Guy Daniels, TelecomTV (00:24):
Hello, you are watching the NextGen Telco Infra Summit, part of our year round DSP Leaders coverage, and it's time now for our live Q and a show. I'm Guy Daniels and this is the first of two q and a shows. We have another one at the same time. Tomorrow, it's your chance to ask questions on all aspects of infrastructure, including of course, the impact of AI workloads and traffic. Now, as part of today's summit, we featured a panel discussion that looked at how to build a future-proofed telco infrastructure and the key considerations for telcos and their partners. If you missed the panel, don't worry because we will rebroadcast it straight after this live q and a program, or you can watch it anytime you want on demand. Now, we have already received some questions from you for today, but we would love to have more.

(01:24):
So if you haven't yet sent in one, then please do so now and use the q and a form on the website. Well, I'm delighted to say that joining me live on the program today are Beth Cohen, SDN Network product Strategy at Verizon Business Group. Fernando Castro, Cristin VP and GM Telco infrastructure for Hewlett Packard Enterprise and Sree Nandan Atur, who is Technical Director Cloud at Rakuten Symphony. Hello everyone. It's good to see you all again. Thanks so much for coming back and joining us for this live q and A show. Let's get straight to our first audience question then. And the question is, how are telcos going to modernize their infrastructure and data center assets to drive AI native operations and to meet the new AI demands from both consumer and enterprise customers? Be Beth, perhaps we could start by getting your views on this one because this is such an important question.

Beth Cohen, Verizon (02:36):
It certainly is an important question and there's a number of aspects of it that I think is worth noting. One is that AI and telecom infrastructure aren't necessarily the same thing. A telecom could in fact build its own infrastructure to support its network and then tie it with a second piece of infrastructure to support the AI workloads. Because AI requires a huge data processing network. Workloads behave differently, so they're actually quite different types of workloads, and so to build the infrastructure to support them doesn't necessarily require an integration between the two workloads. Another consideration is how much are those workloads going to be sent to using a third party cloud provider that specializes in workloads that are highly compute intense and storage intense, and how do you get the network to connect? Obviously you need the data, the network feed itself, which forms the underlying data that drives the ai. So I think there's a lot of changes going on to meet those requirements. It's definitely early days and exactly how it's going to work out. I think each telecom is working through it on their own. That would be great if there was more consensus, but I don't think there is yet.

Guy Daniels, TelecomTV (04:28):
Okay, thanks Beth. That's interesting and hopefully we will see some more consensus as we go along. As you say, this is very early. Fernando, what do we come across to you? What's your thoughts here?

Fernando Castro Cristin, HPE (04:41):
Well, Beth said several times the word data and I think that this is the center of what we're talking about. So telecom industry is definitely the one that is transporting that data. So how AI can work without the right infrastructure to support that data exchange and that data manipulation. And I think that, like Beth said, we are in the early stages, so I think I used the analogy of a student going through the different stages and we are now in the stage where we are now creating the environment so we can monetize or reduce the cost depending of which workload we're talking or which aspirations we have for ai. So there is multiple initiatives ongoing in both of the main directions, which is reducing costs, optimizing on the other side, generating revenue, and I think that really telecom industry is there to connect all that together.

(05:42):
Now, telecom industries and industry like any other one that can have its own benefits, but he's right, those are totally different workloads, but are we doing AI already within the telecom? Yes, maybe not GPUs like everybody understands, but the intelligence on how to optimize the network is there since quite some time already in the core side, is the run revolution in terms of having an open run that going into the details, will that enable workloads to be influenced at the edge and generate some revenue for the telco or for the people using the telecom industry or maybe a shared situation in terms of revenue? Why not? All that is at the beginning and I think that the more questions we have around it, the more we can find answers, and that's a very evolving environment where definitely there is room to get benefit managing that data, and I think that's the center, how the data is managed and how we extract value from the data. I think telecom industry is in the center of that because it's transporting the data. So definitely a lot of things to moving today and will evolve in that direction.

Guy Daniels, TelecomTV (07:08):
Yeah, absolutely. There's a lot moving and yes, we will track the evolution of this. Thanks, Fernando. Beth, we'll come back to you in a moment, but Sree, I'd like to come across to you as well for your thoughts on this question. The question about how telcos modernize their infrastructure, what are your thoughts?

Sree Nandan Atur, Rakuten Symphony (07:27):
Yeah, so from the perspective of ai, there are really three things. One is it starts with data, then comes the training and then the inference. Now let's look at this from two aspects, hardware and software. So from the hardware angle you have storage, network and compute. So since you have massive amounts of data that is generated, you really need a really fast storage network. Maybe PCIE three NVME fast drives wherein reading and writing is really fast. Now this data needs to be transported to your compute clusters. So you have really fast interconnect, maybe churning around 25, 50 or a hundred Gs, and your compute clusters is a combination of not just GPUs but maybe even TPUs and cpu. This is because not everything like inference requires A GPU. For example, you can build AI models. If you take the far extreme, it takes about three months to build a foundation model with a large language model.

(08:27):
So for that, really you really need a GPU, but for small models it can be done in 30 minutes. So it really depends on what kind of AI workloads are running on your calco networks and what kind of data it is dependent on and the kind of inference that it needs, which means that the large language models is on your extreme left, which requires really large compute, expensive GPUs and GPUs to actually do your training and inference. And on your right hand side with maybe inference models, small models and inference can be done with cheap CPUs. So telcos really are innovating here as to depending on the AI workloads that are running on your different layers of your edge network for edge, edge and core, depending on your AI workloads, the kind of inference that needs and the kind of data that it is operating on, there are various AI models and that is how I think telcos are reinventing their networks to bring in AI from the perspective of hardware and software.

Guy Daniels, TelecomTV (09:35):
Great, thanks very much for those insights, Sree. And there's certainly a lot of interest around inference at the moment and the perceived volumes and claimed volumes and profitability to telcos. But Beth, I wanted to come back to you. I wanted to pick up on some comments.

Beth Cohen, Verizon (09:52):
I wanted to circle back to some of the things that Fernando mentioned, and I think Sree also mentioned as well. There's really two types of work stream AI work streams. There's the internal ones that the telcos use to optimize the networks. Those are very early ai. I mean, we've been optimizing networks for decades not using ai, and so it'll be interesting to see how much AI can optimize further on that level, but that's within the telco infrastructure itself. There's also been a push to make AI exposed to our customers and give them tools to allow them to optimize their own networks or rather optimize their applications as they are within their networks. And that I think is very much more exciting and newer approach to using AI and using the infrastructure to support it. And that's a much more complex infrastructure requirement because we need to bring in data from our customers, but of course there's privacy concerns, so we have to be careful about that. So all of that is obviously has more potential for revenue, but at the same time I think we have to be careful about what kind of infrastructure we would be building to support customer facing AI tools.

Guy Daniels, TelecomTV (11:40):
Yeah, interesting perspectives. Thanks very much Beth and Fernando, let's come back to you as well.

Fernando Castro Cristin, HPE (11:46):
Yeah, I just wanted to jump on one of the comments Beth just made and it's related to expanding a little bit to the security. I think that when we talk about AI and when we talk about data, you mentioned privacy, I'm mentioning security because you are dealing with the data and telcos transporting data from customers and from individuals and from enterprises. So security will become even more important than now, even it's today super important. It's going to be critical business and life critical. And the telecom industry is very familiar with security. It has been in the DNA of the telecom industry. So I think that there is two anchors there. There is the security DNA, I would say this way that it's provided today and will be kept if ai, it's delivered by the telecom industry as a product and there is a lot of confidence from the enterprise that the telecom industry will be able to provide the security as they do it today.

(12:59):
Then when we talk about the different workloads that we three mentioned from the inferencing for the LLMs, that requires super large computing capabilities to very dedicated and potentially at the edge that security has to be throughout all the chain. So I think that security is really an element that we need to make sure that it's part of the thinking when we want to deploy part of the key pillars of the strategy. And I think that there is consensus in the different actors in the industry that has to be, and it is not a wish. It has to be part of the solution. So it's super important to provide that and I believe that telecom industries in the center and able to deliver that as part of the continuous history of delivering services.

Guy Daniels, TelecomTV (14:04):
Absolutely. Let's hope so. Yes, thanks Fernando and Sree, let's come across to you for additional comments as well.

Sree Nandan Atur, Rakuten Symphony (14:11):
Yeah, I would like to add to a comment that Beth made actually about the AI use cases that are driving the telcos. She made a really good point from the operator's point of view and also what AI use cases would be there for the customer. So going from 3G to 5G, you have seen that the network has exploded with respect to the number of devices connected, the number of nodes, number of clusters, et cetera. So when there's so much data and network and a really large network to manage, AI plays a really critical role for the operator. It plays a really critical role in resource optimization, which leads to energy optimization and providing the best root cause analysis. So you can use your natural language to figure out at any point in time what is the health of your entire network and drill down to find out the exact root cause. Now all of this is good for the operator, but what does it really mean for the end user now because the operator is actually using these AI use cases, the network is cost effective and the customer is getting better value and better customer service. So indirectly the end users are being benefited with ai.

Guy Daniels, TelecomTV (15:33):
Great. Thanks very much for those comments and thanks everyone. Great. First question to open up our program with. So let's move on to another question, shall we? And the question is, considering that major telecoms vendors continue to push their own cloud infrastructures, is there still a business case for operators to build a generic telco cloud for functions from other vendors or what's the concept of a generic telco cloud, just a passing trend? Well, very interesting question and Sree, I'm going to come straight back to you if I can for your thoughts on this one. Is there still a valid use case here?

Sree Nandan Atur, Rakuten Symphony (16:18):
Yeah, I really don't think it's a passing trend. If you see the network, there is disaggregation at the hardware level and the software level, right? And at the hardware there is, you have different layers of the H compute with far H edge and core. And at the software disaggregation you have things that are running in far h, h and core different layers of the software stack. Now, if you take a look at, if you zoom into one of the software stacks, you would see at the base, you would have an operating system that is running, and on top of that you have some sort of an orchestration platform, either OpenStack or for managing your virtual machines or your Kubernetes clusters. And on top of that, you have your applications either in the form of version machines or containers. Now, because of this open nature, the operator has the flexibility to choose different kinds of hardware and software.

(17:12):
Now why would they want to do that? The two reasons, one is for the cost point of view, and the second is the risk. So from the cost point of view, there is a constant evaluation of different software stacks and hardware stacks. To keep the cost low, you really need to keep the cost low so that you deliver best results for your end users and your CapEx and opex costs are low. And to do that, you really need good automation, orchestration and automation. So if you narrowed down on the operating system, it takes about 30% of the licensing fees. Now, not everybody can build their own system. Red Hat has a rail operating system, and in fact the hyperscalers like Microsoft built their own operating system like Avenir, which is Linux based to avoid these licensing fees. So if the telcos start thinking from cost optimization point of view, they would say, Hey, what alternatives do I have for these software stacks and hardware stacks in order to optimize the cost and in order to optimize the risk of bringing in a new vendor? So you need vendor flexibility. So it is not just a passing trend and open telco networks are here to stay.

Guy Daniels, TelecomTV (18:29):
Great, thanks Sree, very positive there. It's absolutely not a passing trend. Fernando, let's come across to you for your thoughts. I mean the question's quite specific, isn't it? Is there a business case for a generic telco cloud?

Fernando Castro Cristin, HPE (18:44):
Yeah, fully agree with what Sree said. Just wanted to comment on another aspect. It's not only the optimization, it's also the capability to be dynamic. I think that the telecom industry went through all those different generations and each generation was bringing dynamic concepts and the fact that the telecom industry has to reinvent itself very fast. So I think that from a technology standpoint, this disaggregation allows diversity in the applications and more capabilities to react to the market. We saw a huge reaction needed during Covid, but multiple things and AI will bring that dynamicity also, I dunno if it's an English way, but this dynamic situation cannot be achieved with a very specific infrastructure. That was the case in the past in two G and 3G. It has to be part of the equation. And for that I think that this aggregation allows the telcos to have that dynamic change and that dynamic exposure to the market and being able to reinvent their network in a faster way.

Guy Daniels, TelecomTV (20:07):
Thank you very much Fernando, and more reasons there why there is indeed a business case for a telco cloud. But Beth, let's come across to you and do you agree with other panelists?

Beth Cohen, Verizon (20:19):
I do, but I want to bring up some other thinking. So Fernando mentioned that telcos need to be very dynamic and change rapidly, but we also at the same time invest in the long-term thinking. So when we invest in infrastructure, it will frequently be around for 10, 15, 20 years. And that's actually gets to be a problem. I know we have some infrastructure that's no longer supported by the vendor, but we can't snap our fingers and take it out. So there's this weird dynamic where yes, the telcos are very, very much cutting edge, providing the 5G infrastructure, the 5G, and now we're talking about six G, whatever that is. And yet at the same time, we only turned off 3G literally a year ago or maybe two years ago now. So it was quite recent. And so we have to commit and we have committed to something 4G for quite a few years. So there's a huge amount of complexity in needing to kind of have that very stable infrastructure underneath. And the disaggregation really just does get us to where we need to be, but at the same time we need to have that flexibility to be able to change things. So I think every telecom comes up with a somewhat different answer to that dilemma.

Guy Daniels, TelecomTV (22:11):
Great. Thanks very much Beth. And we'll come back to Fernando. Fernando, you want to respond to Beth there.

Fernando Castro Cristin, HPE (22:17):
It is not a response, it's just a comment. That's why the telecom industry is so it's a passion because you have, it's a regulated industry. You have to commit things, you have to stay for decades with delivering a service. At the same time you have to be dynamic in the most modern industry that you can find. So I think that Beth just explained how difficult the telecom industry is and why it's so interesting at the end, that's a source for engineering and source for innovation and innovation. Telecom cannot survive with innovation in boundaries that are quite difficult to maintain. So that's why it's a passion. You're there if you're passionate.

Guy Daniels, TelecomTV (23:07):
I couldn't agree more. Well said Fernando. Yes, it is very difficult, but once you're in, you are in. So thanks very much everyone for your responses to that question. And we've got another viewer question here. Let me read this one out to you. What are operators and vendors doing to mitigate the risks of using a single supplier for silicon for their infrastructure? The fabrication and supply of silicon has become a hot topic and the number of associated risk factors appears to be increasing. Well, Fernando, let me come straight back to, we have been talking about this for some time, but I guess a lot of recent coverage around AI has brought the silicon supply chain and ecosystem into more focus. What are your thoughts here?

Fernando Castro Cristin, HPE (24:01):
Well, when we talk about silicon, we can talk about different phases in time and also different areas within the network. So if we talk about core like we just did in the virtualization, then you have different silicon vendors. You have for example, broadcasting that is quite heavily using ARM in the service providers arm is used because it's, it's some technology that provides cost efficient in terms of power consumption. AMD has been there in the core also, many of the packet costs today in different networks are using AMD Intel of course is there if we move a little bit more to the run, yes, there was quite some intel driven situation, but more coming and I think that we'll have more detailed information coming from my peers here. Then when you go to ai, Nvidia of course, but not Nvidia alone and you will see evolution there.

(25:10):
So I think it's all cycles and if you want to drive a little bit C news weeds in terms of golf curves in between the different silicon vendors, they have their apogee, they have their competition coming and all that, different phases that are interlocking together and the telecom industry overall is there to orchestrate those different golf curves. So yes, I think that we need to talk about the complexity of building silicon. It's not easy to build a plant and to build those super high tech devices we are talking about any vibration may impact the quality. So where do you put the plant? Geopolitics will come into the equation, how important the isolation of the plant is from the rest of the world. A truck passing through a plant will generate vibrations and if you are building with those very high-end silicone is not going to fly. So all those things makes that the changes and the dynamicity of the dynamic, I need to run the right wording there, but you need to weigh those phases and those learning in order to be able to beat those silicons in order to see competition happening. But in every sector, like I said at the beginning, competition came and improvement of the technology came with that competition. So it's around phases I think.

Guy Daniels, TelecomTV (26:57):
Yeah, absolutely. Thanks Fernando. And there are many different types of silicon product out there as well at different levels of complexity and different manufacturers and fabricators. Beth, let's come across to you for your thoughts on this.

Beth Cohen, Verizon (27:12):
So to a certain extent, the telecom industry is at the mercy of these chip vendors. And as Fernando mentioned, there's just a handful because there's a very high barrier of entry to build a new chip plant and they literally cost over a billion dollars each. And they have very stringent requirements. And of course the geopolitical issues have been simmering for a while. I think they're increasing right now with and telecoms aren't in the chip business. We're buying hardware that has chips in it. So we can support the industry obviously, but we don't have much control over which vendors are working with which hardware manufacturer. And as Fernando mentioned, it's not just perspective, it's not just the routers and it's not just the infrastructure to support the generic infrastructure to support in the cloud data centers, but it's also the mobile devices and the PCs and everything is interconnected. Another industry that is actually of greater interest to the telecom industry is the fiber, the fiber optics industry, which is also very, very centralized and monopolized by only a couple vendors because again, it's very specialized, very expensive to get into the industry. And so I know that several telecoms have actually locked in contracts with these fiber optic cable manufacturers because they know it's so critical to their industry.

Guy Daniels, TelecomTV (29:26):
That's interesting as well. Thank you very much Beth and Sree, let's come across to you and your observations about the state of silicon and telecom's reliance on perhaps just a handful of vendors.

Sree Nandan Atur, Rakuten Symphony (29:40):
So I would like to add a few comments to what be said. I completely agree with what she said With respect to telcos are not really resource hungry in grabbing thousands and thousands of GPUs, but the pressure is more on the fiber optics. So in contrast, it's not really like Tesla or Meta who are trying to acquire a hundred thousand GPUs. Telcos are not there yet. So the accelerator cards, what it is called in telcos on the FAR H networks is actually used for fronthaul packet processing today it's actually done through maybe FPGA cards or ES ESC cards. And replacing that with an actual GPU is going to have cost effects because if particular the foreign server is about 25 grand, adding a high cost GPU would double it and you don't really need that. There is no use case for it. So the GPU really make sense in your central data center and the requirement of the scale of the GPUs is not that high, maybe a few hundreds and the pressure is still there. The order needs to still go into the manufacturer, but not so much.

Guy Daniels, TelecomTV (30:53):
Great. Thanks for those additional points, Sree. Thanks everyone. Well it is time to check in on our audience poll for the NextGen Telco Infra Summit. And the question we have been asking you all this week is what are the most important investment areas to create a future proof telco infrastructure? And look at that. The real time results are appearing right to my right there, fairly close and evenly spread out service-based architectures and AI factories and AI edge computer just edging ahead there with perhaps surprisingly, perhaps decentralized intelligence to the edge currently in seventh and last place. But if you have yet to vote, then please do so. You can influence these votes and these results. We will take a look at the voting during tomorrow's live q and a show.

(32:04):
We still have time for more questions and we are receiving more. And I must say at this stage, if we don't get to answer all your questions today, we will put them on the top of the pile for tomorrow. But before we get onto the next lot of questions, it's probably worth asking our guests here. Any thoughts about the viewer poll and the question we've been asking, I mean it's maybe early stages with the voting, but I was quite surprised there about AI edge AI implementations scoring quite highly and yet decentralized intelligence is not scoring terribly highly at this stage. Any thoughts from the three of you? I mean, Beth, you haven't had much chance to look at these, but it does seem that it's fairly wide open and there's no one leader there. It is just because we just simply don't really know no strong opinions.

Beth Cohen, Verizon (33:09):
I think that it's a situation where it's kind of a mixed question and if you look at the results, it falls into two things. It's ai, yes, and then infrastructure, basic infrastructure. And I think the two got mixed up in some of those answers. So that's why that's surprised me. Hey, we want some AI at the edge and then oh yeah, but we're not interested in actually putting any compute at the edge. Well, that doesn't make any sense because if you have AI at the edge or rather if the data's at the edge, you're going to have to want to put AI at the edge because you're nobody in the right mind is going to want to pull all that data into the core, soak up tons of bandwidth just to do the compute at the core instead of at the edge. So those are architectures that setting aside ai, that kind of edge architecture has been something that's been bubbling under the surface for a couple years now. I've been involved in that going back at least I'm thinking like 10 years now. And so I think that it's a mostly comes down to not quite understanding the question.

Guy Daniels, TelecomTV (34:36):
Well put, well put Beth. Yeah, I would agree with that. And I think you sprinkle some magic words and acronyms in and you get surprising results. Fernando, any thoughts from you? I know it's very early stages with the voting, but any initial thoughts from you?

Fernando Castro Cristin, HPE (34:54):
Well, I think it shows that there is not yet any maturity in the market about what do we do with ai. And I think that the more than not understanding the question, which I agree, it's difficult to understand the question when you don't have in your head a clear mapping of the, I want to use again question, but a mapping of where would do I put my AI for what and at which cost and there's no magic recipe. And so I think that the questions were being interpreted in different ways by the different population of people answering the questions depending on from which angle they were looking at to it. So it's quite interesting to see that there's a hybrid situation today where I want to do, but I don't know why I want to do it, but I know I have to do it.

(35:57):
And so that goes to these kind of answers where it's quite medium and if it's medium, it means you don't have a strong opinion and you don't have a strong opinion because do you have to date the capability to make your own opinion based on what you really need to do? And so that's I think a big trend in the industry to try to rationalize from concepts to reality. And I think we are moving there and I think that the first implementations and the overall industry is moving into how can I adopt it in a reasonable price point in a reasonable location and for which reasonable outcome and for which benefits. So it's moving there, it's moving on that direction and I'm sure that this poll next year will have way more strong differences in between the answers. I hope. Let's do it. Let's do the test.

Guy Daniels, TelecomTV (37:05):
Yeah, we'll be back in 12 months and see, but you are right, we do have viewers from different backgrounds, more technical, some more marketing, there's different backgrounds and they're going to answer differently. I think we ought to get my colleague real matri to do one of his in-depth perception studies and reports into this area and drill down a little bit further into this. It could be worthwhile. Right? Thanks very much for those quick observations on our poll. And another reminder, if you haven't voted yet, then do so now it's on the website. Meanwhile, back to our view of questions and here's our next one. The freedom and choices afforded by open networking come at a cost. For example, integration is a major additional overhead for telcos. Is it time to reassess the merits of open telco infrastructure? Well, Beth, I'm going to come across to you first. I'm sure you've got views on this one. Is it time to reassess the merits of open Telco?

Beth Cohen, Verizon (38:06):
No. Open Telco is here to stay. It's changing obviously over time, but it is absolutely essential to invest in open infrastructure, open support. And I've said this many times, telcos don't compete on their infrastructure. Telcos compete on the services that they offer to their customers. Their don't care about the infrastructure they work. So it behooves the telcos to use open infrastructure whenever possible. And proprietary infrastructure has its own challenges, does open, and we end up at the end of the day with a mix of both integration. I think it's kind of, I would say a bit of a red herring to say that integration is easier with closed infrastructure, proprietary infrastructure. I think integration is difficult no matter what. Remember telcos have a whole setting aside the network infrastructure. We have a whole ecosystem of BSS systems, billing systems, tech, tech support systems, ticketing, all sorts of these systems that surround the core infrastructure and support it. And those systems are decades old and they're built change. And again, why should telcos invest in modernizing these or making huge of this technology when our customers aren't particularly interested in paying for us to do that? And if they work, they'll continue working. So yeah, open infrastructure is definitely here to say

Guy Daniels, TelecomTV (40:14):
Thanks Beth. That's very, very clear. Fernando, open infrastructure here to stay.

Fernando Castro Cristin, HPE (40:21):
So more than agree here to stay. I think it demonstrated the benefits, but I think that it's very hybrid and it's even within one operator. Let's take a case of an operated deploying in very diverse type of topologies, even in different countries. They will need both. And it's a complex environment. So if you pull out somebody that comes from let's say a totally different industry, even a technical person that would work in the healthcare industry and you put them into a map of a network just by the acronym, it takes six months to understand. So it's complex. And I think that there is areas of the network or areas of the services that have to be delivered by the telecom industry to the customers being individuals or enterprises that will be more adapted to adopt an open technology and in other cases more efficient, cost efficient to be proprietary infrastructure. So it's a mix and it'll stay a mix. I believe the cost and being able to adapt, those are two things that are extremes and I think that it's a hybrid world. It's just a hybrid world.

Guy Daniels, TelecomTV (41:57):
Yep, great with so much it is never one thing or the other. We often there's a middle ground as a hybrid middle here. Thanks very much for those answers. Look at the time, we just about have time for one final question. I think this will be our final question for today's show. The question from our viewer is, will next gen infrastructure be led by general purpose hardware and software or by specialized bespoke hardware and software? Beth, I'm going to come to you first on this one perhaps because I suspect we've heard this question

Beth Cohen, Verizon (42:38):
Before. We have heard this question before, and I'm an industry veteran so I can say the pendulum swings back and forth. So we have general purpose hardware, then we have dedicated hardware and then back and forth over time. And there are some areas where general purpose will always be the right answer. And there are other places where specialty hardware is always going to be the right answer. And then there's the mix in between. So a historic perspective, the telcos always have to keep their options open. Again, we're aware companies, we are dependent upon our vendors. But I personally think that getting back to the earlier question about open source, we need to have a stronger voice in making sure that the conversation about open source and open infrastructure and less proprietary systems is on the forefront at all times. And I think we don't do as good a job about it within the open source communities. And I'm a big proponent for open source making sure that the telco requirements are in fact incorporated into those projects. Hey, it's back and forth, back and forth.

Guy Daniels, TelecomTV (44:21):
It does feel like that at times. Beth? Yes. Fernando, what about you? Because this is a question that interests a lot of our viewers. Are we looking at a general purpose future or a specialized bespoke future? What are your thoughts?

Fernando Castro Cristin, HPE (44:42):
So I think that it's like when the very beginning of this journey of having general purposeful compute, we all thought that one system can be the solution, the magic solution for everything. That's not the case. So Beth said it's back and forth. I think that there is challenges in every implementation, and I think those challenges are good for engineering. It's pushing us to do the best US everybody. I'm not just talking about the infrastructure, but Beth said you depend on your vendors and likewise, it's all together. So all the challenges of very high specific hardware, they are very specialized and they're the best in their situation. It pushes us to innovation in order to meet those requirements in a different way and offering alternatives. And I think that this stimulation that generates that back and forth, we always try to do the best, the more open and the more cost effective. We haven't talked about yet today about energy too much, but how we cool those things, how we make sure that we don't use energy, only the energy that we need and less and less every time that is driving innovation. And honestly, the telecom industry has been king in pushing for innovation. That's where the vendors are, depending from the telecom industry itself. We are here to innovate because we need that innovation just to survive in that scope. So back and forth. Fully agree, fully agree.

Guy Daniels, TelecomTV (46:36):
Thank you very much Fernando and Sree, let's come across to you for your thoughts on this final question about general purpose versus a bespoke future.

Sree Nandan Atur, Rakuten Symphony (46:46):
Yeah. In addition to what Beth and Fernando mentioned, I'd like to give you a different viewpoint. So if you have a number line on your left is general purpose hardware on your and on your right is bespoke hardware for the telcos. There are technical requirements why you really need a bespoke hardware. Let's take one example, which is the Fronthaul packet processing. Now the bespoke hardware gives NH over general purpose hardware. Why? Because the axillary cards, for example, are 50 times faster. Now can you do the same thing over the general purpose hardware? Absolutely you can, but you cannot meet the service requirements or the time requirements. The packet processing has to be done in an order 0.1 to 0.2 milliseconds. So the technical requirement says that you cannot do this with general purpose. So you really need a bespoke hardware. And similarly for AI use cases, so but mentioned it's a pendulum. You go back and forth, right? So it is not like zero 100 or 100 zero, it's like a point on the number line. Maybe like 80% is general purpose, 20% is bespoke or it may be 90, 10, 80, 20. It all depends on the complexity of the network, the number of the scale of the network, the use cases, the customers, the operator, the geographic region, et cetera.

Guy Daniels, TelecomTV (48:07):
The pendulum swings. Sree, thank you very much indeed for that insight. Thanks to all of you for answering that last question because we are now out of time. Thank you so much to all three of you for joining us for this live program. Now do remember to send in your questions for tomorrow's live q and a show as soon as you can. Please don't leave it too late. We have a couple of questions here that came in late and we will look to answer those and address those in tomorrow's show. And please do take part in the poll. There is still time for you to have your say and you can find the full agenda for day two of the summit on the telecom TV website. It includes a panel discussion on how telcos can support AI and the emergence of the so-called AI factories. Plus we have another exclusive interview for you. And remember, you can see these programs on demand from tomorrow morning and for our viewers watching live. In case you missed today's earlier panel discussion, we are going to broadcast it again in just a few moments. So don't go away. We'll be back tomorrow with our final live q and A show, same time, same place. Until then, thank you for watching and goodbye.