 Welcome back everyone to theCUBE's live coverage, day three here at Mobile World Congress, MWC24. I'm John Furrier, your host with Dave Vellante, extracting the signal from the noises of theCUBE. Love going to the events. We've got our studios in Palo Alto in Massachusetts. We've got the hybrid studios, we've got on-site locations. Dave, we're hybrid. We're going to have a space studio someday. Bill Gardner's here, SVP and General Manager of Optical Systems and Optics Group at Cisco Systems. Obviously a huge part of the connectivity play. Bill, thanks for coming on theCUBE. Really appreciate your time. Thanks for having me. So Cisco has been connecting companies for you. Yesterday on the close, I was talking about how much I love, how Cisco has the most successful business model of all time. You connect things, it's hard to unplug Cisco. When you're already connected. So Optics is one of them. So congratulations on that. But as we go forward, there's a huge opportunity on interconnects, new cluster management, clusters are being developed, clustered systems as we call them, from theCUBE research is coming out as new ways to create these new computing environments that's bringing data and AI together in a distributed computing connected way. Yes. Check. So take a minute to explain what you guys do. I want to go into some of the investment areas and implementation changes now going on. But explain what you do. What is your group that you're running do? What's, who are you targeting? Who are your customers? What products do you have? And what's the, how you organized? Thanks. Thanks again for having me. So I have two main business units. One we refer to as the Optics business unit and the other is the Optical systems. The Optics business unit is basically the transceivers that we sell with routers and switches that find their home inside a data center or inside a central office if you're in a telco environment. And that world is really characterized by two attributes. One is that the fiber is plentiful, meaning if you add a new router or switch, you pull new fiber to every port on that router or switch and you put one signal on the fiber. And the other attribute is that the distances are short and in this world short means less than 10 kilometers. So that serves really all market segments for Cisco, campus, enterprise, commercial, public sector, service provider and web. And we sell those transceivers for Cisco applications, Cisco's and routers and switches, and also third party. So customer can buy our optics and plug them into a competitor router or switch. And that's basically a transceiver business, little transceivers that are about the size of a cigarette lighter that plug into a router or switch and contaminate anywhere from one gig to 800 gig today, whatever a customer might need for their given application. That's going to have a huge, huge impact for AI because interconnecting all of these workloads is going to be done optically with these transceivers. Now once you leave the data center or central office, now the problem's very different. You have to now send that signal over many hundreds or thousands of kilometers across a city, across a country or subsea between continents. That's a very different problem. And sending an optical signal 100 meters or a few kilometers is very different than sending it thousands of kilometers. And so that's a much more sophisticated technology. Traditionally it's chassis-based solutions that allow us to process an optical signal and send it, for instance, from New York to LA. Very long distances, we have to preserve the integrity of the signal as it's sent over those distances. And largely speaking, it's the service providers that are customers of that technology, service providers and hyperscalers who are sending massive amounts of capacity between sites, very complicated. That's got two very different attributes. One is that the fiber is limited, meaning if my customer in LA and my customer in New York put in new routers, I can't ask them to pull new fiber between LA and New York. We've got to cross the Mississippi, we've got to cross the Rockies. And so we have to deal with the fiber that is in the ground, which means we have to put many, many signals on one fiber. That's a complicated problem and that's using a technology called DWDM or dense wavelength division multiplexing. And the other attribute is that the distances are long. Very long, it could be hundreds or thousands of kilometers. So those are the two worlds. The optics world is inside the data center, optical systems outside. Right, so it's a synchronous and asynchronous world. Right, that's how my data center brain thinks about it, so okay. So basically it's a contested, highly contested and shared medium on the long distance side, okay? And congested too, contested and congested. Exactly. So routed optical networking seems to be the key technology. Why is that important? Can you explain what that is? What is routed optical networking? Routed optical networking was really born out of an insight that we had that was driven by the fact that we're seeing massive, massive scalability now with new silicon technologies. Cisco has a technology that we refer to as Silicon One. It's based on an acquisition we did five or six years ago. The acquisition was Liba and we've developed now a complete portfolio of routing products that leverages silicon. When I came to Cisco about 15 years ago, we had a 40 gig line card on a router and we had 14 ASICs that were required to deliver that 40 gig signal. Today we have one ASIC that delivers 25 terabits of capacity. And if you think about it, the cost of an ASIC is about the cost of an ASIC. The cost per bit today for that 25 terabit signal is much, much, much lower than the cost per bit of a 40 gig signal 15 years ago. And yet as we think about that, we think now does the architecture that we laid out for service providers and for large enterprises 20 years ago still make sense given that the cost of silicon has driven a much, much lower cost per bit in the network. And what happened 20 years ago, as we said, the router was the most expensive thing in the network. And so we did everything possible to bypass a router whenever you could. And we built a really complicated optical layer underneath this IP layer to basically bypass routers with sophisticated optical switching technology. And that made economic sense. So it wasn't a bad thing. We created an entire industry around that. But- Was that driven by cost pressure or latency pressure or both? It was driven entirely by economics. It was the right thing to do from a cost perspective. So we built this sophisticated optical layer that allowed us to bypass routers. So if you had to go ABCDE through routers and your signal was really going A to E, you didn't want to go through BCND because that was too expensive. So you went around BCND with an optical wavelength. And as we step back now and say, are the assumptions that were valid 20 years ago still valid? You say, no. And in fact, in many cases, it makes sense to go through BCND now, not around them, because the router is no longer the most expensive thing in the network. It's actually the optical interfaces that are more expensive than the router port in many cases. So that was one unique insight. And that's really driven our thinking around router optical networking. It says we should rethink how we're deploying capacity in the network. And rethink, when does it make sense to bypass? And when does it make sense to go through hop by hop? What was the results of that? So that does that change the usage and then target application aspect or use case? What it does is it changes the way we deploy capacity in the network, what gets done at the optical layer and what gets done at the IP layer. And there are really significant economic gains that can be had if we think through that. Now, there was one very significant technology that came along that really helped our thinking as well. And Cisco acquired a company three years ago this month. The company was Acacia. They are based in Mainer, Massachusetts. And Acacia was a leader in coherent technology, which is the optical technology used in the long haul networks and the metro networks. But they were also a leader in driving down the power required for that technology. And what they've enabled is in the old world, optical systems were chassis-based solutions with transponder line cards in those chassis, big solutions, very power-hungry solutions. Acacia's taken that technology and replaced a line card with a pluggable optic that can now go directly in a router. Okay. So today we can have a 400 gig pluggable optic that can go a thousand kilometers and replace the need for that chassis. Now that drives really significant capex and op-ex savings for customers. So was that the innovation that allowed you to lower the cost of the router? Or were there others as well? So that is, the silicon innovation has really lowered the cost for a bit on the router. This innovation lowers the cost of having that router signal transmitted over a long distance. So as an example, we've had independent analysis done that says nominally for a given network, a customer will save 35% capex by moving to this pluggable technology. But in some cases, even more importantly, they'll save over 50% on op-ex. And the reason they save 50% on op-ex is because number one, they get massive power savings. A transponder sitting in a chassis is about 250 watts. A pluggable going into a router is about 22 watts. That's 90% power savings. Yeah, right there. So if you're a telco or a hyperscaler, I mean that is, both the capex and op-ex savings are just ridiculous. Massive, massive, and with equivalent performance to what they were getting a transponder. So the economic arguments are just so compelling it's hard for anybody to acknowledge. Essentially no performance trade-off is what you're saying. There's no performance trade-off. So it's a no-brainer basically. It's our belief that it's a no-brainer and it's our belief that this technology will be the dominant technology deployed in data center to data center applications and in metro applications and even in long-distance applications over the next three to five years. And we're starting to see that now. So Bill, take us through where we are because obviously no-brainer. Is it available now and you're rolling it out? You're trying to get the meetings, get those attention. So that technology has been available for just a little bit over a year. I would say it's had very, very high acceptance in the hyperscaler community because many of the hyperscalers for things like AI loads that you mentioned are seeing massive capacity needs in moving traffic between data centers. And it's a perfect solution in that application where it's relatively short distance, maybe 100 kilometers up to 1,000 kilometers between data centers. But we've got over 100 service provider customers now deploying this technology as well. Take me through the, I can almost imagine in listening to this as to the way you're explaining it, it's like, okay, I'm sold. No-brainer, right? Is the challenge just getting the attention of your customer base saying, hey, guys, you got to save you money? Or are they all jumping on this now? I think the economics are indisputable. So we don't get a debate from the carriers about economics. I think the key challenge is actually not a technical one. It's more of an organizational challenge because historically the service provider might have an optical team that manages the optical systems and manages the process of deploying and installing and managing that network. And a separate IP team that's managing the IP world. And now what we've done effectively is take a piece of the optical world and moved it into the IP world. Yeah, you got to kick the tires, look at the do-deals. Okay, so now I'm going to put my little network hat on. Well, that organizational point you're making, we've seen this before. Yeah. Right, where they think about with the telephones. I mean, if you didn't converge those organizations, it was a nightmare. And so that, I think, that's a challenge is that we have to converge the processes between these two silos. And in some cases, maybe converge an organization. That, I think, is the challenge. It's not a technical issue. It's an organizational thing. But I do have to ask one, as a network operator, what my network operator hat on, well, I want to know what the disruption is to me. It's okay, organizationally, obviously, there's little, you got to make that happen, that's cultural. On the implementation side, is there disruption? What's the swap out look like? What's the switching costs? Great question. So this is, to be very clear, it is not a rip and replace proposition. It is a proposition that says they can take their existing network where they're using transponders and very traditional approaches. And by the way, we continue to sell transponders to those customers that want to buy them. All they have to do is start deploying the pluggable. They can deploy that side-by-side on the fiber with transponders, so it is not rip and replace. Clearly, for a greenfield application, it's a no-brainer. It's very, very compelling. But for a brownfield application, where service providers have already deployed capacity on a fiber, maybe they've got 10 wavelengths deployed on a fiber, what do they do for the 11th wavelength? We're saying you can start deploying that pluggable technology with no disruption. And do they prefer to let the sleeping dog lie, as they say, and keep the existing stuff, and then maybe kind of slowly kind of migrate? Is it more? Yes, and we would encourage them. I think you encounter much more resistance if your argument is take this stuff out and put some new stuff in. Go at your own pace. Go at your own pace. And frankly, it's good equipment that has a long life associated with it. So they should just milk that. Appreciate it assets, yeah. You answered my last question, which was the implementation, then what changed with AI? Go at your own pace, provision at what you need, if you got more workloads. Guess my final new question would be for you, as an industry leader, you oversee obviously this business as part of the plumbing, critical part of the plumbing. You're involved in all the aspects of it. What do you see, put your industry hat on, and then your Cisco hat, wear two hats, and look at this question from that lens. With AI coming, the big conversation here on theCUBE has been, how does AI change the architecture of my computing environment one? And the network is playing a key role. What has to happen at the network to make some of these new clusters, these new implementations work? Yeah, so I think the question really has a couple different angles to it. One is what's happening inside the data center and the AI architecture there. And I expect that that is going to continue to evolve. Today we have, for instance, discrete optics that are interconnecting the GPU with the switch, and that will continue for some time. That may be an 800 gig optic today. That's probably also one of the areas that we have to look at to reduce power, because as we start to deploy more of these AI workloads, the power consumption needs of the data center become a really significant obstacle for how do you continue to scale that capacity? So I think finding ways to drive power out of that solution, maybe packaging the optic with the silicon or the GPU will make sense over time. That's a long-term problem for the industry. That's, we're going to see some evolution of that technology inside the data center. Then outside the data center, it is how do you scale the capacity because we're going to have these AI workloads that are driving much more significant capacity outside the data center. I think the pluggable optic technology that I mentioned earlier is a big part of that. That can help carriers do it in a much simpler way. And one of the things I think we have to recognize is a big challenge for our customers is their capacity needs are growing much, much faster than their ability to scale the operations. And the answer can't be just scale your operations in line with that capacity need. It has to be, we automate that so that we can simplify their operations. Exactly, they get no operating leverage if they have to get the operating leverage. And the only way to get that is through automation. So I mean, as an example with routed optical networking, the answer isn't take your optical team and make them IP experts and take your IP team and make them optical experts. The answer is to automate out those things that really require human engineering and let those people be the experts in the areas that they need to be experts. Bill, great to have you on theCUBE. Love that last bit there, great insight. And congratulations on your business opportunity. Thank you for having me. I've always been a big fan of, especially the long haul. We need to move that bandwidth. No one wants more bandwidth. Got to have, bring me more power, more bandwidth. I don't think we'll ever see the end of that. It's like Star Trek, more power Scotty. Word. Bill Gardner, SVP and general manager, I appreciate your time here, bringing you all the action. Mobile world, I'm John Furrier with theCUBE with Dave Vellante, extracting the signal from the noise, bringing the best commentary and analysis from our CUBE research team theCUBE at the Silicon Angle, here inside the Barcelona studios from theCUBE. We'll be right back after this short break.