Cisco Systems, Inc. (CSCO) Earnings Call Transcript & Summary

Hello, everyone. I'm Katie Lindner, facilitator for today's webinar. In a moment, I'm going to turn the session over to our fabulous host and speakers. But before doing so, I have a few housekeeping notes to cover. You were automatically muted when you came into this session. [Operator Instructions] This session is being recorded, and you will receive a link to the recording later this week. Also, we'd like to encourage you to take a moment to fill out our survey. It's going to pop up when you close your Webex window. With that, let's go ahead and get started. Fahad, you have the floor.

Thank you. Good afternoon, and welcome. My name is Fahad Najam. I am the network equipment analyst here at Loop Capital. And today, it's my pleasure to host Cisco's Bill Gartner, Senior Vice President and General Manager of their optical business unit. Before I ask Bill to comment, I'm going to turn it over to Marty Palka from Cisco IR to go over some very important disclosures.

Thanks, Fahad. Today's session, the Tech Talk webcast that we're on right now is educational in nature. So no new financial information will be discussed. However, we will be making forward-looking statements. These forward-looking statements may differ materially from actual events and subject to the risks and uncertainties found in our filings with the SEC in our 10-Q and 10-K. I will turn it back to you, Fahad.

Thank you, Marty. Bill, welcome. Good to see you.

Hi, Fahad. How are you doing? Thanks for having me.

For the benefit of many who are on this session today, if you could just briefly explain your background and what your role is at Cisco, and then we'll quickly get started?

Sure. So I've spent most of my career in the optical field in 1 way or another, optical systems and optics. And at Cisco, I have responsibility for our optical systems business as well as our optics business. And the way you can think about those is the optics are the transceivers that we sell typically with our routers and switches that find their home inside a data center or a central office where fiber is plentiful, meaning if you add a new router or a switch, you pull new fiber to every port on that router or switch, and every port gets its own fiber, you put 1 signal on each fiber. And we use transceivers, little pluggable transceivers in the routers and switches in that world. Distances are short, typically less than 10 kilometers. So it's a relatively simple problem to solve because distances are short and the -- and you're putting on signal out of fiber. Once you leave the data center, now you have a much bigger challenge, and you're trying to send a signal across a city or across the country or even between continents, that's a much more complicated problem because the distances are long, fiber is scarce, meaning if a service provider puts a new router in New York and a new router in L.A., we can't ask them to pull new fiber between them. We have to use fiber that's in the ground. And that means you have to put many signals on 1 fiber in existing fiber. So that implies what we call DWDM technology, dense wavelength division multiplexing to put many signals on 1 fiber. And we're transmitting those signals over very long distances up to several thousand kilometers. That's a much more complicated technology. That's where optical systems come into play, leveraging coherent technology. So optical systems or you can think of outside the data center or central office for signals that are traveling around a city or a country, and optics are the little pluggable transceivers that go with routers and switches.

Thanks, Bill, that was pretty comprehensive. So I guess we'll start with the very first thing, which is hard on top of every investor's mind and that's component shortages. Chuck, on the earnings call, signaled that you're seeing some stabilization in the component availability and potentially even improvement in next year. If you look at the Internet of the future revenue declined 10% year-on-year. If I got my numbers correct. So it seems like component shortages are still impacting your business unit as well. So can you give us a quick update on where we are and what you're thinking?

Just to kind of reinforce what Chuck said. I think we're not out of the woods yet on supply chain challenges. There are still significant shortages that we're managing. Cisco has, I think, taken very aggressive steps with our suppliers to make sure that we're in line as the capacity expands. But there's a huge demand that is creating that constraint that than derives from things like the automobile sector that is using much more semiconductor products than they had historically, but also consumer and things like 5G and networking that are placing significant demands on the semiconductor fabs. And those challenges, I think we've seen very significant stress over the last year. We do see a bit of daylight and that things seem to be improving a bit, but we're certainly not out of that -- out of the woods yet. And our supply chain team is spending an enormous amount of time working with suppliers and qualifying second sources where we can make sure that we can come out of this as quickly as possible.

All right. Bill, you and I have discussed this in the past. But I also get this question from investors and that is, we all know and are aware of the significant lead times, long lead times for certain components, especially in the optical domain. ROADMs have been in short supply. And a number of investors have expressed concern that if ROADM lead times and supply is not improving meaningfully anytime soon, are you beginning to see a shift away from ROADM-based architectures towards more router-based switching? And I think there was a little bit of confusion in the investment community. They think that the routed optical architecture, your -- you've proposed replaces some of the functionality of ROADMs. But why don't you elaborate a little bit more on what you're seeing and what your thoughts are?

Okay. You hit on 3 or 4 points that I'm going to try to address there, Fahad. All good questions. And I understand there's some confusion in the industry over this. So let me start by just talking a little bit about router optical networking and then answer your question about whether I think that can replace ROADMs and whether customers are shifting their architectures based on lead times. So router optical networking is a new network architecture that does a couple of things. It leverages coherent pluggables. So rather than have an optical system with a chassis and line cards a plug-in that chassis, you take that function and move it into a pluggable that goes directly into the router. That's one element of it. And another important element is that you take private line services like TDM services and OTN services that many large enterprises might be buying still not just Ethernet by buying things like T1s and E1s and E3s, and you put those services on to the IP layer, you consolidate them onto the IP layer. And you take advantage of these 400-gig pluggables to basically send signals between routers over coherent signals, these coherent signals. When you've done those things and you step back, one of the things that we looked at historically is ROADMs were really put in the network to bypass the most expensive resource in the network. The most expensive resource in the network was the router. So 20 years ago, we looked at the architecture of a network and said, "Boy, you don't want to go through a router if you don't have to". Because the cost of that router port is very expensive. The cost of the switching capacity is very expensive. We should find a way to bypass routers. And we invented very sophisticated technology, ROADMs, optical switches that allowed us to bypass routers whenever you could. So if you're going to go from -- if you have traffic going from New York to L.A., for instance, you didn't want to hop through every router on the way from New York to L.A. You wanted to go basically an express path from New York to L.A., and that's what ROADMs allowed us to do. But it stepped forward now 20 years and you look at what's happened on the cost of a router port, it's come down dramatically, thanks to advances in silicon. We've seen just dramatic, dramatic improvement in silicon. We're now delivering silicon that can offer 25 terabits in a single 1RU chassis, just massive, massive amounts of silicon, 800 gig on a single port. That used to be an entire lineup of bays of equipment to get 800 gig worth of capacity. Now it's on 1 port. And as that's happened, the cost per gig has come down dramatically. So if we step back and say where the assumptions that caused us to build a very complicated multilayer network, IP layer, DWDM, maybe an OTN layer. Are those assumptions still in place? And the answer is no. The router is no longer than the most expensive thing. And in fact, it's cheaper to go through the router rather than around it in many cases, which can be an argument for eliminating ROADMs. You say, "Look, you don't really need the ROADM to bypass the router, you can go through the router now". And that's a more cost-effective way to carry capacity. And in fact, when we've looked at customer networks, and we've modeled many, many customer networks with real traffic, we see they have a lot of lightly loaded wavelengths that are bypassing routers. And rather than have a lot of lightly loaded wavelengths that are pretty expensive wavelengths, by the way, why not pack that capacity onto the IP layer, doing what it does best, aggregating traffic, and carrying that over 400 gig pluggable optics between routers, that can lead to a conclusion that you don't need ROADMs. And for some customers, that may be the conclusion. But some customers will say, look, I still want to have a ROADM to deliver wavelength services, for instance, or private line services that are approaching the rate of a wavelength. And so we're not saying ROADMs go away. We're saying that in some cases, the need for ROADM may be mitigated. But I would say, to answer your question directly, I don't think customers are going to make an architectural shift, which is something they might do once every 5 to 10 years. They're not going to do an architectural shift for a lead time problem that they perceive to be relatively short term. So they're making architectural shifts because there's an economic argument to do it and the long-term economic or technology trend that they want to leverage, but it's not going to be because we've got a short-term supply chain issue. So does that answer your question, Fahad?

It does. But it also dovetails into a very hot topic, literally hot off the press. And I realize it may be a little bit premature for you to comment, but at Sikom, Google just made some announcements about their Jupiter intra data center network, and they talked about bypassing -- switching with an optical switch layer inside the data center. What are you talk about that? And I suspect a lot is still unknown and a lot of the details are still missing. But it looks like this argument is being not just in the ROADMs but also now spreading inside the data center, too.

Right. So I saw your note about 5 minutes ago. I'm not sure I can comment on that specific issue. But let me say generally, optical cross-connects have been around in -- certainly in the research domain, and in, I would say, the sort of the science project domain for a long time. I have a patent on an optical cross-connect that's 20 years old. And that optical cross-connect never hit the light a day because it just didn't make economic sense, which is kind of the problem that we've had with optical switching in general. And people have experimented with optical switching that is effectively everything from a patch panel replacement kind of a fixed static switch that switches wavelengths to actually trying to do sort of optical packets. I'm not sure exactly what Google is doing here. We're certainly interested in finding out more about that. And I do expect that over time, optical switching will play a stronger role in -- certainly in data centers and in customer networks generally. I think initially, there's a clear application for replacing things like manual patch panels. The economics have not proven into date, but I think they're getting there. They will certainly get there. And then I think we'll see more interesting ideas around true optical switching. But I don't think we're at the stage yet where you see an optical packet switch. I don't think we're going to see that for quite some time.

Yes. And I totally agree with you that what this highlights is the increasing importance of optics and optical in network infrastructure, especially as we go higher in bandwidth.

No question.

So switching topics and away from the component shortage debate to demand. Pausing question for all of us is, if I look at the -- not just Cisco, but your peers, every 1 of you is highlighting significant backlog. And the question comes, what is driving this exceptionally strong demand I'm sitting here in New York, my bandwidth certainly has not improved over the last 2 years. So it continues to -- demand continues to exceed even my expectations, and I suspect investors are still also puzzled. So what is driving demand for your optical systems and solutions? What can you tell us a little bit about the demand cycle?

So one thing I would say is, first of all, I think, as an industry, we should be very proud of what -- how the networks behaved throughout COVID. And I think back 20 years before we had ubiquitous Internet broadband services. And before we had video really in a ubiquitous way like we do now, take it for granted, whether it's Webex or Zoom or whatever service you're using, it would have been a disaster. It would have been a complete global disaster for how companies interact with their employees and with their customers. And frankly, we got through COVID in a pretty easy way. And many of us learned new ways of getting work done. But I'm really proud of the way networks held up. And you think about the stress that we put on networks almost overnight, moving from very enterprise-centric world where a lot of the capacity was going to enterprises to very consumer-centric world where a lot of the demand was going to consumers. And what happened is that many of our service provider customers basically built networks for an average capacity that they expected to see and a peak capacity that they would tolerate, but they didn't want to run their networks that hot. And what happened was the networks were running very, very hot during COVID in many areas. Their networks are running way hotter than by design they should have been running. And so I think in part what's driven some of the demand is that service providers are now trying to get their networks back to a nominal average utilization as opposed to running it hot because frankly, if you're running at that hot, then when things go bad, you don't have capacity to restore. So you don't want to be in that state long term. And so part of it is, I think, just catching up with the way networks -- network traffic has changed with COVID. But then there's other -- there are other things that are happening like the hyperscale continue to deploy capacity that is driving massive capacity demand for all of us. The hyperscalers are moving more of their capacity to the edge, which drives a lot of the metro needs for them and for service providers who have to -- who may be carrying that traffic in their metro networks. And then we have things like 5G, where -- which I think is coming online for many service providers as well. So there are a lot of demand signals out there that say there's an argument for demand. This is not a case where this is like -- they will build it and the demand will come, most of these guys are behind the curve, trying to capture out at this point.

Wonderful. Bill, you've mentioned this in the past that the optical industry is cyclical. And what you described is that each individual customer goes through their upgrade cycle. But 1 thought arises that with COVID-19 hitting the entire globe at the same time, every service provider around the world had this capacity crunch at the same time. And so the fear is among the investors that every service provider is now upgrading at the same time. So we've essentially now got a super cycle, and can you go see me?

Yes.

So we've got a super cycle and followed by a deep nuclear winter. What are your thoughts? What are you seeing? Any insights you can give of that?

I don't see any signs of that, Fahad. I mean, we're -- I think we've been through enough nuclear winters and cautious optimism that I think we're pretty guarded about how things look. Right now, I think there's true demand out there. I think a lot of the capacity that was deployed and is being deployed as augmenting existing networks. So it's not necessarily a network upgrade, but it's generating capacity to existing networks and leveraging an existing architecture. We've had to extend the life of some products that we were not -- we're frankly planning to end of life, but we've had to extend them for customers that want to just continue building out for a little while longer. So I don't think we're going to miss the next big upgrade. I think we are seeing customers that, for instance, deploying our 8,000 series routers now. That's very, very early in its life cycle. And customers are beginning to deploy that, but that's a next-generation architecture, the router optical networking architecture, we've got probably 20 customers for that right now, but that's very early stage as a next-generation architecture. What we saw is a lot of customers augmenting capacity on the existing architectures.

Bill, are you beginning to see any of the benefit from the massive stimulus that are being spent here in the U.S. and in Europe or is that still yet to come?

I think that's yet to come, absolutely ahead of us. I don't think that -- I don't -- we're certainly looking to capture some portion of that. But I think that's all ahead of us.

Okay. And I know you mentioned this briefly on the 5G upgrade cycle. We in the investment community still think about 5G as a wireless technology and primarily focused on the massive TAM for radios and antennas, but that is a concomitant associated wire line opportunity. Why don't you double-click on the opportunity you're seeing with 5G deployments? I suspect you're beginning to see some of the initial rollout fab.

Yes, absolutely. I mean 5G, absolutely is a wireless access technology, and we'll see things like private wide 5G and large enterprise applications as well, but ultimately has to land on a wire somewhere, land on an optic generally. And that puts stress on access networks that have to carry that capacity back to a cord or metro network and it puts stress on metro networks. I do think that we're probably not going to see going forward, we're going to see more stress on metro and access than we will on core networks because as one of the network trends is that I think there's a lot more traffic that's going to be terminating in the edge of the network, not necessarily going through the hierarchy that we used to have of sort of access to metro to core, but rather staying in the metro network. And so we're going to see a lot more stress in those metro networks. And it's driven in part by things like 5G. It's driven in part by things like services that are terminating in the edge.

In terms of the edge and access opportunity, I suspect you're beginning to see some of the RFPs already happen. And probably those are still going to follow the radio deployments. But we all get a little bit confused on the timing. Is it that the service providers have to do the radio upgrade as well as the wireline upgrade at the same time? Or is this they can go...

They're decoupled. In some cases, the service provider has gotten the network in place ahead of the radio network. We've been doing 5G motivated deployments for the last couple of years ahead of radio network deployments in many cases. So I think this is going to be like a 5-plus year transition. Some service providers is going to take a long time to get to 5G, some are getting much more -- are much more aggressive about it. But I don't think they try to synchronize necessarily the radio and the access or metro networks to happen at exactly the same time. They anticipate both have to be done. But -- and it doesn't have to be done in lockstep. One can proceed the other.

Got it. So I'm going to now kind of focus more on the intra data center opportunity, optics part of your portfolio. I was a little bit intrigued by Chuck's comment on the earnings call, where you mentioned that you guys had an 800-gig win with a hyperscaler. And to me, at least, we were still thinking that Microsoft and Facebook are still doing 400 gig upgrade cycle. So can you -- at least the way leases suggest that there will be no meaningful pause between a 400-gig upgrade cycle and an 800-gig upgrade cycle. But can you tell us where we are in the 400-gig upgrade cycle and what's going on with the 800 gig?

Yes. Let me actually back up a little bit from that, Fahad. I think we're not done with the 100 gig upgrade cycle in many cases. Certainly, if we focus on the hyperscalers, they are the tip of the spear and you can look at them as sort of what's going to happen in the rest of the industry. And they are absolutely on the 400-gig upgrade cycle now. But I would say early stage still, not everything is a 400 gig. There's still a lot of 400 gig to be deployed. As we look at service providers, it's a mix. Many of those are at 100 gig and just dabbling in 400 gig. And if we look at the enterprise market, there's very, very little 100 gig that's actually been deployed. So I think these have a very long tail. If we look at 10 gig, 10 gig has been deployed for 20-plus years, and we're still seeing massive deployments in 10 gig. So these -- all these technologies tend to have a very long tail overlapping for certain, but a very long tail. For 400 gig, we are at the early stage of the 400 gig when we look at the total life cycle and total customer base for 400 gig, we're at the very early stages of 400 gig. And 800 gig is just -- it's not even a blip on the radar yet. We don't have 800 gig optics, for instance, to support 800-gig router ports. And the initial 800 gig optics are really going to be 2 400-gig optics that are packaged side-by-side in a transceiver. For quite some time, the 800 gig is really a port density issue driving more density into a port -- it's not driven by a need to have an 800 gig optic per se. So we'll have 2 x 400 gig optics plugging into 800 gig ports. That's going to be the deployment for quite some time. And even that, I don't expect to see that in any significant volume for at least a couple of years.

You mentioned 2 x 400. I guess it dovetails into the next question I had, which is this 100-gig opportunity serdes, 100 gig PAM4. The thought arises that if you're going to upgrade -- especially in the cloud data center, if you're going to upgrade 1 portion of your network to run at 100 gig serdes, you're better off upgrading the entire data center architecture to 100-gig serdes. So the first question is, one, do you think 100 gig serdes as really another big upgrade cycle akin to the 10-gig serdes upgrade cycle that we had previously?

It will come down to economics, Fahad. I don't think that customers that have deployed, for instance, a 50-gig serdes based solution or 25-gig serdes-based solution are not going to jump to 100 just because 100 is there. They're going to jump to 100 because there's an economic argument for doing that or there's an architectural reason for doing that. So as an example, the early stage 100-gig optics were all based on 25-gig lanes. On the optical side, let's talk about -- not necessarily the host side, on the optical side, like a CWDM4 had 4 25-gig lanes. Then you come along to the point that you can have a single 100-gig lane on an optic, okay, using something like PAM4 with effectively like a DR1. Single 100-gig lane, it's still a 2-kilometer optic that carries 100 gig worth of capacity from point A to point B. So why would a customer upgrade from that very effective 4 lane solutions they've got deployed to a single-lane solution. The answer is, in most cases, they're not going to do that. But here's the architectural got you. That single lane of 100 gig is the thing that connects to a 400-gig optic that has 400 gig lanes. So if we want to connect something to that 400-gig optic that has 400 gig lanes, you need 100 gig lane on the other end. And that's an architectural reason why you might do it. There's also an economic reason that says, look, if this was significantly cheaper for me to do the 100 gig rather than 4 x 25, then I might consider transitioning to that, but I probably wouldn't replace that until I need to make the next step, like to the next bit rate, 400 gig or 800 gig or whatever it might be. So I think that's going to be an architectural question and an economic question. We generally won't see a customer swing from 1 technology to another that gives them equivalent capability unless there's a really strong economic argument. And even in that case, it would be a cap and grow type scenario rather than a replace scenario.

So in a way, what you're saying is that the 400-gig has a very long runway. 400-gig is...

400-gig lane has a very long run rate yes. And it will evolve from 50-gig lanes to 100 gig lanes, and that will provide some economic benefits over time and maybe some technology or architectural benefits. But customers that are going to deploy early are going to keep what they've got, and then they'll look for the next logical upgrade cycle to consider an upgrade, but they're not just going to jump because the technology has changed.

All right. So the next big hot topic inside the data center is bit 1.6 terabit pluggables or co-packaged optics. I think Broadcom just announced their Tomahawk4 with a co-package version of it. And I think Tencent is the lead customer for that solution. But it sounds like it's still a big debate in the -- especially among the cloud scale customers about co-packaged optic versus pluggables. First of all, what is your take -- what do you think your customers are telling you?

So Again, I think when you look at something like co-packaged optics, it's a question of economics and it's a question of architecture. And really the primary issue that drives co-package to optics is heat dissipation, being able to manage power. As the bit rate increases of those lanes, you talked about going from 25 gig or 50 gig to 100 gig, then we have to go to 200 gig really to get to 1.6 T. Every time you do that, the power consumption just driven up significantly. And so at some point, it becomes an impossible problem to cool that effectively. And so then you have to look at ways to actually take the power -- the things that are generating all that heat out of the equation. And that's really what drives co-packaged optics, as say, take out things like lead timers, take out maybe some of the serdes between the ASIC and the optic and eliminate those power consumers and put the optic very close to the ASIC. And in doing that, you can reduce the total power. So that's the primary driver. There may be an economic driver as well. But I would also say that in doing that, there's an entire industry ecosystem that has to evolve around packaging that optic. We have a good ecosystem around packaging optics today. When I talk about optics, I'm talking about something like this. We have an ecosystem that knows how to package these things and do all the work inside this to package all the electronics and the optics. We have an ecosystem who knows how to package ASICs, whether you're Intel or TSMC or whomever, we have an ecosystem knows how to do that. We don't have an ecosystem who knows how to do both together. So an ecosystem in the manufacturing sector has to evolve. That's going to take time to really mature that. And so I think that what's going to happen is we're going to have very, very smart people trying to do as much as possible to keep pluggables alive as long as possible because that's a technology we understand. It's an operations model our customers understand. And as long as it's technically feasible and the economics make sense, we're going to see pluggables continue. And for me, it's almost like people predicting the end of Moore's Law. We've had people predicting the death of Moore's Law for 20 years. And then we got smart people that overcome the last hurdle. And I think we're going to see that in the area of pluggables as well. We are investing in co-packaged optics. We do expect that at some point, we will hit a wall where you really do have to do co-packaged optics. I don't think it's at 50 terabits. I think it's going to be much higher than that. It's going to be at 100 or 200 -- 100 terabits before we have to consider that. So I think we will see 1.6 terabit pluggables, and maybe 3.2, but we'll definitely see 1.6 terabit pluggables. And that may be the end of the line before we have to move to a co-packaged optic, but we're talking like '25, '26 time frame for 1.6 T. So it's post that before I think you'd really need to see co-packaged optics. Fahad, one other point I think I want to make there. I think we have actually delivered co-packaged optics solutions to customers. We have delivered that, it's not deployed in volume and what the reason customers wanted was to understand what operationally is going to change. When I take a switch that has 32 ports on it that have pluggable optics like this, and now basically have a monolithic switch that has everything built in. What does that mean from my installation process? What does that mean for a failure of a port? How do I manage that? What does it mean for how manage the maintenance and life cycle of this thing? And that's a learning process that's going to have to take place for our customers. And we've had customers that say, "Look, I know I don't really need this right now at 25 terabit, for instance or 12 terabit. But I want to understand the operation history, so I'd like to deploy some of this and see what it looks like". And we've done that. So we've got some experience working with customers on that, but there's not a true need for it yet. And I don't think that we're going to see that true need until we get to around 100 terabit.

And that's certainly very true. I mean we've seen co-packaged optic being deployed primarily in PC use cases to date and there to addressing a very specific use case. I guess you mentioned something about co-packaged optics and Cisco doing it in-house. You acquired a company called Luxtera a couple of years back. There was a pioneer in silicon photonics. Haven't heard much about them. Can you share with us a little bit of an update on how Luxtera is doing, how the synergies have played out since you've acquired them?