Ray Le Maistre, TelecomTV (00:05):
So TelecomTV recently published its Telcos and AI report, one of the many series of DSP Leaders reports we've published in the past couple of years. And today here at the DSP Leaders World Forum in Windsor, I have a distinguished panel of speakers to talk through some of the results from the survey that led to the report. So welcome, gentlemen. I know who you are, but I'm going to get you to introduce yourselves before we get into the questions. So Diego, if we start with you at the end.

Diego R. Lopez, Telefónica (00:36):
Good. I'm Diego Lopez. I work for Telefónica and recently I was appointed as responsible for data and enablers in the innovation branch.

Ray Le Maistre, TelecomTV (00:46):
Okay, great. Thank you for joining the panel. Warren.

Warren Bayek, WInd River (00:49):
All right. Thank you. Warren Bayek. I'm a member of the office of the CTO at Wind River overseeing strategy and futures for intelligent edge and AI.

Ray Le Maistre, TelecomTV (00:57):
Okay. And Robert.

Robert Curran, Appledore Research (00:58):
Hi, I'm Robert Curran, consulting analyst with Appledore Research. We publish research on the development of telco in the era of cloud and AI and edge.

Ray Le Maistre, TelecomTV (01:07):
And edge. Absolutely. Okay, fantastic. So the Telcos and AI report was published a couple of months ago based on a survey conducted in February and March. We got 260 respondents from the industry, so a good spread of people and about a quarter of those work at telecom operators. Now there was a lot of positivity from the respondents about the potential that AI offers to telecom network operators, but is that optimism warranted? This is what we want to talk about here on this panel today by looking at some of the survey results. So a lot of the focus on the use of AI in the industry is related to efficiency, shortening and automating processes, improving productivity. And there's also a lot of excitement around Agentic AI. So as part of our survey, we asked, in your opinion, is Agentic AI mature enough for deployment in telecom networks?

(02:10):
And although our respondents in general are optimistic about AI, on this particular topic, the majority don't currently think that Agentic AI is mature enough for deployment in production telecom networks. In fact, 57.5% of them said the complexities of telecom networks have not yet been properly baked into Agentic AI developments, and only 30% of the respondents believe that the use and understanding of the Model Context Protocol and the agent-to-agent or A2A protocols is a game changer for telcos currently. So panelists, in general, are we seeing a positive impact from AI on telco process automation and what are your views about the current applicability of Agentic AI in production telco networks? Diego, I'm going to start with you. So in general, and then Agentic AI.

Diego R. Lopez, Telefónica (03:12):
In general, certainly, yes. I mean, there have been experiences in applying AI in many fields. And we're not talking about AI, I'm not talking only about LLMs, talking about machine learning, we're talking about mechanisms for ... It went totally out of my ... For detecting anomalies, anomaly detection, things like that. Yeah.

(03:36):
We are applying this, it's being applied. It's in production. It's been productive and it's really useful. When it comes to even using LLMs with the conversational interfaces to users, et cetera, as well. With Agentic, the problem, I believe the main problem apart from the maturity of the protocols—because the protocols were formulated probably a couple of years ago at most—is about the fact that right now they are being applied in very closed environments. It's about "I am provider X, I have my agents, I run my agents on my tools" and so, well, that's perfectly fine for a certain kind of task, but for dealing with a complex, heterogeneous, critical infrastructure like we have in the case of telcos, I think that it's not only about immaturity, it's about the lack of openness of those mechanisms that are very much focused on particular providers.

Ray Le Maistre, TelecomTV (04:30):
Okay. Warren, what about you? What do you see overall in general, AI having that positive impact?

Warren Bayek, WInd River (04:37):
Yeah, I'd agree that we're already seeing measurable impact from AI in certain parts of telco. As you mentioned, customer interaction obviously is a huge one. We see it a lot in the operational side. Telcos have been automating for many, many years as we know. AI layered on top of the automation has created a tangible improvement in their ability to perform operations at scale. Anomaly detection for sure, predictive maintenance, things like that, AI has definitely enhanced in measurable ways. I'd agree that Agentic AI is early in terms of current usage in the telco space. And I agree with the point Diego made. I also think a big piece that is yet to be tackled is the operational trust of Agentic AI. I've been in telco for many, many years and I've heard the closed-loop automation promise since the very beginning, literally 40 years ago. But in certain parts of the telco world, that's just, to me, it's something that's not quite there and maybe never will be.

(05:46):
Some of these services we're delivering are frankly too mission-critical currently to fully trust them to Agentic AI yet. Now, will that change in the next year, two, three, four years? Probably. But right now, I think the operational trust just isn't high enough for telcos to fully embrace Agentic AI in terms of their rollout of features and their infrastructure.

Ray Le Maistre, TelecomTV (06:07):
Okay, great. And Robert, I mean, you're at Appledore now, but previously you worked in the software sector in telecom and obviously over the years the promise of automation and indeed the use of AI over the years has been discussed around. Where are we now? Have things moved on noticeably and what's your view on Agentic AI?

Robert Curran, Appledore Research (06:34):
Yeah, I think Warren and Diego have said, we've seen results from AI in general, the machine learning kind of era that we've been in for the past 10 years or so. That's all fine and dandy, as it were, but we're at a kind of important transition. The sorts of technology and capabilities we're seeing with Agentic have really taken us into some different problem spaces. So the first base for that is looking at the use of agents within specific technology domains, things that we understand already pretty well. So automating, enhancing the automation that we already have to some degree. I think the next challenge is where we'll see the real benefits in the long run, which is that kind of cross-domain, end-to-end service application of Agentic AI to do things that are really very complex and difficult. I think, as one point aside, the question about trust and responsibility, it's not just a steady continuum.

(07:24):
There's a real step-change as you go from managing things that you understand pretty well and processes you understand well into saying, "Well, here's a process that's never even been defined. I'm going to let Agentic AI work out what the process is and then let it regulate itself." I mean, that's an enormous step up. So we're just at the start of that journey. I think it's quite tentative for operators to look at that, but they are definitely looking at it. The potential is enormous. I think there's no question about that. The degree of productivity enhancement that is on offer is potentially absolutely game-changing for telcos, but it's going to take quite a while. I think the question about protocols is critical in certain areas, but in time that will also become just an embedded part of whatever is taken for granted. I think the real question is where are the pain points for operators?

(08:16):
And looking at that from that business perspective, again, we've heard a lot of that in conversations today already. It's not about technology. It's about understanding how the business works. And I think if you look at how Agentic AI in general is talked about in other contexts, it's good old-fashioned systems analysis people come back to, which is: what business problem are we solving? What can we expect? What's the payoff if we actually address that? So early stage for sure, but we are starting to see some benefit there.

Ray Le Maistre, TelecomTV (08:40):
Yep. And we're even seeing, I mean, somebody always has to be first. So we are seeing some early deployments. And in fact, we heard about one of them earlier on today,Ahmed at Deutsche Telekom, first steps, but somebody's always done something first and then others follow. But it's good to see there's some momentum there. Now, like I mentioned earlier, a lot of discussion around AI and telecom is around efficiencies, but it's not all about efficiency and automation. The telecom operators need to generate new revenues and find new revenue streams as well. So we asked our survey takers, does AI offer telcos a sizable new potential revenue stream? And I came up with the questions, so I know why I put "sizable" in there because you could treat "sizable" how you wish, but we're talking about something meaningful, something that might move the needle on the top line.

(09:47):
So the responses there, the answer was positive in general, yes, according to our respondents, again, it was 57.5% of the 260 or so said yes, they thought that there was a real opportunity for AI to move the needle on revenues and only 17.5% of the respondents saying they didn't think that AI was going to impact revenue levels very much, if at all. So are our respondents—is there a reason to be optimistic here? How do you feel about the prospect of telcos developing sizable new revenue streams from AI-related products and services? So Robert, we'll start with you this time, because I'm sure you are asked by your clients to look at this as well, to maybe find some opportunities that they are looking for.

Robert Curran, Appledore Research (10:41):
Sure. I think I don't want to spike the question by referring back to 5G. The same kind of question asked of 5G 10 years ago, I'm pretty sure it would have resulted in the same kind of answer. So we won't go there. We are in an optimistic industry. And of course, we're looking at how technology can enable new potential revenue streams. The question of sizeability... I think one of the aspects of this is segmentation. I think there's an opportunity which is immediate AI reselling, which we've looked at telcos getting into that business. So everything from hosted AI to AI in security and so on. And then I think there are, for enterprise, there are different areas where AI-enabled products might add new revenue streams. I think one of the other things to consider here is whether you're going to actually grow revenue or just avoid losing revenue.

(11:39):
That can be a benefit in itself. I think as more AI becomes more embedded, even in things like voice channels, for example, what you may see is AI just become a standard feature of certain products. Before you think about growing your customer base, let's stop the losses that are happening already. So in some areas, churn is a major factor. So we might see AI—it'll play the role as you're describing, but will it be a sizable revenue stream? Not too sure yet, but I think it really depends on segmentation, where you're looking in the market, where is the need? And there may well be lots of opportunities. The question about data, I know we're going to talk about some edge applications in a different context, there are opportunities there. Sizable? I think the audience has been a little bit over-optimistic on that in terms of evidence today, but one could be optimistic.

Ray Le Maistre, TelecomTV (12:25):
Yeah. No, and that's a great point you make because if there are sizable new revenue streams, they might come as other revenue streams fall away or are, in fact, replaced. Diego, what's your sense here? How might this develop in terms of revenues that might come directly from AI services or AI-enabled services?

Diego R. Lopez, Telefónica (12:53):
That's something that I believe is not going to happen in the coming couple of years, but frankly, because to make it sizable, you need a certain scale.

(13:05):
I do believe that, and this is something that I mentioned this morning as well, that there is a good opportunity for becoming ... I have started to use the term substrate, a common substrate for using AI applications because I don't see telcos selling AI applications themselves or AI models or whatever, but I do believe that there is something that is about moving data, making data available, making a trustworthy source of decisions, et cetera, is something that could have an opportunity the same way, or very much connected with things that happened with IoT. Telcos are not manufacturing IoT services or the devices, they are not providing direct IoT services, but they're providing the infrastructure for this, perhaps as an additional source of revenue. I see something similar and probably even going a little bit beyond, because we have the experience and we have the already-in-place agreements, experience with the regulatory frameworks.

(14:13):
We don't have the regulatory frameworks fully, but we have the experience with handling regulatory frameworks. This is a very delicate issue that looks like it will be regulated and that implies that, well, our experience with that would help. And we're in the position of establishing this kind of trustworthy, trust links with data providers, data consumers, inference providers, and build an ecosystem there. And the example, something that happens, and the example of the CAMARA initiative about exposing certain data, et cetera, is an example. I would not say that this is the solution. Currently, what we have right now is something that is usable for other purposes, but I think it's a first step showing how we can progress in that direction and build an ecosystem that will be beneficial for everyone and especially for telcos.

Ray Le Maistre, TelecomTV (15:13):
And of course, we've had some conversations earlier today about how things might be possible, but you need the foundations to make these things possible. So Warren, how much does this conversation about telco ability to generate sizable AI revenues—how much is that linked to where they are on their telco cloud and cloud-native journey?

Warren Bayek, WInd River (15:44):
Oh, I think it's fundamental that they're in lockstep. And I think Diego hit it, that the telcos have built—and I mentioned it earlier on a panel this morning—this incredible real estate empire. And to your point, telcos are used to behaving in a highly regulated, highly secure environment, highly distributed. All of those factors are a major play for their ability to provide an AI-enabled infrastructure, which people will trust. And it's an ecosystem that, frankly, they're never going to be LLM providers. They're not going to compete with hyperscalers or Anthropic or that sort of thing, but they will be able to provide an infrastructure and an ecosystem which, if they open it to the development community, these services can be developed and deployed on their system. And the data they have is another goldmine that, managed properly, could help them with some of these sizable revenue numbers.

(16:46):
And in fact, I'm really encouraged to see those numbers. I was frankly surprised when I saw this report at how high that percentage was in the telco world, because frankly, when I sit with telcos, they're pretty pessimistic on their ability to monetize this coming AI revolution. And we try to explain to them sort of the things you mentioned, the thing I mentioned here, that Robert mentioned earlier, their play in this is a little different, I think, than what most people simplistically think of as AI revenue generation. So they're going to have to be a little creative, maybe create some partnerships to capture some of the coming AI money that's there. And in fact, if you listen to Jensen at NVIDIA for more than 13 seconds, all he talks about is physical and edge AI. So telcos are primed and in a prime position to be a major player in that space.

Ray Le Maistre, TelecomTV (17:36):
And this is such a fast-moving topic that we asked these questions in late February, early March. If we asked it now, like as of today, I think the response would be much higher because now we're starting to hear about in China the operators starting to sell tokens and use that as a billable thing rather than voice or data, et cetera. So it depends how you interpret it as well, but of course if you're selling that, that might be the package and not so much voice or data. So it's going to be an interesting dynamic, a challenge for the marketing teams as well as the technology and other teams as well. Now you touched on something, trust, really, really important, and there's a big link between that and AI and sovereign services. And there's a lot of talk these days around sovereign services and that's clearly driving regional and local investments in AI and infrastructure around the world.

(18:46):
So we asked our respondents, should telcos develop their own sovereign AI factory to offer AI services to the enterprise sector, including government customers in their domestic markets? Again, we got a very positive, very bullish response. Our respondents are glass-half-full types. 57% believe that AI services delivered from a telco's own AI factory represents a strong business opportunity. And we asked that question actually a year earlier and it was already 54% then. So people have been listening to the AI factory talk from Nvidia as well, but also seeing some of the early examples, but it's gone up a little bit there. So I guess the question here is, should telcos invest in their own AI factories and does this make strategic sense, Diego?

Diego R. Lopez, Telefónica (19:42):
I'm not that much convinced if we're talking about an AI factory as big installations with thousands of GPUs consuming power plants.

Ray Le Maistre, TelecomTV (19:55):
Let's say it doesn't have to be that massive because I think a lot of people think of it that way, but I think these days, especially for telcos, an AI factory can be something with a hundred GPUs as a starting point.

Diego R. Lopez, Telefónica (20:07):
No, no, that's a very good point. And probably yes, we're talking about, let's call them edge factories or whatever we call them, smaller ones that are connected very, very close to the users, very much focused on the edge, probably in small models that are very much focused on particular tasks. And to preserve the sovereignty in different degrees, I would say yes. And there is an advantage that we have to consider that with AI, the development teams that you need are not so big. One of the reasons why telcos stopped developing their own software and one of the things was it's about the huge amount of people that you needed for doing so and then the revenues were not supporting it. Right now, at least for basic tasks or for training a model or for developing a piece of code that is needed for you, et cetera, this is something that is feasible right now with a much reduced team. And that's something that could play in favor of telcos and our limited capacity of hiring one million developers.

(21:18):
And that's something that I have the hope that this is something that will change as well, the way in which we are related in general with all the software that we run and in particular with the AI-based software.

Ray Le Maistre, TelecomTV (21:30):
Okay. Warren, are you seeing a lot of excitement around AI factories?

Warren Bayek, WInd River (21:34):
Yeah. In fact, I'm glad you made that distinction because I was going to make that my first sentence, because I agree that—the entry into the classic AI data center space, absolutely not. But in terms of building—

Warren Bayek, WInd River (21:49):
For some, maybe, especially in Asia.

Ray Le Maistre, TelecomTV (21:50):
Yeah, for some of the larger ones in certain—

Warren Bayek, WInd River (21:51):
In general, yeah. Right. Now that said, I think building some form of ... And I'm going to say again, we're not talking about just building GPU farms like data centers, right? We're talking about very specialized, very sovereign data centers. I think they absolutely have to get involved, particularly in some of the more highly regulated environments like Europe where the CRA and the GDPR come into play, having the ability to deliver sovereign AI solutions for very specific edge and telco problems I think is something that telcos absolutely have to be a part of if they're going to see even a sizable AI revenue stream. They're going to have to provide some of that because again, we're not talking about building massive LLMs that can do anything in the world. We're not in a chatbot environment, right? This is a very specialized edge, telco, physical AI space with very defined problems.

(22:52):
So the data center types they're going to have to build are much smaller and more defined. And I think telcos can absolutely have a big play in that space.

Ray Le Maistre, TelecomTV (23:01):
Okay, great. Robert, we're seeing some early examples, aren't we?

Robert Curran, Appledore Research (23:05):
We are. Look, let's not separate this from the geopolitics of the world that we're in today. The key question here is not about AI factories. It's about sovereignty. It's a sovereignty question. And that then leads you to the question of who's the customer here. And the reality is that there is going to be, to Warren's point, there's a small number of customers for whom that's absolutely essential. And we're also looking at the context where the stack has changed. This is critical national infrastructure for telecom, and now we've got a piece in the middle of that, which is AI, which is also now becoming part of the critical national infrastructure. So the question of sovereignty and resilience and control—for whom does that matter most in a way that's completely non-negotiable? And that starts with government. There are other entities as well, but it starts most obviously with government.

(23:51):
And then from there, everything else follows. So partly the answer to the question "should telcos do this?", I think they're going to be forced into acting that way. Any national government, especially in Europe but also elsewhere, will want to know: what are the risk points? Where do we have a risk? Where do we have a resilience continuity risk? And right now, there's an increasing risk point in the whole question of how AI is provided, different aspects of AI. And what the governments are certainly trying to do, and then beyond that, enterprise, is reduce that risk point—reduce the risk that an arbitrary political decision made in another country will interrupt their operations. So I don't think, again, to Diego's point, getting into the cloud business—no, no, we know what that looks like and it doesn't work very well, but it doesn't mean we shouldn't have something to offer in there.

Ray Le Maistre, TelecomTV (24:35):
Okay. And then very quickly, because edge has been mentioned a couple of times, we did ask our survey respondents about whether telecom operators should deploy AI infrastructure at the edge of their network in particular. And there's a sense in the sector that AI is giving the telco edge a reason to exist at last. And our respondents agree—more than two-thirds believe telcos should be investing in distributed AI service deployment. So very quickly as a wrap-up here, Diego, is this something that makes sense?

Diego R. Lopez, Telefónica (25:13):
Yes. To me, it's something that, if we are realistic and we size appropriately the deployment, for sure. Yeah.

Ray Le Maistre, TelecomTV (25:23):
And I expected you to say that because Telefónica is dealing with it. But with good reason and—

Diego R. Lopez, Telefónica (25:31):
No, no, definitely.

Warren Bayek, WInd River (25:32):
Yeah.

Ray Le Maistre, TelecomTV (25:34):
Okay. And Warren, do you see this as a natural step, an advantage maybe that telcos have that others don't?

Warren Bayek, WInd River (25:40):
Yeah, 100%. The fact that they have that edge space means for sure they need to enable the AI capabilities that affect the devices that require low latency, resiliency, always-on, five nines—things that just don't exist in standard AI kind of chatbot environments. So only at the edge can that be delivered, and only telcos are on the edge. So they almost by definition, they have to become involved in that.

Ray Le Maistre, TelecomTV (26:10):
Okay. Robert, does the edge have a raison d'être now with AI?

Robert Curran, Appledore Research (26:13):
Well, it's interesting. We've seeing two forces in play. We're seeing, particularly in the mobile space, the dependency on AI, an increasing desire to depend on AI just to keep the network running. And that automatically requires a build-out of some additional AI capacity at the edge. And the other thing is looking at what kind of applications might need to migrate from a centralized location to an edge. And so some of the enterprise applications are around robotics and security and so on. There's a need for some of that. I think you've got to look at the geography of a country, and it's completely different in Switzerland than it is in the USA or India. So that's also a factor. How far does the edge really need to be? In France, the delay from Paris to Marseille, I think, is pretty negligible no matter where you put things. So there are questions about it from a latency point of view.

(27:01):
That's not the only reason to do things at the edge, but it's certainly a factor.

Ray Le Maistre, TelecomTV (27:05):
Okay, excellent. Well, gentlemen, thank you very much for joining us for covering a couple of the topics. The report, it's 26 pages. There's a lot covered in it. And although things are moving very quickly, a lot of this is still very relevant. The data's still very fresh, and it's available to download from the TelecomTV website for anybody to do so. So check it out if this has been of interest to you. What we've discussed here today on the panel, please do check out the report and download it, but I'd just like to thank Diego, Warren, and Robert for joining us on the panel today. Thanks very much, everyone.

Warren Bayek, WInd River (27:45):
Thank you, Ray.