00:00 [music]
00:03 >> Hi everyone and welcome to our channel Rock to Run where we discuss with
00:06 industry leaders issues around AI infrastructure and data centers. Joining
00:11 us today is Mark Bieberich, Vice President of Portfolio Marketing at
00:14 Ciena. So Mark, thank you very much for joining. Thanks for having me, Juan.
00:18 Look forward to the conversation. Great, Mark. So, how is the rapid expansion of
00:23 new clouds reshaping the optical and routing requirements you are seeing from
00:28 customers? Yes. So, the first thing to understand
00:31 about the new clouds or what we refer to as neo scalers is that there's a growing
00:37 number of them. They are increasingly well capitalized and they have a lot of
00:41 investment to go out and fulfill their business models. And the other key thing
00:46 to understand is that they're growing very fast. So, as they get their
00:50 businesses off the ground, they see incredible demand from enterprises, from
00:55 hyper scalers, from other cloud providers that need to support AI
00:59 workloads. So, what that's driving us to do at Ciena is to innovate at the pace
01:06 of their businesses to make sure that we can address their scalability
01:11 requirements, make sure that we can address what they need from an
01:14 operational standpoint. You know, many of the neo scalers do not have a deep
01:20 bench of optical expertise or network engineering know-how. You know, their
01:26 businesses are all around how to bring customers uh compute infrastructure and
01:31 compute services. That's really the main thrust of their businesses. So, the
01:35 network is not always something that they know exactly how to operationalize
01:41 and to scale and to build for the long term. We can help them, you know, with
01:45 our expertise, with our services staff. So, what we see is a rapidly growing set
01:51 of requirements. We see the need to help them scale their businesses and to
01:56 activate network capacity flexibly and fast. And we also see a need for
02:02 expertise, you know, and the ability to operationalize and maintain those
02:07 networks over period of time in which they're growing.
02:11 Great. And Mark, what unique network challenges arise when AI providers scale
02:17 without owning fiber or traditional data center footprints and how does Ciena
02:22 adapt to that model? Yes. So, in the case of the neo scalers, you know, many
02:27 of them now are looking at building their own networks. But if they do not
02:31 own their own fiber, if they do not have the ability to light that fiber, you
02:36 know, between data centers and in their network core,
02:40 then they're relying on the service provider community for wavelength
02:44 services or, you know, what we refer to as mofin or managed optical fiber
02:49 network services. And those services can be incredibly valuable to help those
02:55 customers get their businesses off the ground. But for the largest neo scalers,
02:59 as they really grow their business, they need more control over how the network
03:03 behaves. They need more control over how the network scales and grows over time.
03:09 They need more flexibility and control over how to add capability to the
03:14 network, things like security, for example.
03:18 So, we've seen this trend of the neo scalers build their own networks and
03:22 want to manage and operate them as well with our help. So, those challenges of
03:28 scale, of expertise that I mentioned before, you know, certainly apply to,
03:33 you know, their decision over whether to continue to use the services from the
03:39 service providers, wholesalers, or to build their own networks. We, of course,
03:45 see an incredibly healthy market, you know, for wavelength services sold to
03:50 the neo scalers. That market is growing fast. And not
03:55 every neo scaler is going to need to have their own network infrastructure.
03:59 Their businesses may not necessitate that level of autonomy and scale. So,
04:04 it's very important that they have solutions from the service provider
04:08 community. And those same solutions, of course,
04:11 offer a monetization opportunity for the service providers. They have the ability
04:16 to bring to market, you know, more flexible, more on-demand dynamic
04:21 services at higher rates of capacity, you know, to meet the neo scalers'
04:25 needs. Great, that's really interesting, Mark.
04:28 And also, AI clusters generate massive east-west traffic. What shift in traffic
04:33 patterns are you seeing and where do current networks fall short, Mark? You
04:37 know, it's interesting that you bring this up. We just had a podcast not too
04:42 long ago with our customer Verizon and one of the things we talked about was,
04:47 you know, the rollout of some of their AI connectivity services. Of course,
04:51 they see, you know, massive east-west traffic demands from some of their
04:54 customers and even on their own network. And, you know, one of the key challenges
04:59 there is, you know, the dynamic way in which AI workloads, you know, arise in
05:05 the network. So, the demand for AI applications is unpredictable. The level
05:12 of bandwidth that's required at any given time, you know, really can't be
05:16 predicted right now because we're still in the very early phases of, you know,
05:20 this AI boom. We're seeing all different kinds of traffic patterns emerge. We see
05:26 a lot of different types of behavior from enterprises, from hyper scalers,
05:32 from network operators. And, you know, that behavior
05:36 necessitates a lot of flexibility in the network. It necessitates the ability to
05:40 turn on bandwidth fast and in lockstep with uh customers' demand. And that, of
05:46 course, requires a lot of programmability in the network. One of
05:50 the things that we've done, for example, with Verizon and other customers is
05:55 provide them with, you know, the kind of programmability that they need in order
06:00 to deliver on-demand services. You know, we provide a lot of telemetry from our
06:05 network infrastructure so that, you know, their network operations layer,
06:09 you know, the multi-layer control and the operational support system layer can
06:14 identify what's happening in the network and then dynamically turn on bandwidth
06:19 as those workloads require. Uh so, we think that that's a very important way
06:24 of addressing, you know, these unpredictable traffic patterns that are
06:29 so indicative right now of the AI application uh boom. Yeah. And given
06:35 accelerating AI demand, which part of the transport network hit capacity
06:39 pressure first? Maybe long haul metro or interconnected links?
06:44 You know, where the demand rises in the network actually depends on the type of
06:48 customer that you're talking about. You know, some neo scalers, for example, you
06:53 know, operate most of their facilities within a given metro. Other neo scalers
06:59 are larger and more advanced in their business plans and as a result, you
07:04 know, they're starting to build even intercontinental or global networks. So,
07:08 there are times when some of our large neo scaler customers, you know, see
07:12 demand on submarine links, on long haul links in addition to, you know, the
07:18 metro links where they're serving up, you know, GPU as a service, for example,
07:22 to enterprises in a given metro. So, it really does depend on the operator. We
07:27 actually see bandwidth uh demand in all parts of the network right now, whether
07:33 it's metro or in the campus of a data center or in a regional or long haul
07:38 network and, of course, in the submarine where Ciena has a very long history of
07:44 providing uh capacity to the largest network
07:47 operators in the world. And Mark, are customers uh starting to co-design
07:52 network and GPU clusters together? Yeah, it's a good question, Juan. You
07:57 know, I think that most of our neo scaler customers are laser-focused on
08:02 getting their compute infrastructure in place, you know, so they can start
08:05 offering services as fast as possible. And typically, what they have is a, you
08:10 know, a smaller staff, you know, that's responsible for the network. Now, that
08:14 staff is working hand-in-hand with the rest of the technical organization to
08:17 understand what the network needs are. But I think that typically, you see a
08:22 lot of emphasis on the part of the neo scalers to, you know, activate and
08:26 operationalize their compute infrastructure first. And many times,
08:30 the network actually follows. I think over time, you know, as a lot of these
08:34 neo scaler businesses scale up and mature, you're going to see, I think, a
08:38 stronger bond between the compute side and the network side so that, you know,
08:44 all of the GPU as a service or other types of services that are offered right
08:49 out of the gate are well supported by a high-performing network. Today, we see
08:54 just a little bit of lag, but that I don't think that's uncommon for, you
08:59 know, an environment where these companies are really just getting off
09:02 the ground and trying to operationalize as fast as they can. But it is
09:06 important, you know, for the network to be sized according to the compute
09:11 infrastructure that's getting deployed. And we're starting to see that the
09:15 network engineering staffs, people responsible for the network
09:19 infrastructure, you know, uh really dialing in and making sure that they
09:24 have enough network capacity and that their networks are engineered to scale,
09:29 you know, as that compute infrastructure grows as well. Great. What is the value
09:33 proposition that Ciena offers in this emerging and really interesting space of
09:38 neo clouds? Yes. So, first of all, Juan, I think that Ciena is very well
09:44 positioned as a leader in enabling neo scalers to scale. We have now the
09:51 world's leading coherent optical technology available now at 1.6 terabits
09:56 per second. That is a distinct advantage over our competition and one of the
10:01 reasons why just having more capacity per wavelength is so important is
10:06 because most of the neo scalers don't have a lot of fiber. So, they need to
10:10 maximize the capacity they have on the fiber that they've acquired. They need
10:15 to maximize their bandwidth per wavelength that they activate. So, the
10:21 more they can get out of the fiber, the lower their costs are because they don't
10:26 need to add fiber. They're taking advantage of the fiber they have. So,
10:31 because they're scaling their businesses so fast, you know, that advantage really
10:36 matters. Helps them, add more capacity to the network at a faster rate and at a
10:42 higher rate than our competitors. We also have the ability to, as I alluded
10:47 earlier, come in and really help our neo scaler customers operationalize. You
10:52 know, from the design of the network to operating the network, maintaining it,
10:57 to upskilling, you know, their staff, even providing, you know, resident
11:02 engineers or other types of staff and services that help them get to market
11:07 faster, to scale up with their businesses plans, and to ensure that
11:13 they have scalable business model over a longer period of time. You know, that
11:18 level of expertise is unmatched from Ciena. We now have, you know, a market
11:24 share lead in data center interconnect, for example, with the hyperscalers. So,
11:28 we bring a lot of expertise building and operating the largest networks in the
11:31 world to our neo scaler customers. And so, our customers, our neo scaler
11:36 customers, have a level of confidence and assurance that we can help them, you
11:41 know, with any solution that they need and that we are trusted to get the job
11:45 done. One of the things I also mentioned, Juan, earlier is that, you
11:49 know, neo scalers run the gamut from large multinational corporations now to
11:55 emerging startups. You know, not all solutions for neo scalers look the same.
12:00 So, they need a portfolio that can be tailored to the needs of their business.
12:04 And at Ciena, you know, we have a wide-ranging portfolio of optical and
12:09 routing technologies that can help them operationalize and build a network that
12:14 works best for their business. You know, so, having that range of solutions,
12:20 having the broadest portfolio of optical capabilities in the industry, really
12:24 lends itself to enabling neo scalers to build the network that's right for them.
12:29 So, that's how I think Ciena is attacking this market and so far we've
12:33 been very successful. Great. That's really interesting, Mark. I would like
12:37 to thank you again very much for your time and your insight today. Thank you,
12:42 Juan. I've enjoyed the conversation and look forward to talking to you again in
12:45 the future. Great. And thank you all for watching and please stay tuned for more
12:50 conversations on this critical space.
No matching passages. Try another word.