Sean McManus, TelecomTV (00:22):
Hello, you're watching TelecomTV and our extended coverage of the DSP Leaders Report series. I'm Sean McManus, and today's discussion looks at some of the points raised in our most recent report, telcos and ai. That report is based on the responses to a survey of the telecom sector conducted during late March and early April. We asked a broad range of questions about the impact of AI on network operator, business and operational and infrastructure strategies. We received more than 200 responses from individuals ranging from CEOs to engineers who work at a broad range of companies, mainly telecom technology vendors, network operators, and systems integrators. If you are a registered viewer of Telecom tv, you can download the report now for free. There's a link in the text below this video, and if you're not yet registered, then why not Go ahead and do so. Registered users have access to all of our reports and videos, and now that we've published the results of our latest research, it's the perfect time to discuss our findings with our special guests.

(01:28):
So joining me on the program are Michael Clegg, Cice President and General Manager for 5G and Edge at SuperMicro Mark Gibson, VP Software at Connectivity. Julien Sicart, SVP Digital Services, orange Business, the enterprise Division of the Orange Group, and Robert Curran, Consulting Analyst, Appledore Research. Hello. Good. See you all. Thanks so much for coming on the program. To help us delve deeper into our findings, we're going to focus on some of the points that came out of our research and look at these in more detail. Let's start by looking at an increasingly hot topic, sovereign AI services and the infrastructure that supports such services. There's increasing talk about the value of sovereign AI services and the potential for telecom operators to play a role in providing them. Some have already started to or are planning to from facilities often referred to as AI factories. Currently these are data center facilities, housing graphics, processing units, or GPUs. As survey asked, should telcos develop their own sovereign AI factory to offer AI services to the enterprise sector in their domestic markets? As the graphics shows, 54% believe sovereign AI is a strong business opportunity for telcos. While 27% think that telcos should leave this particular service to IT specialists, a notable 19% are unsure about the potential of sovereign AI services for telcos. So is sovereign AI a new business opportunity for telcos? Robert, what's your view?

Robert Curran, Appledore Research (03:15):
Yeah, our view is sovereign ai, something that is real. I think over time it's likely to be something that's mandated by regulators of telecom in different countries. I think there's a pretty strong appeal for combining secure connectivity and then locally constrained ai. Certainly for government entities, regulators, and certain classes of enterprises. Douglas have a strong reputation for trust and resilience. So there's an opportunity to reinforce both of those and offer an alternative to what's available, for example, from hyperscale cloud companies. And it's worth noting that the hyperscalers have been active, obviously improving data sovereignty offers over time, but in the context of ai, this isn't just about compute location, it's also about the supply chain. I think of AI models and cells, how they're generated, maintained, and so on. So I think in our view, if you are going to want that kind of requirement, then telcos are likely to be a good place to go. So we think sovereign AI is definitely an opportunity for telcos. They're very a good place to be the right partner for enterprises to work with.

Sean McManus, TelecomTV (04:24):
Brilliant, thank you. Michael, let me bring you in there. What's your view? Yeah,

Michael Clegg, Supermicro (04:28):
I agree. As we see you mentioned the number of operators already committed to this. We have operators like SKT saying they're going to be an AI telco. We have syntel doing some activities. It seems to be a little bit more in Southeast Asia that we've seen quite a bit of activity and also a little bit in Europe. For me, I do think it does depend a little bit on the country though. I believe a lot of the value in southern AI is also going to be in government services, government related services where you've got a lot of government information, personnel information that is of value is you don't really want to have it act in terms of identities, theft or so on. So the whole data sovereignty, keeping it local, keeping the private is a key factor. The reason why I say I think it depends on countries in the us we have a very strong hyperscaler entity and we've tended to see the operators even in edge partner with their hyperscalers like Verizon, AWS example, outposts fairly public. I think in other countries where you don't have either a strong hyperscaler or the hyperscaler is not a national, certainly when it comes to national data, they're going to look to the telco who's very well placed in expertise in heavy and familiar with running data centers and obviously has the network to provide that role. And I think their operators to see this will step up and take that opportunity.

Sean McManus, TelecomTV (05:44):
Great, thank you. Well, Julien, tell us the view from Orange businesses point of view.

Julien Sicart, Orange (05:49):
Yeah, absolutely. I fully agree and we are actually one of the player actually moving with this AI factory investments. We are selling demands on sovereignty for AI use cases, not only across the public sector but also on verticals like finance, health and defense. To give you a very concrete example, we are actually currently supporting a very large French bank in transforming the legacy application from a very old programming language to something a little bit more modern. And obviously when you deal with what is at the core of your finance activity, those legacy application, it's absolutely critical to run both the training but also the inference for LLM in a very sovereign environment. And as it was said, actually I strongly believe that telco, we are very legitimate in that space. Obviously we are used to build and operate digital infrastructure, especially on network, but not only on cloud. We've done that for many years and sovereignty is definitely a key part of our play across all our product and services. And at Orange we have this European passport, which is something we are pretty proud of, especially those days. It turns out to be very useful and we're complimenting those infrastructure solutions with services which actually help to support the customers in the AR transformation journey.

Sean McManus, TelecomTV (07:07):
Brilliant, thank you. Mark, tell us what's your view on this?

Mark Gibson, ConnectiviTree (07:12):
So I mean connectivity, we're still at the very start of our journey towards having a fully fledged network, but I think the thing that we see in particular is the need to be able to securely and accurately move learning data sets between data centers. And so having, that's exactly where telcos excel, that's not in particularly the wheelhouse of the cloud providers. It is really much more of a telco thing. So I think that's where we're seeing the conversations start to go is that ability to really securely and accurately move models, move large language models between places and do it really super securely and super accurately. And again, I can't remember the number of times I've heard that if you get some part of your dataset incorrectly transferred, then you are wasting your time trying to use it to learn. So I think that's where we see the main kind of driver and the main potential for us, us to sell wholesale backbone across Europe. I think that's where we see it and then being able to do that within locations and regions. So I think definitely it's something that we're expecting to see more of in the future.

Sean McManus, TelecomTV (08:28):
Brilliant. Thank you all. So it seems that offering sovereign AI services is potentially a money spinner for telcos. It could be a way for them to attract new customers and boost their stickiness with existing enterprise and government clients. Now, much of the chat about AI in the telcom world concerns how it can improve efficiencies and help operators save money. But what about its impact on product development? And the Telco top line? In our survey we asked, does AI offer telcos a sizable new potential revenue stream? As the chart shows, the majority of our survey respondents are optimistic, but only just with 52% believing that AI can deliver sizable incremental sales for service providers. However, a hefty minority, 28% aren't sure either way, and 20% believe AI won't boost telco revenues much, if at all. So let me ask our panel now, do our respondents have the right to be optimistic? Julien, let's start with you this time.

Julien Sicart, Orange (09:31):
Well, I have to say in the market with several hundreds of billions of dollars with a CAGR of 30%, I think we all have very good reasons to be optimistic, right? At Orange, we strongly believe that this growth will be driven by the B2B domains, and we have actually structured our strategy around three layers. The first one is really infrastructure, and we've talked about it with those AI factories that will drive obviously significant revenues. The second layer is really the platform layer where we're providing pass and AI as a service type of solutions. Typically, we tend to partner with startups around the AI ecosystems to expose this capability to customers. For example, we've recently launched a partnership with a company called Mistral to address developer needs, and we're integrating the LLMs, which is called cuts in our infrastructure to run this in a sovereign environment.

(10:25):
The third layer is a solution layer, and we've launched a solution called live Intelligence for our enterprise customers. So maybe let me spend few seconds on this one. We are providing a multi LLM conversational experience, and this solution started with a very large deployment at scale in the OR end group that we've run over the past three years. We've now reached more than 70,000 active monthly users. And this all initiative started with a phenomenon we call shadow ai, where typically 70% of the employees are using public than AI tools without the company approval, which obviously put their data at risk. And we've been developing that solution and we are now bringing it to a complete new level with a platform and ecosystem approach where we'll be enabling agent and assistance on top of this platform. And what we're doing is combining those three layers with services.

(11:18):
Again, this is super key. It starts with consulting. Typically we've seen that many customer needs help in identifying where the value is going to be created. Typically 95% of the CIO believe in gen ai, but only 51% of the company are able to find use cases which will bring high value to 'em. So this consulting arm we have is absolutely key in helping customer to transform. We also have training and change management capability because it's great to have a good business idea, but adopting it across the company is also absolutely a game changer in the way it'll drive value. And lastly, obviously the implementation, be it data governance or integration because there is no single standalone AI solution. So it has to be integrated with the customer IT systems. And typically what we see is for one euro of solutions, we're generating two euro of services. So typically this combination of services and the three layers I've talked about is actually key to unleash the potential of AI forgo.

Sean McManus, TelecomTV (12:17):
Brilliant, thank you. It's clear from that that orange has taken a very holistic view of the opportunity here and making investments showing a high degree of confidence. Mark, tell us where you are coming from on this.

Mark Gibson, ConnectiviTree (12:28):
So I think connectivity, we want to make use of the AI techniques to really drive our efficiency and to make us better able to offer homogenous service experience to our wholesale buyers and sellers. So our business model is predicated on being the single place that you buy a service from, and then we then buy sellers services from downstream Alliance partners and sell them to Upstream Alliance partners. So we want to make that whole experience as normalized as we can, and we see things like large language models starting to give us an ability to normalize the response to issues that are experienced across any network segment and to be able to help drive that ability to troubleshoot quickly and to troubleshoot accurately and to give a tiny response so that we can then offer services across the entire conductivity alliance, which are ideally better, more sharply priced, and just give a better end user experience. So I think that's where we see it, and there's been some interesting developments in the space. I think meth last year started to look at how you develop large language models and they've gone down a couple of different routes. So I think that's where we see the initial part of being a consumer of AI so that we can then offer better services to our end customers. So I think we are less in orange's space and much more trying to be a consumer to drive better, more interesting, more flexible products into the marketplace.

Sean McManus, TelecomTV (14:13):
Yeah, that's very interesting. So I suppose whereas Orange is exposing AI externally to the customers, you're using it more internally to compete more effectively. Robert, tell us a bit about your view on the opportunity here.

Robert Curran, Appledore Research (14:27):
Yeah, I think if I'm honest, we're a little bit more cautious about it. I think AI is already everywhere from the devices and bettering devices all the way through into data centers. Going back to the question that you asked to respondents, can telcos generate significant additional revenues through ai? I think there's three parts to that. First is the using AI to sell more of what telcos already do, and that's part of where the initial interest in AI is. So greater personalization, more targeted offers, all that kind of stuff that's certainly happening already. Will it be major contribution to revenue? I'm not sure it's going to be major, but it'll be incremental. I think the second question is will AI be the thing that telcos themselves provide to their customers? I think as the previous response already indicated, that's more likely to be done in partnership with providers of core ai.

(15:24):
So that's still again going to be a revenue share type model of some kind. I think the third thing is can telcos make money by using AI to add value to what they're already doing? And I think there is strong potential there, particularly in areas of security and cybersecurity because that's something that it's clear people are quite concerned about. It's the top priority for telcos and telcos can build on that trust they have and the resilience and the trust that customers put in them to be that channel for that kind of technology, that kind of capability. And that's a problem that is endlessly growing, endlessly expanding. And again, in a world where the bad actors are using AI against us, then we have to be, every enterprise has to be up to the minute in terms of response. So I think there's upside here. Yes, I think just we'd be a little more cautious on saying that it was going to be a major contributor to revenue, but we're not disputing the amount of investment that's going in to building AI related infrastructure. Let's see how that pans out.

Sean McManus, TelecomTV (16:24):
So a more cautious outlook then on the opportunity there. Michael, tell me what's your view?

Michael Clegg, Supermicro (16:32):
Yeah, I'm not too surprised that the charts indicating that people are 50% around or skeptical or questioning. I think this is really just a reflection of our immature AI is despite all the noise we've had, and it's really only been over the last two years when chat GT and l LM sort of exploded on the scene that AI really became known. Telcos had been doing AI for a couple of decades already more in the machine language space. But I think Julien articulated it incredibly well a little bit the journey. And it was really interesting for me to hear him say that they investing in the infrastructure, but then they're also investing in startups and training and consulting, which again, I think is a reflection of how early the market is. But it's really a good way for Orange to start to generate some revenue early and of course learn about the market in the process.

(17:20):
And AI telcos being enterprises themselves are going to consume an enormous amount of AI internally and are going to have a lot of AI expertise built up. And their ability to take that in-house knowledge and expertise and infrastructure package it up and offer it to their customers, I think is a very logical thing to do. So I would agree that we don't quite know what the end game looks like a little bit, but definitely the opportunity is there for those who want to grab it and telcos are going to be very well positioned because of their internal expertise to do that. I would say one thing though, for me, again, it's a little reflection of the immaturity today. We talk about ai, but to me AI is a lot more analogous to the internet. I feel like we in 1999 now, year 2000 when the browser and the internet came along and at the end of the day we won't consume AI directly.

(18:11):
AI will become embedded in everything we do. Like today we're running this thing over the internet, but we're not thinking that this is an internet application. We're thinking this is a panel chat application. So you'll start to see AI will slowly become more and more embedded and you will consume products and applications that are AI enabled rather than going out and consuming AI itself. And that's really where I think the telcos have to the stage two once we get past this initial phase is map their expertise network and where is their best fit into how is it actually going to be embedded and delivered and what are those products that are going to be taken. I think historically telcos have done well in the first phase but have sort of lost out on the second phase that when the technology gets embedded, and that's where the opportunity for them is to this time is stay relevant and gain some of the downstream revenue stream.

Sean McManus, TelecomTV (19:01):
Thank you very much. It is interesting that you bring up the maturity of AI there because it seems like it's all moving so fast that it is also maturing very quickly. Let's come to our next question. Now, among the many architectural considerations that telcos will need to consider is whether they should invest in their own distributed AI infrastructure to help with their operations and service delivery or enablement plans. So we asked should telecom operators deploy AI infrastructure at the edge of their networks? As the chart shows, the majority of our respondents, 58% are in favor of such investments. Only 17% say it would not be a good use of telco resources, which by the way is still very constrained. 25% aren't sure whether telcos should invest in edge AI at all. So let's come to our panel now. Let me ask, how does Edge AI infrastructure rank as an investment prospect for telcos? Michael, let's come to you first on this one.

Michael Clegg, Supermicro (20:02):
I think one of the clear distinction to make in ai, we have training and we have inferencing and training is going to be large centralized data centers, and that's the AI factory. But inferencing, you tend to want to move a little closer to the user and it's going to be a much more distributed and fragmented application. The telcom networks are inherently distributed anyway. Telco networks consist of typically many thousands of central offices that are spread around the country and those central offices are sort of being consolidated, but even so announcement by BT recently, they want to shut down 6,000 remote central offices, but they're still going to end up with a thousand hubs. So telecom networks through the nature of communications are distributed. The question is, do you put more intelligence in that distribution? I believe the real value for telcos will come when they find an application that combines connectivity together with AI processing together with a distributed application.

(20:58):
And I believe there are several things like smart cities and particularly autonomous driving, which is now really starting to take off. I think the opportunity for telcos is when they bridging two or three systems together and you want that inference done locally at the edge. So for example, if you integrate in a smart cities traffic controls with an autonomous driving application and the two start to go together, so now you're managing your traffic. So I think there are many instances where there's this combination and the telco then is the best one put together to bridge the two applications because they own the connectivity of the network, they already carrying the traffic for both of them and then doing that out at the edge, depending on the application, has benefits of latency as benefits of redundancy. If one node goes down, you can just shift the traffic to a different node, particularly as these things become mission critical. So I would say given that telecon Connect are inherently distributed and given that we will see one way or the other AI applications materialize, these two do go together. And over time the question for them is when is the right time to make that investment out at the edge?

Sean McManus, TelecomTV (22:05):
Interesting. It seems like it's building on the inherent strength of the telco network and that distribution to really deliver those low latency applications. Robert, tell us what's your view?

Robert Curran, Appledore Research (22:16):
Yeah, I agree with some of what Michael said. I think the destination here is in quite specific called niche applications, but they're specific applications. I think to answer the question as framed, you asked, does AI edge infrastructure rank highly? I think short answer that is no, it doesn't rank very highly today. I think there's a problem now in considering large scale infrastructure because of the whole 5G experience. I think if you look at this from an investment point of view, I think CFOs are pretty cautious about committing to a strategic investment for Edge AI in the light of still trying to realize investment from 5G. So we're already still part way through in the most recent wave, and now there's another potential one ahead of us. We've seen this change the shift from fully decentralized to centralized, and now we're trying to test the edges of that.

(23:14):
Again, at the end of the day, the real question is what's the right intersection point between, as you say, the strengths of a telco and the end user application, which might be a whole set of devices, not necessarily human interactions. So I think there's a lot more interest, a lot more focus on enterprises in general, which means lots of different industry verticals and their potential needs. And where do you put together the, as Michael says, the need for inferencing with some of the resilience and so on that telcos can provide? It's still pretty early. It's still pretty early. We're still getting used to the idea of even things like cloud native networks, let alone AI enabled cloud native networks. And also I think at the same time there's been quite a lot of machine learning creeping into how mobile networks are run as well. So there's got multiple threads going on here, multiple influences on how the architecture develops. So I think over time, yes, we can see the possibility for the need for greater inferencing in AI at the edge or close to the edge, but for today it's not seeing it as a priority for most telcos just yet.

Sean McManus, TelecomTV (24:23):
Yet. That's great. Thank you very much. It's interesting that you touched on the diversity of industries that they will need to be catering for because I think that's going to prove to be quite a challenge in itself. Mark, let me bring you in there. What's your view of AI at the edge?

Mark Gibson, ConnectiviTree (24:38):
So the edge of my networks looks like a series of NNI interfaces to other service providers. And so my edge needs slightly different and we're still quite early on in our deployment life cycle. So for me at the moment, my primary concern is to get a normalized data set that I can start to apply learning to and to think about that data set in a way where it can be, I can think about how do I build pipelines, how do I build data sources and data sync so that when I'm able to start learning, I can actually apply those learned rules back into the data and therefore use this and looking at the original question to then offer better services and to solve my operational needs. So we are really going back to basics when we approach our OSS design here and thinking about how do we normalize the data within our overall system rather than thinking about how do I normalize it within say an inventory or within a service orchestrator.

(25:40):
So we are looking to go long-term towards a data mesh architecture where rather than the data that sits between the systems being just the back and forth passing between the systems, we actually normalize and homogenize those kind of Kafka topic data sets. And that then gives us a set of congruent data that we can then start to use to learn about how our overall system is working and then start to then ideally optimize our service and optimize our operations away from that. So we're really still at that very early stage, but we are being very conscious to design our systems in a way that we will generate data which will be useful to help us generate the learning pieces. So that's really our approach at the moment is thinking how do we capture the data in a way that we don't have to massively manipulate it to make the things in different data sets match up, we match up inherently, we don't sort of post-process it to match it up. So I think that's how we're approaching this edge, but I take that our edge is slightly different to a more kind of user-driven edge, so we have a slightly different problem set.

Sean McManus, TelecomTV (26:59):
That's great. Thank you. Michael, let me bring you in there. Tell me what you want to add.

Michael Clegg, Supermicro (27:06):
Yeah, I just wanted to go back and pick up on something that Robert was saying and to me it's fairly critical in terms of telco evolution and having been involved in 5G for some time and super micro being essentially a bare metal server manufacturer. So for us, we play, we are more relevant in the network when it gets virtualized and we were talking about cloud native networks and I think one of the big challenges with 5G has been there's an enormous cultural shift that has had to take place in telcos that we've seen them struggle with. They've gone from deploying appliance based networks to really deploying virtualized networks, these cloud native networks essentially they've gone from being network engineers to computer engineers where everything is now virtualized, running our computers and delivered it as an application. And telcos sort of have to finish this migration rarely to get 5G fully deployed and fully effectively, especially as they go to 5G advanced.

(27:57):
But the outcome of this is a few years down the road essentially Tel networks going to be big distributor data centers one way or the other. I mean inherently the network itself is going to be a big distributor data center. At that point you start asking the question, I already have all this compute infrastructure. This compute infrastructure is not necessarily fully utilized all the time just because of the characteristics of the network and the nature of compute is that in many of the workloads, and we've seen this today, many of the workloads that I'm running today, the cost suite point of computers gives me more compute power than I need to run that workload. And I think at that point the light bulb's going to go off, and I'm sure it's with many telcos already because they're smart people, is we've got this huge compute infrastructure that's distributed, that needs to be distributed, that is now available to do other things. And one of those other things for churchly could be ai. So I do think that telcos need to get through this transition to really understand how to build and run cloud native networks to get their internal staff switched over. But once they've reached that stage, they'll have this distributed computing infrastructure sort of by default.

Sean McManus, TelecomTV (29:02):
That's great. Thank you. I love the idea of the network as a distributed data center. I guess it's clear that they've got, as you said, they're on a journey to get there yet. Julien, tell us a bit about where Orange is with regards to AI at the edge.

Julien Sicart, Orange (29:16):
So maybe I'll comment a little bit on the enterprise domain and maybe to compliment with a few example to what Michael was taking. We're clearly seeing a strong move in the industry from training to inference as it was said. And of course that will generate new use cases and there are few reasons to actually get excited and interested in the ads compute, right? There are mainly three according to me. The first one is the most obvious one is a latency for very real time type of use cases. Being able to run your AI workloads close to where data is generated is obviously a lot of value, but it's not only about speed, there's also a case about control. Some of the industry are actually regulated where they can run these type of use cases and some use cases will have to be run either very close to the in-country or even within the customer premise.

(30:17):
So there's a case there for Edge and there's one that was also mentioned on resiliency, right? And here I will take the manufacturing domains as an example, is a smart industry where any outage in the manufacturing domains will lead to a prediction line being stopped and that obviously impact directly the top line with millions at stake for each hours of downtime, right? So the more your AI is integrated to the core of your industrial processes, the more sensitive these type of things are and you want to make sure that your factory will continue to operate even if it's isolated. And as telco, we are very relevant, it was said because we are able to provide this type of solution combined with end-to-end solutions. Obviously security was mentioned that's absolutely critical, but connectivity such as industrial LAN or mobile private network, they compliment very well these type of use cases for B2B customers.

Sean McManus, TelecomTV (31:12):
Brilliant, thank you. And of course telcos have a lot of experience as well with providing very highly reliable technology. So that will obviously be critical for their customers in this. Now there's one more key finding from the report. I'd like to look at one of the more high profile use cases for distributed AI infrastructure in telco networks is AI ran. So we delved further into the AI edge topic by asking is AI ran where the same compute stack is used for AI and mobile networking workloads, a network architecture option that mobile operators should adopt as soon as possible? As we can see in this chart, our respondents are a little more cautious but still positive overall about this infrastructure option. Only 24% believe network operators should be investing in AI ran systems right now, however, 52% think mobile operators should consider this option in a few years time when the dust has settled on the best way to virtualize and cloudify the radio access network. Only 8% of our respondents don't think AI ran is an optimal architecture for mobile operators. So what do our specialists think of these results? Investments in the radio access network are already being squeezed. Do operators have the means and the will to invest in AI ran? Are the benefits of shared workloads, greater network capacity efficiencies and possible revenue generation compelling enough for telcos? Robert, let me come to you first on this. What's your view?

Robert Curran, Appledore Research (32:47):
Yeah, look, when we talk about a ran, we generally referred three distinct strands. We talk about AI in the ran, ai on the RAN and AI for the ran. Those are the three for working groups within AI Ran Alliance. All of that is still pretty new to the industry. So as I mentioned before, machine learning has already been a feature of ran management for at least five years, probably 10 if you go back into r and d. I think as regards the Iran idea, we've seen some strong capabilities, some demonstrators and so on shown off using this kind of infrastructure. I think it's not yet clear that telcos are ready to commit just yet. We're still very much in the network infrastructure life cycle of the typical seven to 10 year period. And a bit like open RAN in general, is AR ran compelling enough on its own to change the trajectory of investment cycle?

(33:46):
Not sure we're there yet, which I think your survey kind of reflects that to some degree in the future maybe is it something we need to do right now a bit more of a cautious response to that? I think we're still getting used to the ideas of virtualized RAN and cloudified ran. So a RAN is still something that's a little bit in the future. At the end of the day, I think when it comes to ran and ran investments, what targets really want to see is that there's a measurable impact for customers that are a benefit or some clear benefit to operations. I think the inversion of RAN the AI RAN implies the idea that sell sites coming from revenue generating centers is really a pretty radically new idea. So I'm not surprised to see a degree of caution in there that they'll be generating revenue from spare capacity.

(34:31):
It's just not something that telcos are used to thinking of. So this is still pretty new. The other question is whether or not there's enough aggregated demand to warrant the degree of scale investment and automation required to do this. There might be, we're not sure, we haven't seen it yet. There's not an obvious standout case. There are examples of specific instances here and there, but this one's going to take a little bit longer to see if it really tick off and becomes the preferred structure. That's not to say there won't be examples of using the Iran architecture where it makes sense and whether that's dense urban environments or particular industries. We may well see those things, but is it the right strategic architecture? The jury is still out I think at this point.

Sean McManus, TelecomTV (35:17):
Thank you. Yes. I think showing those benefits is really going to be key to getting operators on board with that and making start to act sooner. Michael, lemme bring you in there. What's your view from Super Micro?

Michael Clegg, Supermicro (35:31):
Yes, for me there was two part question there. Should we do something right now? And that more skepticism to do AI and ran right now obviously implies that there's a significant and decent amount of AI applications that are ready to be monetized. And I agree with Rob, but where we are today, it's a little premature. We're still in the early phase of that, although some operators of leading the charge like SoftBank. So we will see always there's a pioneer that shows the way for people, but probably for the industry as a whole is a little bit wait and see. Coupled with that, of course these operators have just made a huge investment in deploying their current RAN infrastructure and we see that in open ran and ran sales in general from the major providers. So there's a little bit of a lull at the moment and telcos are famous for sweating their assets.

(36:19):
So they're going to go run out and put a whole bunch of new infrastructure in place when they've barely fully utilized what they've just gone and spent on probably not. So I think we will see some pioneering work happening but not mainstream. But going forward, I think this becomes a much more logical thing to do. As we said, basically Van is the ultimate distributed workload in application and as the van slowly gets virtualized and just becomes a software application on standardized compute, now you've got this compute infrastructure and as I said before, the way silicon works, there's always a sweet spot and you buy a PC today, 16 gigabyte of RAM is the optimum price point. Tomorrow it's 32 gig. That's just the way semiconductors work. There's a point that reach, you get a certain performance that is the optimum from them to produce from a silicon point of view.

(37:04):
And then really whether you use all that performance or not is up to you. If you saw Intel recently saying we've gone from under distributed DU cell site where we needed multiple servers to run a cell site. We can now do that all on one server. Well, the next iteration of silicon implies you can do multiple cell sites in one server. And that ties back to something I said earlier with the transition towards fiber networks. As you get more and more fiber networks, you can pull some of these edge nodes back and pull them. So you get this idea of DU Pools and that's where centralized van comes in, where you have a hub site that is serving multiple cell sites. Now, once you've got that hub site, you can start to pull some of your capacity and that's much easier to generate excess capacity in that site as well, particularly again as the silicon comes along.

(37:51):
So I think we will naturally see a situation in a couple of years where the available compute at a C site is pretty significant and will be available to do additional workloads. One of the big argument, and just think for yourself in the middle of the night, the 5G network isn't carrying a lot of traffic. Everybody's sleeping. So all that compute infrastructure, if it was general purpose compute, could now be repurposed and run a different workload. And these are the type of workloads we are speaking about. So I think inherently as networks become more computer based, they become more virtualized. People will seek to continually maximize the utilization of that network and AI will be one of those workloads that they can run at the edge.

Sean McManus, TelecomTV (38:33):
Brilliant, thank you. Well, at TV we very much look forward to following the work of those pioneers you mentioned just there. Now we must leave it there. Unfortunately. Michael, Mark, Julien, and Robert, thank you so much for taking part in our discussion today. If you haven't already done so, please download the DSP Leaders report. It's available to all registered viewers of telecom tv. Registration is free, so sign up now and you cannot only get our AI report, but you will have access to all of our DSP leaders content. And if you're new to telecom tv, please take a look at our dedicated telcos and AI channel where you'll find the latest news and analysis and videos including all of the discussions from our AI Native Telco Summit series, plus the recent telcos and AI event. For now though, thank you for watching our DSP Leaders Reports panel on AI and goodbye.