think this is a distinction that's important because you're working with all the chip players, Broadcom, NVIDIA. You're creating a harmonization layer between systems and ultimately enabling applications and data to work together. This is an important, very nuanced point, but it does connect the dots for me. But if you can be an underlay and an overlay, you see a lot. Why is that important and what's that enable?

Shekar Ayyar

>> Yeah, so think about it, again, I think the analogy I would have is through a lot of the work we did at VMware. So trying to simplify this a bit, think about us as going as deep as possible into the hardware processor and silicon architecture as we would be allowed to. Okay, so to the extent we can work in deep partnerships with the Broadcoms and the NVIDIAs of the world, we go down to the primitives and essentially start building out our operating system environment from there up. We're building the entire BGP stack, the entire route infrastructure, all homegrown, and these are being done by experts who have done this before for companies like Cisco and Arista and so on. And so the reason that is important is because we get to have the best of both worlds. If somebody comes up with a processor that's, whatever, 800 gig or terabit speeds, we would take advantage of that and then put our best operating system software on top of that so that the customers get the best of both. On the other hand, if you look at somebody as a customer and they say, "All we need is a simple switch at one end of our infrastructure, and we really want to basically make this a simple tour or even a management switch," we can effectively narrow down the applicability of our operating system right down to the basics of the silicon and then add that to make this a very high performance simple switch. The only way you can do this entire spectrum of architecture, all the way from simple switching to complex routing is if you do what I just described. You have to be able to get down to really the knits of what the hardware architectures offer us. We have to work closely with the hardware vendors in order to be able to do that. And then the benefit of this to customers is they see an interface that essentially they can take comfort in the fact that-

>> They can tap into it.

Shekar Ayyar

>> They can tap into it. They can basically create services. They can build functions on top of it. And we are evolving ourselves. So not everything that Arrcus has built at the operating system level is already exposed through an API, but as you look at our company growing and as you look at the customers adopting, the fabric itself is becoming or will become more and more a basis for our customers to start programming functions on top of that.

>> If I can translate that for the folks that aren't in the networking business, think of your Windows operating system. Every time a new processor comes out, it runs faster, gets better, they update it quicker. You don't have to wait two years for the new version.

Shekar Ayyar

>> Exactly.

>> That's been the problem in the past. You guys are more agile. You can move your OS around, keep that harmonization, but also take advantage of what's next and have that backward compatibility. Is that right?

Shekar Ayyar

>> Absolutely. And inside the OS, and if you dissect this and we have a nice picture that shows you the services that we can break our OS into, there's a component there called the D-PAL or abstraction layer for the data plane, and that essentially allows us to bring the Arrcus OS on top of a new hardware pipeline, even a new hardware vendor for that matter, quicker than anybody in the world can do this. We can do this faster than a Cisco or an Arista or a Juniper. And this allows us to become so tightly collaborative with our customer that they could come up with a set of new functions and services that they've dreamed up yesterday, bring that list to us, and our engineers are then able to go in, and if we don't already accommodate that in our OS, then we can actually upgrade that to get that in quickly.

>> Well, I wish we could go for an hour, but we have a big day ahead of us here at our Media Week. Let's get into some of the other points I want to highlight because I think it's worth noting and calling out. They always say, as an investor, big market, great team, go after it. You're going after a big market. Obviously, enabling is this super cycle. It reminds me '86 to '96. During what's created a massive set of entrepreneurship and wealth creation and opportunity and innovation. I think the next 10 years are going to be good. It's going to happen in the next five years.

Shekar Ayyar

>> Yep.

>> That'll set the table for the next 10. You've got a great team. Key are your partner. He's the founder, CTO, ex-Cisco, distinguished engineer. He knows networking, so there's no doubt there. Your career, you've connected the dots. You've done M&A, you've been technical, so you can see the puck that's coming. Skate to where the puck is, another cliche. I love, I'm a hockey fan, Wayne Gretzky, who said that. But you've got some great advisors. I see Basque Ayres on there. You've got some really big customers, CTO of AT&T. You've got good, healthy investors. It's almost like you're moving fast in this AI world with the classic, I won't say old-school investment thesis, but the reliable one. Good partners, funders going to back you, great team. This is going to enable you to get into a lot of customers. Take me through the customer value proposition because, again, Cisco took a couple of years to get baked into these companies. Their use case was simple, connect buildings and campuses, and next thing you know, PCs are on desks, they're connected on LANs, packets are moving, people are buying switches. The rest is history. What is your version of that in this super cycle? You come in, you knock down a simple use case. Why would the CTO of a big bank, big network work with you?

Shekar Ayyar

>> I think the foremost use case really is programmability. So if you think about what is happening today, the pace of technology change is tremendously fast. I mean, before you know it, you've gone through cycles of whatever, IoT and crypto, and we're into this hopefully long-standing cycle of AI. But things change extremely fast. And so you cannot have an infrastructure that is basically a static, non-programmable legacy box that you find hard to create, manipulate, change, and make new applications around on top of it. So fundamentally, our customers and our customers, again, not all of which we can disclose publicly, but for example, many of the top market cap companies, the top 10 market cap companies, three or four of them are some combination of customer, partner, investor, or all of the above. These are the kinds of customers that they are looking at taking the best advantage of all available infrastructure, the latest in technology, and they want to move fast. I mean, they want to build supercomputers fast. They want to build data centers fast. They want to launch applications fast. Right here, NYSE trading floor, you know the speed at which things move. So programmability coupled with the speed of the underlying infrastructure and the speed at which applications can be created and launched is paramount and high importance for all these customers. If you then take it from there, of course, then the next level issues around things like reliability and cost and ability to inter-operate and operate in green field and brown field environments, supply chain, getting diversity across vendors and not being locked in. These are all value propositions that we also then bring on the table and add to the value proposition of our customers.

>> Yeah, I mean, I think data networking was a classic. Data processing was an '80s thing too, all coming back because LLMs and the foundation models are being trained now. That's like going to school, but once you get out of school, nobody stays in school forever. You go out and you reason in the real world.

Shekar Ayyar

>> Exactly.

>> And so I think the killer app of inference and agentic systems, which will be embedded in applications, is going to have to use data and be harmonized with distributed computing architecture, of which you're enabling. I think you are going to be the killer app from a networking perspective because the inference has to be done at the point of the human and or device, machine, wearable, whatever you got. It's got to have compute and everything tied in there. I mean, this is networking concepts applied to data.

Shekar Ayyar

>> It is. And if you think about it, a simple switch or router used to be doing just a simple job. Switch went on or off, and a router basically sent a packet in one direction. That's really what they did. Now, when you start looking at the technology advances, in particular, we're working with SoftBank on this thing called SRV-VI mobile user plane, and when you look at the power of the SRV-VI mobile user plane, it actually allows you to create things like network slices, et cetera, using an IP-based programming methodology and then deploy that on top of your mobile or 5G infrastructure. So something like that is now enabling customers to go in and deploy an application or a service on their infrastructure by bringing architectural both innovation and simplicity into this network. So this gives you an example of, if you had a few more packet bits to code, you could impose a lot more policies, both operational and security on top of that. And this makes your network such a rich environment for therefore manipulating the workflows and the data flows that happen at that level.

>> Shekar, I'm a fan. If you're out there watching and you're in networking, you're going to be important because it's moving up the stack, GEN AI and all the graph and databases, there's just nodes and arcs. It's going to create more traffic patterns that we haven't seen before that you guys are going to take care of. Thanks for coming on theCUBE here for our Media Week. Appreciate you.

Shekar Ayyar

>> Thank you.

>> All right.

Shekar Ayyar

>> Wonderful.

>> CUBE alumni here inside theCUBE Studios in the NYSE, part of the NYSE WiredCUBE Open Network. Of course, we're doing wall-to-wall coverage here all week. I'm John Furrier, your host of theCUBE. Thanks for watching.