How AI Is Fueling a New Fiber Supercycle

Hi everybody, I'm Linda Hardesty, executive editor and chief analyst with Fierce Network. Today I'm here with Noah Taylor, he's head of market intelligence and growth strategy for broadband at AFL. AFL is a global manufacturer specializing in connectivity solutions, including fiber optic cable and hardware, as well as test and inspection equipment. So, we're going to talk today about the fiber landscape, both as it relates to data centers as well as fiber broadband access networks. Welcome, Noah.

Great to be here, Linda.

I wanted to kick it off with how is AI creating a new fiber supercycle or fiber landscape? I mean, what all is AI doing to change things?

Yeah, great question and quite the hot topic these days. I think it's good to start with, AI is really driven by data centers, I think a lot of us understand that and know that, but what many people not realize is that data centers are no longer just in one single building, it's often multi-campus buildings that have to be interconnected. And so because of that, we're seeing a huge influx of fiber demand just to support the infrastructure required for AI, for both the training and inference side as well. And a lot of it too is there's a constraint in power, it's pretty well-defined in the news cycle, and because of that, they're looking for sites where they have access to power, can they bring along the fiber access to that? And so, spreading that apart really means adding more fiber to the network. RVA did a great study with the Fiber Broadband Association, and what they showed was AI is really going to require over 92,000 new route miles over the next five years.

So, an incredible amount of fiber, and I think when you think about how that scales with fiber count, most of these routes are several hundred fibers, if not thousands of fibers in a single cable, that could be over 213 million miles of fiber just to support this AI supercycle. So, a lot of influx here and I think there's a lot of cyclical business to the broadband industry. So, we see fiber demand go up and down. But so far with the AI boom here, it's really just an upward trajectory from a fiber consumption standpoint, and quite frankly, I think we're going to see this over the next five to 10 years.

Oh, that's interesting that you brought up the power stuff because AFL is also involved with power solutions. Is that a synchronicity for you all?

Yeah, no, exactly, I'm glad you brought that up. So, you kind of mentioned it in the intro, but AFL was founded back in 1984 as a joint venture between Alcoa, which is the Aluminum Company of America, and Fujikura. So, Fujikura brought the fiber expertise and Alcoa kind of had that heritage in the energy market. So, it's woven into our DNA, Linda, we have been in the industry since 1984 in that energy market, and of course now we've moved into broadband and data center as well. So, we touch all three sides of the AI equation, if you will. We've got the energy piece, we've got the broadband to connect people to the models, and then we also have the data center, which of course is standing up the AI in general.

Well, what's special about data center interconnect known as DCI in today's market?

I think about this in really three ways, and I think it'd be helpful to shape the conversation around these three things. One is inside the campus. So, fiber counts are continuing to climb, and I think we're starting to see copper retirement. A lot of that, back to our power conversation, is driven by power and economics really. Fiber is the cheapest way to move a bit of information of any medium. So, whether you take copper, you of course compare that to fiber or wireless infrastructure, or even in satellite, across all those different mediums, fiber's the cheapest way to move a bit of information. And so, the hyperscalers know this, and so when you think about building out these massive data centers that have to move tons of information, fiber's clearly the best way to do that. And so, inside the halls, we're starting to interconnect all these data center halls, a lot of that copper's coming out and now we're starting to see a trend for co-package optics of moving fiber to the chip, so bringing it even closer.

And a lot of that is a latency play, but I think this is a trend we're going to continue to see is a lot of fiber consumption inside the data center. So, that's point one. Point number two is looking at the metro or regional connection sites, and so if you go to Ashburn, Virginia or any of the hotspots in Texas, it's not just a couple buildings in one area, there's multiple building campuses or clusters, if you will, in that region. Atlanta would be another example of that. And so, you have these metro or regional connection sites that all need to be interconnected for that training phase. And then the third piece of this is the long haul. So, this is connecting region to region. And so, a couple examples I would give, you have Ashburn, Virginia, quite a hot spot, Atlanta, you've got Texas, you've got California, Kansas City, Chicago...

All of this needs to be interconnected, and those are very long routes in between these major cities and they have to be high fiber count as well. And so, all of this is starting to drive even more pressure on the fiber infrastructure. And if you're old enough to remember back when we didn't have interstates, Eisenhower realized we need to build an interstate system to connect all our cities, and we're doing the same thing in this modern day for the digital super highway, if you will. We've got interconnection today, but the fiber counts are really being maxed out, there's only so much we can do with spectral efficiency, and so we're starting to see a push to really build up and improve the fiber super highway that we have in the United States. And looking at this from coast to coast, from California all the way to the East Coast, we have to interconnect our country better. And that's again for training and it affects the inferencing side of AI as well.

Yeah, it's really been a boom for the fiber industry and a little bit unexpected, at least for me.

Yes. Yes. And I mean, an article you just wrote about of course was BEAD, and how Illinois just finally signed off on the NTIA, that's another program that's coming online and hitting about the same time as this AI infrastructure. Of course, you had RDOF that's been being built out for several years. And then we have commitments from the major tier ones like AT&T and Verizon, they're also looking to continue fiber to the home at least until 2030, the end of this decade.

So, Linda, when you think about all of that happening at once, you really put a constraint on the fiber supply, and then, oh, by the way, there's conflict in the Middle East that disrupts helium, and of course, a lot of people don't realize that, but both helium and germanium are top of the funnel inputs into the fiber process. And so, a lot of that helium comes through the Strait or Hormuz, of course, which we all know and see at the gas pump, is being constrained at this time due to the conflict. So, there's a lot of pressures, I think all kind of confluencing together at once and creating the supply dynamics that we see today.

But it is boom times, so that's good.

Yes, yes, yes.

Well, let's talk about the old stuff, legacy stuff, when will the great copper retirement happen?

Great question there, and I think it's something we've been watching for a while. We continue to see improvements in DOCSIS, and I think we start there. A lot of work has been done both at CableLabs and with the MSOs to improve DOCSIS and copper as much as we can. But again, you go back to the fundamental first principles thinking fiber's the cheapest way to move a bit of information. And so, I think they know that, they want to move to fiber at some point, a lot of them, again, are operating a hybrid fiber coax network. You see some of the major system operators, they'll advertise as powered by fiber, or they'll work fiber into the language somewhere, and it's a truthful statement, right? Fiber is involved in their network, but it may not be involved in that last mile connection.

So, I think they're fighting this battle of trying to get the most out of that copper asset that's in the ground today, but then also we need to think about the future, and they're getting pressure from a lot of different competitors who are building out full fiber to the home. I think from a consumer standpoint, we all want a fiber connection, I think inherently we all know it's better. And so, those are all pressures, I think, leading them to think about how do you upgrade? AT&T, if you think about the tier ones, both AT&T and Verizon have committed to retiring copper, AT&T especially, I mean, they talk about this in their earnings call, they're really targeting to have all their copper shut down by 2029. And a lot of reasons for that, not just the speed, but I mean, it costs them over $6 billion annually to run 4,600 copper wire centers. So, there's a lot of costs involved, they know fiber's better, and so it's how do we turn that off?

But you can't just flip the switch. When you think about 911 access, this is a fundamental right that we need to have in case of an emergency, a lot of that is supported by copper infrastructure today. And so using AT&T as an example, I know they're doing a lot of work with both fiber to replace that copper, but also fix wireless in those hard to reach areas to make sure those consumers are connected. And I think we see some of this copper turndown being tied up in California, I think that's a large part of the concern is are they adequately covering the customers that they had so they can get access to emergency services and things of that nature? So, kind of a complex thing that's going on, it's not just as simple as turning it off, and it's also building the fiber infrastructure. And I think when you think about hurdle rates and investment and when's the best time to switch over to full-fledged fiber, during a fiber boom when fiber's constrained may not be the best time to make that move.

So, I think all of this is going into the calculus for them. I think internally we're starting to model, and a lot of analysts have done a great job of looking at this, of how much does it cost copper to continue to run? And really over a 10-year life cycle, you're going to spend more on copper than if you would to go ahead and switch over to fiber. And so, I know a lot of these copper plant operators have a lot of smart people on their team, they're probably already looking at this and understand the dynamics there. And so, I think we're going to continue to see them try to switch that copper off. We've got the commitments from the tier ones, and I think we'll quickly see a switch over from the MSOs and other tier two and tier three operators as we get into 2030 on into 2035.

That's funny you're talking about all the pressure that they're under at really maybe the worst time when we're building so much fiber across the country to try and do this retirement. But at the same time, I think they want to repurpose all those central offices that handle all the copper.

Yes, yes.

So, definitely hand it to the actuaries to figure out cost benefits.

Yes, exactly. And I think that's an important thing as we talk about edge too, because those central offices are a great spot to put edge data centers. And there's a lot of negative public sentiment today about data centers, there's no doubt about that. And I was at a conference last week, Linda, and something that cracked me up... And we're talking about people complaining about data centers on Facebook, which was using a data center to complain about. And there's this this circular conundrum that I think consumers don't understand is most of what you do online comes from a data center. If you watched Netflix last night, it was from a data center somewhere, hosting those files. We think about the cloud, but because we don't see it every day, we don't really understand what's all involved in the background.

And I think that's important, especially as not only manufacturers, but providers in this space to educate the market and let everybody know data centers are what power the modern day life and really give us access to all the things we enjoy, like connecting with our families on Facebook or watching a Netflix film in the evenings. All of that, again, is powered by data centers. So, a lot of concern there, and there's definitely a huge public debate going on about data centers, the not in my backyard sentiment seems to be very strong right now, but everyone needs to understand we definitely need them to power everything we love and use in the digital world.

Yeah. All right. Well, so segueing from that, I wanted to ask about convergence. So, how do you see the convergence of fiber, wireless and edge playing out in the next five years?

Convergence has been talked about for a while and maybe even the term convergence has been overused. So, we'll start there, but there really is no wireless without wireline. And I think a lot of people don't understand that. I've talked to members of my family and they're like, "Why don't we just go to everything wireless?" And I'm like, "Well, everything wireless is powered by some type of backhaul and front haul that is powered by fiber." So, you need that infrastructure. I mean, if you think about it, every radio that's on a tower somewhere that your phone's connecting to is powered by a fiber optic cable that eventually goes into the backbone of our superhighway network. Same thing for satellite. Satellite is fantastic for remote areas where no connection is possible, it is definitely a fantastic technology. But again, all of that satellite eventually comes down to a terrestrial network and is handed off on the back haul.

So, even for a satellite connection, there's still fiber infrastructure being provided there. As I look forward over the next three to five years, some exciting developments is in the next PON evolution. So, a lot of networks today are operating on maybe GPON, our XGS-PON, 10 gig PON. As we look to 25 gig and even 50 gig PON, there's a lot of exciting things happening from a technology standpoint, where we may actually be able to finally converge both wireless backhaul and standard broadband PON access networks. And so that would be extremely powerful. We've spent so much investment and continue to spend investment on bringing fiber to everyone's house, it would be fantastic if we could also run over that PON wireless backhaul as well because that's going to give us access to more small cell sites, give everyone better connectivity and that we all come to really take for granted, quite frankly.

So, I think that's super exciting, I think we're going to see that continue to invest and evolve as we move into the 2030s, and concurrently, that's on the fixed broadband side. On the wireless side, we have 6G standards coming out, they're supposed to complete around 2028 was the last timeline I heard, and then we'll start to see some of those networks hit around 2030, I think there's some talk of trialing it at the Olympics in Los Angeles come 2028. So, again, that's being developed during this whole AI fiber supercycle, but if 6G does require some densification, it'll be fantastic if we can put some of that backhaul on the 50G PON as well, and that's just going to make those upgrades a lot easier, and maybe we'll finally get the promises that we all heard in 5G. I think we all heard the rave about 5G, but it didn't quite pan out from a speed standpoint as we thought.

It's just taken several years, and I think from a consumer standpoint, maybe there's some fatigue from the marketing there. So, for both of those, I think there's a lot of exciting things happening, and when you combine this with AI and what it can do to our networks, it's only going to drive efficiency and make the experience better for everyone.

So, we've covered a lot of ground, there's a lot of exciting things happening in wireless, and wired obviously with fiber, and with all the AI activity. So, what do you see telecom investments looking like through say 2035?

Let's start with fiber, we'll work through several technologies here, but with fiber, the FBA reported in their build report last year that 84 million people have been passed, so that's unique passing, so that's about 60% of the country is where we're at today with fiber. We're targeting around 90% by 2030, and I think that's fairly in line with some of the marketing materials seen coming from our tier ones, they're trying to wrap up a lot of their builds by the 2030 timeframe. I think RDOF will wrap up, so will BEAD. Again, that's supposed to be a three-year build cycle, so that'll be around 2029, 2030. So, from fiber to the home standpoint, we're really going to start to see some saturation as we get into the 2030s. From a wireless standpoint, a lot of exciting things there, and one thing I think to keep an eye on is most of us use AI from a mobile application standpoint.

AI helps us think, and we're all mobile, we don't just stay in a fixed location. So, although we need to upgrade our fixed wire line as well, the wireless side also needs that upgrade. And from a wireless standpoint, it is the upload. We're typically very download heavy today, and I think AI shifts that paradigm to make it more of an asymmetrical, from asymmetrical to more symmetrical network. And so, we're going to see a lot of investment there. I saw a report from Ookla, they have fantastic research, highly recommend, you can check out their work over there, but they model what China has done compared to the US, and we're kind of behind from a symmetrical network standpoint on the wireless side.

So, I think we're going to see a lot of investment in that. Regardless of how 6G pans out, I know there's a lot of debate on whether it's going to add densification or not, but regardless of that, I think we're going to see some investment at least on the radio side to make sure we have symmetrical capacity. And then, copper, I think we're going to continue to have that retirement again for all the reasons that we've discussed, but I think that'll start to wrap up around 2030 on into 2035, we'll see a lot of pressure to get that retired. And then edge, how do we monitor AI? A lot of it is latency-based, I think is where the problem is. Today, a lot of people have access to gig speeds from a bandwidth standpoint, but maybe it's the latency that's more of a concern. And when you think about AI and all the applications that it will have for real time interaction, we're going to need lower latency.

And so, a way to solve that problem is really pushing stuff closer to the edge from a data standpoint. So, think about it, I'll use Netflix as an example. Instead of watching a show and it has to go all the way to California to pull your video file and come back to your house, what if it only had to come to a local neighborhood? And so, we can do that with some of those copper retirement sites, Linda, that you mentioned, but also maybe at some of these macro cell locations. Can we place a small modular edge data center that can help offload some of that load and really increase the latency? And I think we're going to see a push for that, especially as we see AI devolving all its application usage. And from small things, I mean, anytime I open up a PDF, we have Adobe on our company laptops, but anytime I open it up, it gives me a synopsis of what did the PDF say.

And if you think about how many times you open a PDF and it has to go to a model, get inferencing information and then come back, that's just for PDFs, and then of course, Microsoft is integrated into everything. So, there's a lot going on here, and I think the market is quickly changing to adapt to AI and all the benefits that it has. And so, when we look out into 2030 into 2035, I think there's no doubt we're going to continue to see data center interconnect remain strong and just building out that super highway. That's not something we're going to do overnight, especially with the fiber constraints, this is going to be a multi-year, if not a decade cycle that we're going to monitor. And then, as we move into edge, how does that help with the latency standpoint and round us out to make sure we're fully prepared to live in this AI world that we're all building and taking advantage of?

Well, exciting times to be in telecommunications.

Yes, very much so.

Well, thank you so much, Noah, it was awesome talking with you today.

Thanks, Linda, great to be with you.