Hello, everyone. Thank you for joining. I'm Heather, your Webex host for today's event. In a moment, I will turn things over to the speakers, but before doing so, I have a few housekeeping notes to cover. Please note you've been muted. Should you have any questions, please submit them using the Q&A panel down at the bottom right of your screen. When submitting a question, please be sure to select All Panelists, and if you happen to experience any issues, please reach out to me directly. With that, let's get started.
This meeting is being recorded.
Thank you, everyone. This is Sami Badri. Actually, can you guys hear me just fine, Heather?
Yes.
Okay, great. Sorry. This is Sami Badri, Head of Investor Relations and Market Insights at Cisco. I want to welcome everyone to Cisco's Tech Talk on our Optics and Optical Business, live from OFC in San Diego. I would like to introduce today's speakers. Bill Gartner, currently our Senior Vice President and General Manager of Cisco's Optical Systems and Optics Business within the mass- scale infrastructure group. We also have Simon Leopold, Managing Director and Senior Analyst covering the technology and networking sectors at Raymond James. Bill will give a presentation, after which there will be a Q&A session hosted by Simon. If you have a question, please enter it into the Q&A chat box, which you can find at the bottom right-hand of your screen.
Before we begin, I'd like to remind everyone that today's call relates to Cisco's optics and optical bu siness only, and that no new financial information regarding Cisco's overall performance is intended or implied. We may make forward-looking statements, which are subject to risks and uncertainties outlined in detail in our documents filed with the SEC, including our most recent filings of Form 10-K and 10-Q. Actual results may differ from statements made today. With that, let me turn it over to Simon, your hosting analyst.
Great. Thanks a lot, Sami. Appreciate you guys having me host this with Bill. Something about OFC always feels a little bit like coming home, so certainly it's always exciting here, and a good opportunity to dig into the world of optics. So, Bill, I understand you've got some prepared remarks to give folks an overview of the optical business at Cisco, and then I've got a few questions I'd like to ask, but we'll also give the audience a chance to ask questions as well. But with that, Bill, let me hand it over to you.
Oh, very good. Thank you, Simon, and thanks for hosting this, and thanks, everybody, for joining. I'm just gonna provide a little bit of an overview of the business and, kinda where we play. And again, as, as Sami said, please enter questions into the chat, and we'll be happy to take them live. I'm sure Simon's got a few questions wrapped up as well, but please feel free to ask questions via the chat. So let me just start by framing the world in which we play, and just, just to define some of the terminology that we use as well. The optics business for Cisco is really what we're talking about when we're talking about inside the four walls of a data center or a central office, if it's a telco environment.
In that environment, it's really characterized by two things. One is that 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 every fiber is carrying just one signal, whether it's 100 Gb or 400 Gb or 800 Gb, it's just one signal on the fiber. The other attribute is that the distances are short, and in this world, short means less than 10 km. That market is served with transceivers that you can see in the graphic there, that plug into the router or switch ports, historically using a technology called direct detect, which basically means if the light's on, it's a 1. If the light's off, it's a 0.
It gets a little bit more complicated than that, but the game is basically turn the light on and off as fast as we can. And that's been sufficient for typically the less than 10-km applications. Now, once you leave the data center or the central office, now the problem becomes much more challenging. Now, you're sending a signal across a city, or across a country, or even between continents on subsea links. That's a much more challenging problem, in part because fiber is scarce, meaning if we add a router in, in L.A. and we add a router in New York, we can't simply pull new fiber between those two routers. We have to use the fiber that's in the ground, and that implies that we have to put many signals on one fiber.
So it's we don't have the luxury of just putting one signal on a fiber, we have to put many signals on that fiber because it's a very precious commodity. And the other attribute that makes things challenging is that the distances are long, and in this case, long means more than 80 km, but can be up to several- thousand km. And those two attributes, the fact we have to have many signals on one fiber and send them over very long distances, require a much more complicated technology known as coherent or DWDM technology that historically has been delivered in the form of chassis, as you can see in the right-hand graphic there. Those are two chassis that Cisco offers. The lower one is NCS 2000, the one above it is NCS 1000, and those chassis-based solutions have been deployed by carriers and large enterprises worldwide.
The very interesting dynamic that's occurring is that we can now take the transponder that sits in that chassis. The transponder is the thing that actually generates the wavelength that's carried on the fiber, and we can put that into a pluggable form factor. And so that is a very interesting dynamic for Cisco. We're, we're gonna talk much more about, but it is the dynamic that really motivated our acquisition of Acacia. Now, one economic trend that I'd like to just put out here, because it does influence a lot of our thinking, it influences a lot of our investment, is the relationship between the port cost of a router and switch and the, the optic cost. At 10 Gb, the optic represented about 10% of the cost of the total BOM. The port was about 90%, the optic was about 10%.
At 100 Gb, that approached about 50/50 ratio. The reason is that silicon has gotten much, much more substantial in terms of the capacity it offers on a port basis and on a cost basis. What I often tell people is I came to Cisco about 15 years ago. At that time, we had a 40 Gb router line card. It had 14 ASICs on it, and the total capacity for that line card was 40 Gb. Today, we have 50 Tb in a single ASIC on a single pizza box platform, and the scale from 40 Gb to 50 Tb resulted in massive, massive savings in cost per bit on that port. The optics costs have come down as well, but they just haven't come down as fast because the optic is not just silicon.
The optics has generally a photonic element that can be offered in silicon photonics or in discrete optics, has an electrical element, a DSP, and a laser, and it all has to be interconnected. It's a much more complicated assembly, and the result is that the optic is now becoming a much more significant portion of the cost. At 100 Gb, it represents about 50% of the cost, and at 400 Gb and 800 Gb, it's well over half the cost, approaching 70% of the total cost.
And this influences our thinking around optics investment, because if we're going to remain relevant for customers who want to buy a whole solution from Cisco, this would include enterprise customers, public sector customers, as well as service provider customers, we need to have a full portfolio that includes the optics and the ports. So over the past 10 years or so, we've done a number of acquisitions, and I'm not going to drain this chart, but You can kind of look at the upper half of this chart as acquisitions in the area of silicon and optics. And there's really three investment pillars for Cisco. Foundational technologies are silicon, optics, and software. And most of our investments in the area of Cisco Networking or the mass-scale infrastructure group are in one of those three areas.
And so, as an example, Leaba is the acquisition that we did that has brought us Silicon One. Most of the others on the upper half are silicon photonics technologies or CoreOptics was an early-stage coherent technology, and Acacia was the most recent coherent acquisition. It also brought us silicon photonics technologies. Accedian was an acquisition that helps us with network assurance and performance monitoring. And on the lower half of the chart are software acquisitions that help us support customers in monitoring their network and providing insights into their network. ThousandEyes is an example of a technology that allows customers to get very deep visibility into what's happening across the internet, not just in their network, but really across the internet. SamKnows takes that insight all the way to the endpoint at the customer prem.
Sedona is one interesting one. I think, you know, we've been talking about IP over WDM for 15, 20 years. And I think the original incarnation of IP over WDM really didn't succeed in the market, in part because it was really a hardware-centric view. We took a portion of the optical system and put it into the router. It had some drawbacks in terms of density, but it also had significant drawbacks in terms of how we thought about operating the network and when we had siloed operations, in many cases, in our customers with an optical team and an IP team. Sedona helps bring IP and optical together for our customers and converges their operating processes in the network.
So I think we've thought pretty deeply about how to bring together silicon, optics, and software, and I don't think there's any other vendor that's investing at the level of Cisco in these three areas that really help drive down costs and simplify networks for our customers. We completed the Acacia acquisition three years ago this month. Acacia has been a terrific acquisition for Cisco. They've continued to innovate and have very significant share of the coherent ports worldwide, whether they are the high-performance ports that are used in transponders, or the ports that leverage the pluggable technology that is now in a QSFP-DD form factor and is, I think, really becoming the dominant technology for DCI and metro applications and increasingly long-haul applications.
We're continuing to focus on both the very high-performance solutions that are transponder-based, as well as the pluggable solutions that are certainly supporting DCI and metro today, but it really supports all, all market segments, everything from DCI all the way through subsea and even satellite applications. Some of you may be aware that when we acquired Acacia, roughly 20% of their business was Cisco, 80% was other network equipment manufacturers. We've worked very hard to maintain their business through this acquisition and continue to do that, and we're committed to that. I think we... the Acacia team has done a very good job in maintaining business with what are, in some cases, Cisco competitors, and we are committed to do that.
And the reason we're committed to do that is because it's a silicon-based business, which basically means volume helps costs, and we need to drive volumes in order to really drive costs down. I'm not going to drain this chart, but just to show Acacia's got a very rich history of deploying these technologies, both for high-performance applications as well as pluggable applications. The most recent solution that we're offering in the market is based on the Jannu DSP, which was first introduced in 2022. And Acacia introduced for the first time a pluggable solution for transponders that is called the CIM 8. And just before that was introduced, the 400 Gb DCO in a pluggable form factor, and that is now shipping in massive volumes to hyperscalers as well as service providers around the world.
We are just beginning the ramp of CIM 8 at this point, and you've probably seen some press releases for subsea demonstrations, as well as some long-haul demonstrations with everything from 400 Gb all the way to 1.2 Tb using the CIM 8 module. So we are, I think, very, very focused on pluggable technology, that where you take a DCO pluggable and put it into a router and replace a transponder. But we have many customers that are continuing to deploy transponders, leveraging the CIM 8 pluggable that you can see on the right-hand side there. And that is employing the most recent and innovative technologies in terms of things like 3D shaping, using an eighth generation DSP, that gives us 1.2 Tb of capacity on a single wavelength, which is just staggering, staggering capacity on a single wavelength.
That might be appropriate for a DCI application. At 800 Gb, we view this technology as being able to support about 90% of the terrestrial applications globally. So very, very high performance for those applications that demand it, like an ultra-long haul application or a subsea application. Now, I want to shift and talk a little bit about what we have called routed optical networking. At about the time that we acquired Acacia, we had a very firm belief that the pluggable coherent technology would play a very significant role in disrupting the optical market, but helping our customers to simplify their networks and also reduce the cost of their networks. And I think we've been delighted with the progress that we've seen in that.
I think we're still early in that journey, but we're delighted with the progress we've seen in hyperscalers as well as service provider customers that are now deploying that technology. I think interestingly enough, when we first announced that, many of our competitors argued that it would be a very tiny niche of the market that would be served with pluggable, coherent technology. And now we see most of the competitors announcing technologies like coherent routing or derivatives of Routed Optical Networking that sound an awful lot like Routed Optical Networking, but might have the word coherent or routing in there. I think the train has left the station, and the industry is now rallying around the idea that pluggables are going to become a much more significant part of the way customers deploy optical technology.
Let me just show you sort of the cartoon view of what why this technology is important for customers. If you look at the top half of this graphic, you can see on the left-hand side is a router, and on the very right-hand side is a fiber, a DWDM fiber. How do we get the signal from the router to the fiber? Well, there's a short-reach optic in the router that is then... That signal is then carried over a fiber jumper to a short-reach optic that's sitting in a transponder. The transponder has a DWDM optic, so the job of the transponder is basically to take the short-reach optic and convert it to a wavelength or color of light. That's the entire function of the transponder.
That transponder sits in a chassis, and from that DWDM optic, the signal, now a wavelength, is launched through a ROADM into the fiber. That's the path that every single service provider or hyperscaler uses today to get a signal from a router onto a fiber. Again, this fiber is typically going across a city or across a country. This is not a fiber inside the data center. And when we contemplate replacing that transponder with a DWDM pluggable, you can see all the things that go away. Two short-reach optics go away, the transponder goes away, and in fact, that chassis that's carrying the transponder also goes away. The other thing that's very significant about this technology is that for the first time ever, it's complying with open standard DWDM interfaces.
So that offers the customers the promise of being able to have interop occur at the DWDM layer, which has never happened before, and offers them the opportunity to have a multi-vendor environment, where in the past, with transponder-based solutions that are proprietary, they're basically locked in with one vendor for the life of the product. So there's huge customer benefits here from a savings of equipment that results in CapEx savings and power savings, and from the fact that they are no longer locked in with a single vendor once they make a choice to deploy this technology. I want to just get a little bit more specific about those savings. This is an actual RFP response that we responded to recently.
We had a North American service provider that said, "I want to size a network that includes a router with gray optics or short-reach optics, going to a transponder that goes into an open line system. So give me a price based on the entire network for that type of architecture." And we also gave them a response that said, "Well, if we replace that transponder with a pluggable optic, we think that could result in significant savings. And in this case, and this was a, this was a North American provider, a very large metro network, nationwide metro network. In this case, we saved 66% reduction in CapEx for transponder versus a pluggable, and that resulted in a 92% reduction in power for a transponder versus a pluggable.
So for customers that are worried about how they're going to meet sustainability objectives or green objectives, this is a very, very clear and compelling case. And for customers that are always worried about how they can drive down CapEx, it's a very clear and compelling case. And just from a sort of a box perspective, the left-hand graphic here kind of shows the present mode of operation. You've got big routers, you got big line, optical line systems, and big transponder chassis. Routers are now collapsing from what were many racks of equipment down to a pizza box that can support 25 Tb or 50 Tb worth of capacity. Transponder chassis are now being collapsed into pluggables, and line systems, open line systems are moving from pretty large scale chassis, half a rack, down to pizza boxes.
So there's been significant reduction in size, power, and cost as we've seen these, these transitions driven by silicon, like Silicon One, as well as innovations that Acacia has brought with silicon photonics and coherent technologies. And in aggregate, you get significant reduction in power, space, and size. And that does not include things like distribution costs and packaging costs, and things like that. So we've got about, as I mentioned, about 100 customers that are now deploying this. We've seen significant growth year-over-year. Interestingly enough, over 70% of these deployments are over third-party open line systems. That means it's a Cisco pluggable riding over a Ciena or Nokia or Infinera or ADVA open line system.
And that's an important part of the message here, is that we can now disaggregate the line system from the transponder layer, whether it's a transponder or a pluggable. And Acacia has established itself as the market leader in deploying 400 Gb DCO optics. These are just a few of the customers that have announced the deployment of this technology. There have been some that are very vocal. Arelion has done a number of webinars advocating this architecture, as has Colt. Colt has actually advertised a webinar that basically mirrors our claims about CapEx and OpEx savings. That's public and available for any of you that may want to take a look at that. And just in summary, I think Cisco is making huge investments in silicon, optics, and software, both organically and inorganically.
We have our eyes on the market always, and we'll look for ways to fill gaps in the portfolio if they exist, or help us move into adjacent areas. We announced recently Routed PON, so we'll move into adjacent areas where it makes sense. We'll invest organically and inorganically to plug those holes. Some of these things are going to drive architectural transitions. Routed Optical Networking is an architectural transition, and it will take three-to-five years, I think, before we see this fully embraced by service providers, because architectural transitions take time. They're not just an incremental next generation of an optic or a router.
I think as you look around the industry, nobody is investing at the level of Cisco in silicon, optics, and software, and I think we're seeing the results of that come through with things like Silicon One architectures, as well as what we're doing in both short reach and long reach optics. Thank you very much. I'm happy to take questions.
So great, Bill. So good overview of sort of where you've been and where we are today. I want to sort of take us to the moment. So we're kicking off the OFC trade show. Attendance looks very, very high. What's been the buzz? What do you sort of perceive as the hot topics this year?
Well, I probably can't get through the first 30 seconds without saying AI. So AI is clearly the buzz of the industry. I think a couple of years ago it was 5G, but now it is all about AI, and I think that's real. There, we are seeing the effects of AI, primarily in the hyperscaler market today. We're seeing the effects clearly inside the data center, but also in inter data center traffic, driving things like additional ZR and ZR+ demand. I think that the AI curve is, you know, very early stage here. We have a lot to learn about AI and how it will, how it will impact traffic patterns, but there's huge opportunities there. NVIDIA is, you know, obviously on fire with AI. I think there's a lot of room for others like Cisco to play in this market.
That's one topic. Let me just rattle off a couple, and then you can-
Sure
... you can dive deeper. I think, linear optics is a big topic of conversation. I think there's a wide-ranging debate about linear.
Maybe help folks understand what that term means, because that may be a new one.
Yeah, thanks.
Yeah.
Thanks for asking that. So today, you have a host router or switch platform that has silicon on it, and that silicon has a technology called SerDes, which is basically the technology that drives the signals from the silicon to the faceplate. And that SerDes technology is very, very sophisticated technology that has to maintain the signal integrity, even though we're traveling maybe, you know, 12 inches across a board. As these signal rates increased from 10 Gb to 25 Gb to 50 Gb, to now 100 Gb, and now 200 Gb, the problem doesn't just double, the problem actually gets much, much more complicated every time you double the signal rate, and the SerDes has the job of getting the signal out of the silicon and then transmitting it in a high-integrity way to the optic.
On the optic itself, you have a DSP ASIC, digital signal processor, that basically receives the signal from the host and then cleans it up and prepares it to be delivered across a 2-km fiber interface. That DSP has to be communicating with a transmitter and a receiver. That's part of the optic front end. And all of that consumes power. The SerDes power consumption grows significantly as we double the bit rate each generation. The DSP consumes power, the optic itself consumes power, so all of these things are contributing to power growth. And even though we can say the power per bit is going down, the absolute power is going up.
So every time we go to a new generation of a pluggable optic or a new generation of a, of a silicon, the total power for that solution is going up, even though the power per bit's gone down. And so the challenge for many of our customers is power. Cost is always a challenge. We never take that off the table. Space may or may not be a challenge, depending on the customer, but power is a challenge for everybody, especially for the hyperscalers, who in many cases can't even get sufficient power into the data center. And if they can get it into the data center, if they can deliver power in, they have to figure out a way to manage that power, and that may mean cooling, and cooling can itself consume significant power.
So power by itself is a huge and growing problem, and so the idea behind linear is to find ways to make the SerDes better and, and stretch the SerDes performance and to be able to eliminate the DSP that's in the optic. And there's varying views about how that's done, that we can talk about. There's a spectrum of beliefs about whether it can be done and where it makes sense to do it, but on paper, pulling the DSP out saves 25%-30% of the power. So it seems like a good idea on paper. Technically, it's a very challenging problem that then creates a lot of debate about the best way to do that, and we can talk more about that.
But when people talk about linear, they're talking about a linear interface between the SerDes and the optic that basically removes the DSP function.
Right. Any other big themes we wanna touch on?
So the other big one I would say is, you know, I think two, three years ago, Cisco was out there talking about the role of pluggables in customer networks, and I think virtually all of our competitors are now talking about the role of pluggables in customer networks. So I think the, the fact that pluggables are catching fire with customers is a very significant dynamic and shift in networks, a good thing for our customers, as very, very rarely we can go to our customers and say, "I'm gonna, I'm gonna save you space, I'm gonna save you power, and it's gonna cost less." And we're able to do that with this technology. So it's a very compelling argument, and I think the industry is now embracing that.
So let's go back to the first hot topic, AI.
Mm-hmm.
It's mandatory, I guess. Cisco has given this forecast of $1 billion of AI business in fiscal 2025, which fiscal year ends in July.
Mm-hmm.
... so that's not that far. Can you talk somewhat about what the composition is, and what the mix of what's contributing to that $1 billion?
Yeah, I don't have a specific mix that I can share, but I can say that it includes—I mean, one of the things that Cisco's very excited about is that we have a lot of the elements of the AI infrastructure, and that includes things like switches, it includes CPUs, it includes interconnect technology, optics technology, and it includes technology that gets things out of the data center, to the extent customers have to actually network their data centers in order to build AI clusters. So that's that includes our Ethernet switching portfolio, and includes Silicon One inside that portfolio, or directly to customers that may wanna incorporate Silicon One into their own switches. It includes the software that goes with our routers and switches, to the extent customers are buying those directly.
It includes our short-reach optics, as well as the ZR class optics, ZR+ optics. So it's a pretty wide... And compute, the UCS compute platform. So it, it's a pretty wide range of technologies that Cisco can bring to bear for AI. We don't have the GPU, but I think nearly everything else that's part of the AI solution set is something Cisco can, can act on.
I guess one of the debates in the investment community has been, one of your competitors has forecast $750 million of AI cluster-related revenue-
Mm-hmm
... in calendar 2025. How can we map that, that $1- billion Cisco forecast to sort of an apples- an- apples comparison?
Well, I would say, most of our competitors maybe have a fraction of the portfolio that we have. So they might have the switching portfolio, but they don't have the optics, they don't have the ZR+ optics. They may not, may or may not have compute. So I think it does come down to the fact that Cisco just has a broader portfolio to leverage into AI, and we see this as a very significant opportunity, not only for the hyperscalers, but also for the enterprise customers, large enterprise customers that will be not necessarily using training models, but will be using inference models, and we see that as a very significant opportunity for Cisco, leveraging our relationship with enterprises.
The other topic that's come up on AI builds is this timing or expectation for a transition from using the InfiniBand protocol-
Mm-hmm.
Inside the cluster to using Ethernet. Last week, we heard from some of the semiconductor suppliers in that market. I walked away last week being more confused as to what the reality is, because the argument for InfiniBand sounds compelling, the argument for Ethernet sounds compelling.
Mm-hmm.
I can guess which camp you might be in.
Right.
But I'd love to sort of get your perspective on how that transition can occur.
Yeah, I mean, I think you have to recognize that, you know, every vendor's got a certain set of arrows in their quiver, and they're gonna leverage those arrows, so including Cisco. But, but I would say, you know, we're hugely behind Ethernet. We're behind the Ultra Ethernet forum. We have significant participation in there. We're driving that, and we're driving that out of a belief that open is good and standards are good for our customers. And everywhere we've seen open interfaces, whether it's Ethernet or IP or mobile networks, everywhere that we've seen open interfaces, it's been good for the industry and good for our customers because they have choice. And frankly, good for the vendor base as well, because supply chains consolidate and costs come down, and that doesn't happen when you have a proprietary solution.
That's typically one vendor or a very small set of vendors that have access to it, and it doesn't come down anywhere near as fast. And I look at DWDM as a great example. You know, DWDM is sort of the last bastion of networking that, that is not open, has not been open until the ZR standards came along. And it's a very fragmented market. Everybody's got their own solution, and customers get locked in, and they don't like it. And so we are very much behind Ethernet for, for AI. And to answer your question, I think the Ultra Ethernet forum is working very aggressively to come and address many of the issues that will be required for super high-end applications, where InfiniBand plays today.
But for many applications, like, you know, 24,000 GPUs, I think was mentioned last week by one of our customers that can be served with Ethernet today. So yes, Ethernet can be deployed today in AI applications, and over time, I expect it will be more ubiquitously deployed in AI applications, even the very, very high-performance applications and super high-scale applications where InfiniBand may have an advantage today.
And Cisco has its implementation of Silicon One, which I think of-
Mm-hmm
... as a proprietary overlay to Ethernet. So you can leverage Ethernet scale, but it's still proprietary. How is that gaining traction? What's been the response from customers?
I think Silicon One's had a terrific, we've had a terrific response from customers, in part because by itself it stands as an outstanding architecture that customers, just on a basis of a technical evaluation, think is very, very strong relative to what's available in the market. But also because it provides customers with choice that they may not have otherwise. There's a pretty narrow set of players in the market, and there's some monopoly behavior at times, and I think it gives customers diversity in the supply chain that they are seeking.
So Silicon One has had great success in the market in every form, whether it's selling the ASIC to customers that wanna go build their own switch, or when we sell our standalone switches and routers fully integrated with software, IOS XR, or when we sell our switches and routers to run SONiC. So all of those, all of those applications, I think have had—we've had great success. Our 8000 Series of routers has experienced the fastest growth of any router in Cisco's history, and I think that's entirely driven by Silicon One.
So I wanna go to that second topic you mentioned, which is linear pluggable optics, and this is clearly a hot topic. We were getting questions even before the show began. It does feel to me that it's highly controversial.
Mm-hmm.
I think maybe one of the first things I'd like to see if we can understand is, what are the sort of the deficiencies, and sort of where is it applicable, where would it not be applicable?
Mm-hmm.
How much of the market could be transitioned to that architecture?
Okay. Let me, let me just step back a little bit and, and, tell you how I think linear originated as well. So two or three years ago at OFC, you would have—if you were paying attention to Cisco and some others, you, you might have seen that we were demoing co-packaged optics. And we have a belief that co-packaged optics, it will have a role to play in networks. And basically, the idea of co-packaged optics was, was exactly what I mentioned earlier when we talked about linear, which is: how do we drive power out of the solution?
As the SerDes rate continues to increase 100 Gb-200 Gb and maybe 400 Gb, we're seeing the point where we just can't get the signal across the line, the router line card with any integrity, or the power required to do that becomes just prohibitive. Power in the optic, power in the silicon. So co-packaged optics basically says, take the optic, and instead of having a pluggable on the faceplate, put the optic right next to the silicon and just package them as one monolithic unit. All of the optics that would appear on the faceplate just basically come out as fiber jumpers, but all the heavy lifting in terms of creating that optic interface is done right next to the silicon. Cisco has the capability to do that.
There are a couple of others in the industry that have the capability to do that, and if you just think, who has silicon and optics, it's basically that set of players, but it is a complicated problem. It's a very complicated packaging problem. We probably have to develop a new ecosystem of manufacturers that have that packaging expertise. There's test issues, there's maybe standards issues in terms of what is the optic that comes out? Is there a standard between the silicon and optic that we have to think about? A lot of issues. And I think as people looked at that, they realized that's a pretty complicated problem that's gonna take some time to sort out, though it does deliver probably the best performance in terms of power savings.
And so I think a number of folks in the industry started to think, "Is there a way we can just use the existing architecture of the pluggables with the existing hosts in that, in that, in that form, and somehow just drive to better performance in either the silicon or the optic that allows us to drive power savings?" And that's kinda how linear was born, is to say, "Hey, if we really pushed on the SerDes performance and we improved the optic transmitter and receiver performance a bit, we can probably take that DSP out and basically have the SerDes in one router or switch, talk to the SerDes in its peer router or switch at the far end, and do that channel, the necessary channel equalization and cleanup in order to deliver a clean signal across 2 km, for instance." It's a very challenging technical problem, a very challenging technical problem, and I don't think we should, we should underestimate how challenging that is.
And just to give you one anecdote, we had in a first generation of our 250-Gb transponder several years ago, an ACO optic that was basically a sort of a mostly coherent optic that had the digital piece on the host, and we used a linear interface between the host and the optic. And we did the linear interface for all the reasons that we're talking linear today, and it took us two years to get one optic vendor to make that optic reliable. Reliable, meaning you could plug it in, and it would work without a whole lot of manual intervention, and that it could be manufactured in volume. It took two years to get one optic vendor to basically work with us, and since then, we have no other optic vendors that have made it work.
And that was obviously, you know, four or five years ago. So SerDes have improved, the technology has improved, so we have promise of doing better, but we have a lot of scars from that experience, so that we shouldn't underestimate the challenge in building linear interfaces. Everybody can build one, you know, everybody can build a few, and so you're gonna see demos all over OFC of everybody in the world, including Cisco, demonstrating linear interfaces. And I can promise you that everybody worked really hard to get that demo working, and the issue is really: Can we achieve scale? Can we achieve scale, and can we achieve interoperability? So I'm pretty confident we can get one host to work with one optic, and that may be a deployment model that works for some customers.
But the ideal is really to have any host work with any optic and have interoperability end to end as well. That's the ideal. That's a very, very challenging problem, and we should not underestimate how challenging that will be to get to that future state where we have anybody that's advertising a linear optic can work in any host that has a linear interface, and you could have the far end with a different optic and a different host. That's really the challenge. We will probably have to back off from that desired end state and maybe have, you know, matched optics or bookended optics or bookended switches or routers in the interim, and that may be an acceptable interim state, but it's probably not the best end state for the industry.
Yeah. Now, my follow-up, I have to—we have to be careful because I don't wanna get too far into the weeds, but I know I'm going to be asked about this half solution.
Okay.
So there's this talk this week about, well, what if I do half of it-
Mm-hmm
... re-timed?
Mm-hmm. Right.
How does that work? How is that different? Is that a-
So you can just think-
-good solution?
You can just think there's basically a transmit path, and there's a receive path. You know, one, one's out, and one's in, and each one of those has its own unique set of challenges. We have to transmit the signal from point A to B. We have to be able to receive the signal that's coming from B into A. And the idea of the half linear is that it turns out that there may be some things that in the transmit path, you could. If you kept the DSP effectively in that path, you could simplify the problem. So, but there's trade-offs always. So you make the technical problem simpler for sure, but you don't get the full power benefit, of course, because you're still using the, effectively, the DSP in that direction.
So there are advocates for that in the name of simplicity or in the name of expanding the number of players that could participate. It does not deliver the full benefit of a full linear solution. And to be clear, a linear solution doesn't deliver the same benefit that a co-packaged solution would deliver. So everything is sort of inching its way towards better power performance, and the question is: Where is the industry gonna settle? And that's an unanswered question at this point. As you may know, there was an announcement last week about an MSA being formed that has 12 or 15 very interested parties right now, and Mark Nowell from Cisco is gonna be chairing that and hoping to bring some of these viewpoints together and coalesce around something the industry can rally around.
... So my next question is going to flip this completely around. So we just spent a few minutes talking about the need to get rid of DSPs to save power-
Mm-hmm.
But the challenges. Now, we get this other topic of, well, how about bringing coherent optics inside the data center?
Yeah.
So c oherent has traditionally been in the long haul, and that's sort of the Acacia technology you talked about at the beginning. And what will it take? 'Cause I've had the impression that, well, coherent, super high performance, relatively low volume, very high price. So how can that play inside the data center, and how does that sort of live in a world where you just explained why we want to get rid of DSPs?
Yeah. So there's this is a very large trade space between power, cost, performance, and we're not at the point yet where all those trades have been analyzed, I think. So, bringing coherent into the data center is going to happen. That will happen at some point, and the question is when? And if you asked me a couple of years ago, I might have said, "Well, 51.2- Tb generation of switches is when we might have to start thinking about doing that." But the folks that are delivering IMDD or the sort of the classic direct detect technologies, where we're talking about linear, have managed to improve things on SerDes and have managed to improve things on the optics so that we, we don't have to bring in coherent in order to get to 800 Gb, for instance.
And we probably don't have to bring coherent in order to get to 1.6 Tb.
And so I kind of view it like the analogy of Moore's Law being, you know, people have been predicting the end of Moore's Law for 20 years, and every year we kind of put a line in the sand and say, "This is when you—this is when we're out of gas with the current approach, and we're gonna have to do something more sophisticated in order to drive to the next generation." Yet the folks that are building the technologies for SerDes and optics in the short reach world, the client world, are continuing to innovate and find ways to push that out in time, because it's more costly, it's higher power, and so that has to be traded against, do you want that performance leap, right?
Maybe does the performance leap outweigh the penalty on power or the penalty on cost? We're not there yet, but we will get to a point, and I can't, I can't say exactly when that's gonna occur, but we will get to a point, you know, it might be 3.2 Tb, it might be 6.4 Tb, but we're gonna get to a point where we're just out of gas with the existing set of technologies. We're just at the limits of physics, and we have to, we have to get more sophisticated in terms of how the optical signal is processed in order to send even, you know, 500 m or 2 km. We're not there yet, but we will get there.
Would you be okay maybe sharing at least, not, not necessarily Cisco price points, but just to help people in the audience understand the relative price difference between, let's say, an 800-Gb transceiver used inside a data center for sub- 2 km versus a coherent transponder used in a long-haul network?
It's, I'd say, nominally 10-to-one difference.
So, big gap.
At 800 Gb, yeah, I mean, big, big range around that-
Right
... depending on which customer is buying, but, but it's a big gap. It's not 10% or 20%.
The relative volume in the marketplace, is it also 10 to one?
Easily 10 to one, yeah.
To some degree, this does feel to me like a little bit of a chicken and egg, that if we can get the volume up, we can get the price down.
Yes, for sure. I don't think cost is gonna be the issue. I think it's gonna be things like power. I think power will be a much more significant factor than cost itself.
Great. So I wanna maybe pivot to the wide area network-
Mm-hmm
... applications. But let me check with Sami. Do we have any questions online, or you want me to go ahead to share with you?
Simon, g o ahead. You can move some to WAN questions.
Great. So, one of the sort of the fundamental drivers around network demand has always been sort of traffic growth rate. And there's been this, I guess, buzz related to AI, of course-
Mm-hmm
... that AI is stimulating traffic or will stimulate traffic. Mentioned some of it's anecdotal, but what are your operating assumptions when you're doing your forecasting and thinking about the business?
Well, one thing I would say is it's very hard. We do have customers coming to us today saying that they are driving demand to us based on AI demand. It's very hard in the WAN to separate, you know, how many bits are normal traffic versus AI traffic. That's just not known, not visible to us, and I don't even know if it's visible to the customers in many cases. They're just seeing a demand for more capacity required between data centers, for instance.
It's very hard for them to separate that out and say, "This is exclusively AI demand." But I can say that we are—qualitatively, we are seeing, especially in the hyperscaler space, they are coming to us and saying, "This incremental capacity that we need is driven by AI." And that's impacting our transponder business, it's impacting our router business, and it's impacting our pluggables business, the ZR- class pluggables. So I think we're—we're beginning to see that. I think it's still early, early days, but we're definitely beginning to see that as a driver.
There's a. I'm gonna call it a conventional wisdom or a belief that the telco vertical in particular, and the wide area networks will improve in the second half of 2024.
Mm-hmm.
So I think we've generally all conceded that the beginning of this year is pretty tough, but things will get better as we get into the second half. What's your sense of that as an industry perspective?
So I think we have, I don't know if somebody's gone on mute, off mute there, but we've got to get-- if everybody could go on mute, we're getting a little bit of echo. I think we, we've basically said, "Look, we see a couple quarters of inventory consumption ahead of us," and that still remains the case. There's a couple quarters of inventory consumption that we're seeing across our customers, and beyond that, I think there is a, there is a, a bit of an unknown, and it, it's very hard for us to, to separate what customers are telling us is inventory absorption. Customers are sitting on millions of dollars, in some cases, of inventory in the warehouse that they have yet to deploy. And then the question, the obvious question is, what happens when that's all deployed?
Is there something more that's going to appear like they're stepping on the brakes? And that's fundamentally an unknown at this point. We do see, you know, some green shoots out there, certainly, that customers are beginning to especially in the hyperscalers, but even in the service providers, beginning to place orders. But I think it's still early to declare, you know, we're around the corner.
And one of the things we've observed is your sort of more traditional systems business, so not the pluggable part of the business, but the line systems, open line systems, that's generally been losing market share overall. And I guess I'm trying to understand from you, how much of that is intentional because you're trying to migrate the world-
Mm-hmm
... to your pluggable strategy? How much of it is competitive? What's the dynamic as to sort of that traditional Cisco line system business?
Yeah, well, let me just say, I think some of that is intentional. We are still out there seeking business on line systems and transponder-based solutions for customers, but we've seen a very significant shift, especially in hyperscalers, away from transponders and to pluggables. In some cases, it's not clear we would have won that transponder business, by the way, because we're providing pluggables over third-party line systems that would have naturally gone to those third parties with transponders. So it's more that we're taking business away from competitors. But when we did the Acacia acquisition, we did the Acacia acquisition with the intent of recognizing that we were probably gonna be cannibalizing a portion of the optical business if we were successful in driving customers towards pluggables.
We were gonna effectively be taking transponders out of the business and replacing them with pluggables, and that's a very intentional move. We think that's a good, a good answer for our customers. We think it's economically a good answer for our customers. It simplifies their network, and so it's just one of these, one of these cases where you have to kinda eat your young, and, and we're willing to do that. Across the portfolio, I think we have very mature view about how we're gonna manage the portfolio, both optical and optics, and, I think we're gonna see this over time. That's, that's not to say that we're not competing for transponder business. We are still aggressively competing and winning transponder business with things like CIM 8 and line system business.
But, you know, wild success in Routed Optical Networking would mean diminished performance in the optical business.
So that sort of nicely takes us to talk about the ZR and ZR+ business. So that market, to me, started out nicely, but primarily was really two hyperscalers, and then it paused. It seems to be coming out of that pause and maybe evolving. Could you really help us understand the dynamic of who's buying and why?
Yeah. So I think there was a pause. I think we've seen green shoots more recently there, and we're now shipping to, I count five hyperscalers and over 100, well over 100 customers that are in the service provider segment, some public sector customers as well. We just won a large public sector deal in India, nationwide, network, all pluggables. And, as we've announced, customers like KDDI in Japan deploying for their metro applications. So, I think that we are early stage in this, in this still, and we're early stage, though the economics are just inarguable, it is a super compelling economic argument, customers still have to get comfortable with the operations changes that may be required. It is a disruption. It's a network architectural shift. It's not just an incremental change.
So customers have to think about things like, if I had an optical operations team and siloed operations, and I had an IP operations team and siloed operations, what do I do to bring those together when I've now taken, effectively, the transponder that was in the optical world and put it in the router? We're helping them. We're bringing solutions to the table that help them solve that problem, but, but those are- tend to be slower- moving dynamics when you're going to a major tier one and saying, "Look, we'd like to introduce this new operations system that is gonna replace an EMS, or it's gonna bridge two EMS domains." That's a slow-moving dynamic, so that's why I say three-five years for this transition to really take place.
But what you're seeing is a lot of the, sort of the more agile players that can make that move fast are doing that right now.
Now, it's my nature to worry, and so when I look at this market, to date, it's largely been a duopoly.
Mm-hmm
... you and one other vendor. Now it seems as if we could probably pretty quickly identify 10 vendors-
Mm-hmm
... with ZR pluggable devices, at least. Is this going to be one of these markets where it turns into, you know, profitless prosperity, price competition, how should we think about how this matures?
Well, I certainly think there's gonna be price competition. That's always the case. But today, our price competition is against a transponder. You know, when we go in with a pluggable optic, we're comparing it to a transponder that the customer's deploying. I think as the market matures and more customers adopt this technology, we'll start to see more price competition in the pluggable. But today, the competition is really against a transponder, which is a pretty easy mark in terms of hitting an attractive price point. As you suggest, there will be more players coming in. I think big markets attract new entrants, and we're gonna see that. But I would also say there's relatively few players that have the underlying technology.
A lot of them are gonna be reliant on guys like Acacia and some of the other DSP providers, and then go build modules, and, so we'll see that dynamic as well. The other thing I think we have to pay attention to is the customer behavior. You know, customers, I think, will gravitate long term towards the suppliers of the technology, and, and will wanna know that they're getting it from the source. And so I think long, long term, I expect that dynamic to play as well.
Well, great. So Bill, we're just about out of time. I like to sort of wrap up, to sort of put a button on it, is what do you, what do you think is the least appreciated aspect of Cisco's optical business segment?
So I'd say a couple things. One is, we don't talk publicly on earnings calls about our optics capabilities and business. Optics is a very substantial business for Cisco, multi-billion dollar business, and we sell to all customer segments, everything from our enterprise customers to commercial sector to public sector, service provider, and web. It's a very substantial business for Cisco, and we are now. In the past, we've basically viewed optics as sort of an accessory that goes with routers and switches. We are now going out pretty proactively with customers and saying, "Look, if you are thinking about optics as a buying center, and you wanna consider using optics, third-party optics in your routers and switches, we want a seat at the table for that conversation.
We want, we wanna be considered your optics supplier." I think that's not fully appreciated in the market. And the other thing I would say is, I think Acacia has done a terrific job. We're obviously not gonna go talk very publicly about Acacia selling into markets where Cisco might be competing, but Acacia's done a terrific job since the acquisition of innovating and really sticking to their original roadmap for high performance as well as pluggable technologies, and I'm super proud of that team for what they've accomplished in coming to Cisco.
Well, great. Well, Bill, thank you very much. Folks, thanks for joining us today. Sami, thanks for having us.
Thanks, Simon.
Appreciate it. We'll catch everybody on the show floor.
Bye, everybody.
Thanks.