To embed our video on your website copy and paste the code below:
<iframe src="https://www.youtube.com/embed/XnpOlGgl4og?modestbranding=1&rel=0" width="970" height="546" frameborder="0" scrolling="auto" allowfullscreen></iframe>
Ray Le Maistre, TelecomTV (00:07):
Thanks everybody to come back for this afternoon's sessions. So we'll start off by taking a look at this afternoon. So we've got three more sessions for you. We've got this one on AI, R&D and sovereign. Then we've got securing the digital supply chain that's come up quite a few times already so far today. And then we're finishing off this afternoon with the cameras will be turned off and it'll be a Chatham House Rules unthinkable session on digital sovereignty covering well, whatever we decide we're going to cover in that final session, driven largely by Andrew and Dean, but we'll find that out a bit later on this afternoon. But this is the session on AI, R&D, and sovereign capabilities looking at the role of AI research, AI factories and service ecosystems in advancing sovereignty goals. So just like the other sessions, I'm going to ask the speakers to introduce themselves one by one so everybody knows who they are and the perspective they're coming from. So Atul, if we can start with you.
Atul Purohit, Nokia (01:18):
Hi everybody, my name is Atul Purohit. I work as the market CTO and head of solutions for Nokia's Cloud network services business for Europe based here in the UK. Come with a lot of background from telco's, worked in Vodafone and BT before. Happy to be here and contribute to the subject.
Diego R Lopez, Telefónica (01:33):
Thanks. Great, thank you Diego. Diego Lopez. I work for Telefonica and well I lead a team right now that is mainly focused on applied research in network infrastructures and well aspects related to the standardization of these results.
Kristian Toivo, Telecom Infra Project (01:54):
Okay, pressing the button. I'm Krristian Toivo, executive director of Telecom Infra Project or TIP as we call it. TIP is a nonprofit trade organization focused on driving open connectivity platforms and lately of course AI and I noticed pleasantly some of the previous speakers connected openness very much with the AI journey as well as the sovereignty, which I firmly believe in. Tip has about 200 member organizations, a lot of basically all the European large operators as well as some North Americans and Asian ones as well as lots of vendors and there's about close to 1000 engineers or experts working in our six project groups, different initiatives driving technology forward. So really nice to be here talking about the super exciting topic. Thank you.
Ray Le Maistre, TelecomTV (02:46):
Okay, thanks everybody. Right. As with the other sessions as well, each of our expert speakers is going to have three minutes to start off. So we're going to start with Diego, come to Atul and finish with Kristian. So Diego, your three minutes starts now.
Diego R Lopez, Telefónica (03:04):
Okay. So basically we're talking about AI and research and development firstly about the critical thing of how we apply research. I mean something that has been around in Europe, we are very good at research and not so good at the application of this or to real business. I think that there is something that has to do very much with some divorce between academia in the widest sense possible. And from time to time in inside my company, myself and my colleagues are considered part of the academia as well and how we breach this gap. We are trying to address this in Telefonica with something that is, well we are constituting what we call joints research units with several universities and research institutes, et cetera. Just to explore this. This is on the one hand, on the other hand is the importance of open interfaces and open implementations and to avoid something that is totally false, that is the dichotomy that you have to choose between standards and open source implementation.
(04:14)
The idea is that they compliment and sometime ago in ETSI presented a diagram that has been used several times. I'm very proud because this is kind of things that you see from 10 to 10 say, hey, that's mine. Which is also sort of an infinite loop between what you can achieve with open source and we can achieve with open standards. Let me just remark a couple of things. When coming to AI has been said before and let me exist when it comes to ai, the important thing, the real thing to guarantee sovereignty about the data, not the models. Models are something that are for sure they're important, they're complicated there. But think about it, it's about how you manage the data and how you change the data and how you warranty that. The data remains where it should be. And just to finish, and this is the kind of things in which my team is working right now related to this is on how we facilitate that kind of data infrastructures that support AI of whatever the flavor of ai you can think, agents, LLMs, whatever.
(05:18)
We can facilitate this in an open environment, what allows an open market for data and for the results. And second, with the dream of a multi-party secure computation, storage, communication. And for that what we have has started to work and during the break I was talking with someone about this cannot spot him right now where he's in the room about the promise of quantum technologies for that and we are exploring these results. They are quite promising for the moment we are confined to the lab but we hope to have something to show very soon. That's it. Thank you.
Ray Le Maistre, TelecomTV (05:58):
Excellent. Great. Thank you Diego. And exciting news on the quantum developments there as well. Okay, Atal, you have your three minutes starting now.
Atul Purohit, Nokia (06:09):
Alright, so yeah, I just wanted to take help of this picture to basically just convey my point about the subject and now Nokia as everybody knows is mostly a technology supplier working with telco partners to enable critical communications, enable network infrastructure across the globe. The role of AI is obviously as we understand there are models, there are enablers, there are technology building blocks, et cetera. But from our perspective, basically AI is basically a distributed AI. So we call this entire thing as a distributed AI continuum, which means that AI basically have models at the core, at the edge as well as near to where the service points are from a Nokia perspective. We basically enable if you like the infrastructure needed to build the connectivity where for example, if the governance suggests that the AI model should be broken down into a distributed pattern, then obviously the connectivity becomes important.
(07:05)
There are certain expectations from a networks perspective, which are basically not explicitly called out in this entire euphoria around the AI adoption over the last few years. Nokia plays as well as telcos play a very important role in terms of making sure that the connectivity is available. So I've got these two key call outs that I wanted to specifically touch on. So one is that from a networks perspective, from a technology perspective, there's a lot of expectation, unset expectation around the autonomous automation of AI control loops. And the idea is that AI would work, but it has to work with a lot of support from a network perspective. A fast, stable, reliable and secure control loop at every stage of the entire network economy is what is required as well as a continuous reasoning correlation and improvements near it to where the decisions are made or near to where the services are consumed is required as well. So there's a lot of work that has to go in making sure that the network fabric is in place to make it effective as well as the ai, the SOM of the AI for example, requires the data to be secure, to be available in a specific geography, to be available across distributed models within a certain geography. So all of these basically translate into this type of a distributed AI continuum and basically from our perspective basically play an important role in the entire network fabric available. So that's my
Ray Le Maistre, TelecomTV (08:27):
Okay, great, thank you. And Christian.
Kristian Toivo, Telecom Infra Project (08:30):
Okay, so in tip we work a lot with how to apply AI to make network operation and telecom operations more efficient. I'm not going to talk about that. I'm going to try to see as the task was if I understood it right, what does the telco sector and mainly the operators, what role can they play in enabling this sovereignty? And I think the previous panel before launch touched on many relevant things, so I hope I won't repeat any of that, but I started by trying to figure out what is sovereign AI and of course I use chat GT and a few other sources and there was a number of steps and for me it was very clear I'm not going to waste the time to read through those seven points, but I think it's very unlikely that hardly any nation except for a few examples could master even nationally all of those steps from national compute to data pipelines to foundation models to security, industrial capability.
(09:29)
But I think there are aspects of this which the telecom sector and the operators themselves are particularly well placed to play a role in. And we heard some examples in the previous panel in certain markets where their operators have taken that role, but be it whatever there is an opportunity and I think the short term one is really key is actually to get in there and start helping to build all that connectivity and infrastructure that is needed. I just read this morning in a report that there's expected almost 3000 new data centers to come online in the next three, four years. They all need transport, optical transport by the way, in between them there's all the aspects about the real estate and bringing in AI as part of the B2B or mission critical offerings to public sector or even defense. But I think we haven't talked about so far, maybe unless I missed that in the morning when I wasn't here, is the whole process flow.
(10:30)
There's a supply chain discussion later on. I think operators are very well placed to manage and bring those different partners together that are needed. And there's also a talent and internal governance aspect when you talk about sovereign because it will come with regulation or at least requirements for guardrails that I think our sector and telecom operators particularly are quite good in implementing. Nobody's perfect but at least it's ingrained in the DNA. And last but not least, there's a longer term perspective. Maybe there will be AI fabric, a continuum like my Nokia colleague said, or maybe there will be an edge cloud of some sorts remains to be seen and I think there's going to be quite a few things that have to happen before that is there. Maybe that's going to be the basis for six g, I don't know. But the key thing right now, operators are well placed to help both build out partner and guarantee that sovereignty as we go forward. So let's just get on with the work. That's my view.
Ray Le Maistre, TelecomTV (11:32):
Okay, thank you Kristian. Well I mean we've heard this a few times already I think this morning that there's a real opportunity for the telcos for the network operators to play a key role here and one that would be good for their business as well in terms of providing infrastructure connectivity and associated services around sovereign needs. But if we can maybe start with a question here for the panel about whether there are any specific AI use cases that the telecom network operators can enable that are considered so critical that the underlying AI models and training data must be developed and controlled by sovereign entities. So any specific use cases that telcos can get involved with in the sovereign domain. Diego, do you want to start there?
Diego R Lopez, Telefónica (12:33):
No, actually it's good that your question to ask that because we are now working and we are planning our first semi-public demo of a collaboration that we are having with Carol is with Red Hat and a small startup that is called Shanti on geofencing, on demonstrating the physical location of servers and hopefully in the future of any kind of device using the mobile network as the physical anchor because GPS the problem with GPS, I dunno if you have heard about this hacking that happened with the Swiss railroads application because the GPS depends on your terminal. If you can hack the terminal, you can hack the GPS. The idea is that you have a third party that is a trusted third party that can provide this information and right now we are trying to demonstrate that we can combine SPFI as defined the ITF with camera APIs and with all the whole thing to provide proof of location as part of the general extended way of dealing with the identity of our workload or a source of data on the 17th of the end of December will be our first public demo. We hope to have something to show you very soon to everyone and probably next year we can come here. I'm bringing something that is totally sobering to vision but I do believe it's a real quite interesting use case with many implications when we are talking about providing something that apart from fibers to some extent an antenna as we have that this is the physical widespread distribution that tel operators have now
Atul Purohit, Nokia (14:28):
Yeah, so if I understand the main question you're asking is the specific AI use cases from a controlled by the sovereign entities. Now from a telco perspective, I suppose we can divide this into a couple of areas. The first is what are the use cases required for survivability of the telco, the operational viability, criticalness as well as the security. So these are the use cases that a telco has to deliver related to these characteristics have to be sovereign in nature. Examples are many, for example the resource optimization on the radio side for example the spectrum optimization or anything to do with the critical communication for governments or for security agencies, et cetera. So a lot of such use cases have to be sovereign in nature, which telcos currently support and should continue to support with the sovereign envelope around it. Then you move up a little bit in terms of more on the management layer, more on the control layer if you like, the day-to-day running of the operations for example. So this is where also you can look at for example a bit of high level use cases for them. The sovereign requirement may or may not be as critical however, when it comes to some of the use cases that I mentioned earlier, I suppose that sovereignty is actually one of the key thing that there has to do. There has to be there in terms of use cases.
Ray Le Maistre, TelecomTV (15:47):
Okay, thank you Atul, and Kristian.
Kristian Toivo, Telecom Infra Project (15:50):
Yeah, so maybe somewhat controversially I would say I don't think there's particularly aspects on a telco that would differentiate in service use cases in general for AI unless we refer to things that partly Diego refer to as well. I mean if we connect it with the user with the identity location of a person, obviously telco has all that information so I would look at those. And then the other angle is of course they very often would be good partners to more public sector use cases and applications if you like, but I think I wanted to come back a little bit to the identity and the terminal or the user and now maybe having a bit too much of a mobile angle to it. I think it actually goes back to some of the things and challenges I sometimes see with the discussion around this edge cloud and the operator involvement in that.
(16:43)
I think these services, if they should be broadly applicable, they always have to be able to be accessed by any user, can't be connected to one particular operator and its subscribers, which means I think you need cooperation in a nation between operators for this to be feasible, applicable. Hence network APIs require full coverage in the market if you want to have any meaningful business developed on them. So I think those are the type of things where telcos are well placed, whether it's trust public sector, you use the location or user identity in some form. That's where I would go and start.
Ray Le Maistre, TelecomTV (17:23):
Okay. Any questions from the audience at this point?
Francis Haysom (17:30):
Okay, yeah Francis, just interested in what the panel think. When I look at these diagrams of what telecoms will do, how it will apply at the far edge, I see the same diagrams I saw about five years ago in terms of the edge use case for what does the panel see is the one that sovereignty brings to this and the differences that AI bring to this, is it truthfully different or is it just us taking a second dib on the same edge use cases that we didn't really take up five years ago?
Ray Le Maistre, TelecomTV (18:06):
He wants to go to that first.
Francis Haysom (18:08):
Hey
Kristian Toivo, Telecom Infra Project (18:08):
I dare to start. It's a loaded question or maybe a loaded answer and I think to some extent it might be a little bit that we are looking to solve the same problem another time. Now I think there are some real open questions on that. What inferencing is meaningful to do at the edge and what is the edge it's going to be the device. If it's the device, it doesn't really matter, you don't need the edge. You understand there are cases where it can't be the device because if it's VR glasses they probably don't have the form factor and the battery and all that so you have to go to the network. The question is do you have to do it in the edge or do you go up to a core site because latency wise there's no practical difference between that. At least that's what expertss tell me. So I think it's to be proven that there is a case for inferencing at large scale at the edge at the same time we don't know what those use cases are. Somebody said in the previous panel, it's very early days, so very often in this type of technological transformation things that we never thought about happen and they bring up new opportunities. But for me there is not an obvious case at this stage.
Diego R Lopez, Telefónica (19:24):
I would say that when it comes to ai, there is a difference that I foresee. I mean not in the architectures or where you put the dots and the lines et cetera because it's going to be a network that, but this in the way in which you're using data because data is used not only for being processed but as well to, I mean data is now becoming the way of programming the whole thing, but depending on the data you provide, the behavior mean that you provide for training the behavior of the thing that your training is going to be different. That's different because the sources of the data being assured that the sources of the data is what they're intended to et cetera becomes more important. That implies that telcos may not necessarily, that's the only ones that may have a role there when it comes to so sovereignty et cetera is the concern that that depends very much on the perception of the different actors there from the end users to the small companies to the big corporations to the governments, et cetera.
(20:31)
The concern of how important is for them to preserve a certain degree of control. We were talking this morning about control and resiliency. The important thing is that when it comes to the control, if you are dealing with your tech operator simply, and it's not because we are better than others for that, it's because we are obliged by local rules and not by the rules of the state of Delaware. That typically, I dunno why it's the state of Delaware, somebody could tell me why because it's simple achieve no I guess or because it's very, very, very expensive to litigate against them there. I dunno why this but yeah, but I don't know. But it is curious because
Audience member (21:16):
It's tax, it's tax and also you can incorporate very easily, quickly and it doesn't cost you anything.
Diego R Lopez, Telefónica (21:22):
So thank you. It's something I have learned at the end and it is quite important because you are obliged by local regulations and this is something that in the case if you are concerned about locally or local issues, it's quite natural for sure that's, that's nothing to do with against that someone else can appear and occupy that market niche. But is there and apart from that is, as I said with this use case that I was mentioning that by the way, forget to say that it's going to run in, well not so new, but we are very proud of it. Our Tesla lab in the Madrid area that is called next is that we can take advantage of these physical presence. There are cases for this I believe. So they're going to develop to become a real market. I'm not that sure because things change and in these days they change much faster than they used to.
Atul Purohit, Nokia (22:27):
There's actually something interesting that comes to my mind based on your question, Francis, A couple of years ago, I remember in London we had a Mac conference and one of the things I remember I made a point back then was that we have to look at this entire latency mumbo jumbo in three different numbers. So you have 70 20 10. Basically if you talk about the latency from the point of access to the point where the application is hosted, the overall latency, assuming it is about a hundred millisecond, you can roughly divide this into 70 20 and 10. Where at the edge the latency that you're dealing with is only about 10 out of the a hundred overall because the 20 comes from the edge to the central office and then from central office to the application hosting is where about 70 millisecond latency is what's happening now compared to that five, six years ago.
(23:17)
Is that the point where the application is hosted is shifting rather than an application setting only in a data center here are applications which are sitting closer to the edge. What Diego mentioned about Tesla, it's a good example you mentioned you can't think of a use case. I can because a connected car for example or physical ai, that's obviously going to come because of the evolution that happens in the hardware. There are certain requirements that for example, car has to do or a physical robot has to do, right? So Tesla opt whatever that robot is called, which means that you need essentially the workload to be sitting inference to be happening next to the workload which is nearer to the edge. So I suppose time is right now, five years ago it was an idea. Today it is an idea about to be realized with the advancements that we are seeing. So I suppose that's where we are. It's just one point I wanted to add.
Ray Le Maistre, TelecomTV (24:06):
Well, okay Dean, is it related to what we Okay, yeah, go ahead.
Audience member (24:14):
Inference currently probably accounts for about 10 gigawatts of power globally that's going to grow it over time. How many telcos have nuclear SMRs in their roadmap for sovereign energy source?
Diego R Lopez, Telefónica (24:32):
Well, among other things I do hope for the sake of the planet that we can reduce the energy cost of inference. This is something that you think that to make a school assignment for a 10 years old guy, boy you have spending that huge amount of energy and resources. I mean its tantalising when you think about it, I do hope that this is something that has to be reduced on the one hand. Second, the kind of inferences that are so much expensive has to do very much with natural language and not, probably not with tan language processing understanding but with natural language production to look like you are speaking, talking with a human being. In many cases I believe there are many uses for artificial intelligence that has nothing to do with LLMs, has nothing to do with this and has to do with an adaptive control of whatever I use. Very often the examples that for example, just when we have one here, dogs are intelligent, bees are intelligent, probably they're not as intelligent as some humans, but they can do things in an adaptive way and this is the kind of artificial intelligence I foresee that will be populating the network in the future. And my idea was it was going to be that, but things change and then someone came with LLMs and everyone became fascinated with LLMs. But I do believe that the times of LLMs will be over for different reasons. That's my take.
Ray Le Maistre, TelecomTV (26:21):
Okay. And just to point out, we're going to get to that kind of topic as well in the supply chain session coming next in terms of energy and so on, so forth. But great question, but just to come back to go a little bit further from the question that France has asked. I want to ask the panel whether you think that building investing in and running AI factories, whether that's a giga factory with partners involving billions of euros or smaller AI factories that still require a lot of investment but maybe aren't of that massive scale. Is this something that the telecom operators should be getting into? Now some are already there, Lenor was one of the early ones, Swisscom as well, Fastweb in Canada, we've seen Bell Canada and Telus making ways, but there's lots of other examples. But is this something that across the board, is this an investment worthwhile of the telcos to be making to position themselves with physical AI infrastructure to provide sovereign services? Who wants to start,
Atul Purohit, Nokia (27:39):
Who wants? Just one point I wanted to make here is that when you mention AI factory, it's not just physics part of it basically AI factory. I would rather call it more like an AI continuum because there is a lot of unspoken requirement that for example, the AI continuum brings on the telcos and I suppose a couple of them, for example, the edge computing, the distributed computing we already spoke of where the models are distributed and somebody needs to connect them. The secure connectivity backbone for example, if there is a research institute working with business, working with telecoms or somebody has to stitch that from a connectivity perspective. Is data residency sound?
Ray Le Maistre, TelecomTV (28:17):
Sure that's right in the telco's wheelhouse isn't it? I mean that's where they should be starting,
Atul Purohit, Nokia (28:22):
Which is where they have to start. But the point I suppose the challenge for telco right now is that ultimately it's basically just a bit pushing activity. How do you move up the value chain in terms of making it a bit more intelligent? So authentication of AI agents, imagine a future where there are like, I dunno, a billion, trillion agents who can authenticate them from a legit agent to a non-state or a state actor or whatever. So there are a couple of things I suppose that I could think of. Identity access management, et cetera. Telcos have to play a part in this, not just AI factory but more like building the entire AI continue.
Ray Le Maistre, TelecomTV (28:55):
But do you feel like they should be investing in their own facilities to be able to deliver this or just partner with the local branch of a hyperscaler?
Atul Purohit, Nokia (29:07):
No, I suppose this is again, again matter of choice, but for the lack of better word, I suppose it requires more like a framework based push from a local government perspective, I suppose what the overall sovety requirements for the nation is. And if for example hyperscalers are, I don't know, mostly North American based, they're also trying to build their own sovereign footprint. But I suppose equally if the data has to reside within a certain country, then telcos definitely have a part to play and that's what we are seeing like DT, et cetera. A lot of the operators have started to already start participating there.
Ray Le Maistre, TelecomTV (29:43):
Yes, with support of public finance as well of course, but to a certain extent who wants to go next? Kristian?
Kristian Toivo, Telecom Infra Project (29:52):
Yeah, I can comment. So I think you partly answered the question yourself because the examples you gave, I think
Ray Le Maistre, TelecomTV (29:57):
There's only a few examples.
Kristian Toivo, Telecom Infra Project (29:58):
Yeah, there was a couple of more telecom just announced
(30:01)
Germany yesterday. Anna learned a little bit more about that as well and we've seen some in Asia, Axiata, one of our leading members in TIP as well, IOH in DU has invested and beliefs and in Asia many operators will play that role and I think there's a strong link to the sovereign expectation there as well. So I think the smart answer is, it depends because I don't think it's for every operator given the vast sums and efforts and commitments included, I think it has to be partnership based or it makes sense to do a lot of partnerships and I think if you are in a position that you have a significant, what we call a B2B business as well, very often the telcos B2B business is focused on the public sector on top of it. Many of the leading telcos in Europe at 20 to 30% of the revenues is in the B2B side. For me the answer you can't not do it. I think you would exclude yourself from potential revenue growth and market positioning. You can't take the risk
Diego R Lopez, Telefónica (31:08):
Diego. Well if I were talking about putting my money on it, I wouldn't frankly. Okay. No seriously, I believe that there is a niche or is a market that is, I mean it is gone in the sense of the macro facilities for hosting any kind, kind of these big consumers, AI that requires gigabytes and things like that. It would be possible as a tool was mentioning for specialized services, very much connected with this actually with sovereignty or related with identity, with this extended identity was mentioning et cetera. Probably yes, but that we're talking about something that should not be that huge and far from what we now know because the other market is already there and is difficult to justify any kind of huge investments. Specifically right now that telcos are very much let's say in a very short cycles for whatever that is new. I mean the old business is totally understood and we follow deploying fiber. I'm building towers, I'm changing antennas, I'm occupying a spectrum, whatever. But when it comes to something new, and I have experienced that in my company, if in three, five years it's not providing revenue, high revenue, forget about it. So I don't see the point, but there are specialized kind of services related to this that probably we could occupy and telcos could occupy in a better, much better way than others. Okay.
Ray Le Maistre, TelecomTV (33:03):
Robert? Oh sorry. And another one up there.
Audience member (33:07):
Just following up the point you're mentioning here, I mean a lot of this stuff, while it sounds great in principle in a kind of overall market point of view, we seem to keep going back to niches essentially start from governments and work your way down. You probably won't get much further down than governments in lots of cases. I'm just wondering whether from a telco point of view, this just becomes a line of business, it's a feature on a service, not something at an infrastructure level. But what's your thought about that from conversations with customers? Are they looking to apply sovereignty wholesale or they saying, you know what, it's just a feature of our line of business that serves governments and pan national entities?
Diego R Lopez, Telefónica (33:40):
The point is that for certain features of those features, you need specific infrastructure. If you think about that, you are going to be rely much on mechanisms on with confidential computing right now some of the infrastructure we have is not supporting confidential computing. That's to give you an example, if you're talking about this geolocation right now, there is all these work that is ongoing for 6G around ISAC that you can, because with ISAC you can provide a match better position for example, and that would require additional infrastructure focus on that. That's the way I was talking about niches and not necessarily. And then there is the third one that is about the trust. This idea that we can certify at all was mentioning about certifying what this is legit or not alleged agents and things like that. Providing trust anchors to our customers or to customers in the European Union or in the marketing which we're present. That could make sense. And again, that would require additional infrastructure. Well they're not saying it's about gigantic infrastructure, it's about enhancing or updating infrastructure we have right now it's what they boot befall.
Audience member (34:58):
Is that more or less important than quantum safe networks, for example, just as an
Diego R Lopez, Telefónica (35:02):
No, no, it is the same. I mean at the end I was not going to, I didn't want to mention quantum safe networks and quantum compute et cetera because I was mentioning before and I but well if you want me say this is the quantum guy, I'm happy. So
Audience member (35:16):
Yes, go for it. It's always a question of calibration. A CFO is going to say, you want me to spend a billion dollars, am I going to make a quantum safe network that everybody gets benefit from or am I going to put a billion dollars into making a digital sovereignty? To your point, the minimum necessary infrastructure to support a hundred customers.
Diego R Lopez, Telefónica (35:31):
But for that you need quantum safe, otherwise, otherwise you cannot come. That's the point. So is that you can try to get an additional advantage that you're working on quantum safe or in other quantum capabilities to provide these services.
Kristian Toivo, Telecom Infra Project (35:48):
Okay. Can I add? Yeah, yeah, absolutely. And there's no question about the need for quantum safe and that was mentioned also in the previous panel. But back to your question, I mean, if you're an operator in either an European market or some of the Asian markets, you're kind of a trusted supplier to the public sector. Your 20 or even 30% of your revenues is business to business sales and who's going to pay for AI services? It's not you and I as private consumers, it's going to be embedded into our ARPU if you like. But if you want to maintain that 20 to 30% revenue and you have sovereignty requirements coming to your healthcare, to your tax authority, financial services, I think you probably need to think twice if you don't want to go into that and have a potential to offer those services. Spreading the risk, partnering with some players, but ultimately being able to front an offering of AI services, inferencing services that are safe produced in the country, have guardrails under regulation. I think there is a business case to be made. Of course it's going to compete with another things. It's going to compete with all the billions you have to do to upgrade your network. I don't think you can do either or you need to actually, as in any business, you need to manage both your opportunity for growing your business as well as taking care of your infrastructure and your operational safety or then you accept that you are just a connectivity provider and that's a business in itself,
Audience member (37:28):
A very good business if you can do it efficiently.
Kristian Toivo, Telecom Infra Project (37:30):
I think those choices are strategic choices that I pretty sure all the big operators and telcos are wrestling with all the time.
Ray Le Maistre, TelecomTV (37:38):
Yeah. Question up there. If you can introduce yourself before your question
Mark Gilmour (37:43):
Mark. Mark Gilmour, CTO at ConnectiviTree . So do you feel then that a consortium model that was used for submarine or things like that has no place for the telco in building these large sovereign AI factories or something like that? Is that a model that just really the telcos haven't got the opportunity to play in? I do have another question as well, but that refers, I
Kristian Toivo, Telecom Infra Project (38:13):
Just add to what I think there is partnership partnerships are necessary for that, whether it's the same partners and for subsea cables or some,
Mark Gilmour (38:21):
Well, there's a model that exists that kind of exists isn't there with the subsea cables, but really operators kind of got pushed to the side in most cases that was taken over by a different part of the ecosystem.
Kristian Toivo, Telecom Infra Project (38:35):
I mean the ones we've seen, the examples that were mentioned, the operators have a very solid key role in those partnerships. They're still partnerships in many cases, but they are fronted by the operators as far as I have understood some of the examples that Ray mentioned and some of the others you can read in the press. So I do think there's a role that might be different. I don't know the subsea cable consortium very well, but I think there's an operator play. That's what I'm
Mark Gilmour (39:01):
Thinking. Right. I did have another question, but it's not kind of related to this part, the discussion, but the other part earlier on at all, it's really kind of directed at you. You made a really interesting comment about the edge of far edge use cases. You mentioned a Tesla use case in there and connected cars and things like that. Just my question is are they being driven by latency or by something else that would bring a use case to the edge, far edge or that because if it's latency then it's kind of a similar discussion as we've previously had. And most of the latency requirements can be driven by, particularly here in Europe, we're already in an edge environment. If you look on a per country basis. Your comment intrigued me now and I just wanted to delve in on that particular point, whether there's something else other than latency that is driving it is sovereignty.
Atul Purohit, Nokia (40:02):
No, there is. So latency is just one of the examples, right? So sorry, one of the contributing factors, but for example, the inference if at all required nearer to where the execution, where the use cases, sorry, that's another thing that might be required. So latency is just a means to communicate to propagate the results of the inference.
Mark Gilmour (40:21):
Sorry, help. I need to be educated on that. Why does the inference need to be so close by that can't be solved by just a core data center. We're here in the uk, most of the UK can be served by two or three data centers from a latency want. So why does that inference need to move any closer?
Ray Le Maistre, TelecomTV (40:46):
So are you asking why would we need a hundred edge data centers in the UK rather than six regional data centers? Just the latency differences.
Mark Gilmour (40:56):
It's the same conversation as we had with MEC 10, 15 years ago. And I'm really trying to get into what's driving that closer.
Ray Le Maistre, TelecomTV (41:06):
And it's a discussion that the AI RAN conversation is bringing back into the fold as well.
Atul Purohit, Nokia (41:12):
The only thing I can say is, see, we have to go by the numbers I mentioned previously to Francis's point that it is 70 20 10. So just from a numbers perspective, the latency that each stage of where the workload is and where the consumption happens is distributed into So seventies, if the workload is somewhere in the central data center and then consumption is happening at the far edge somewhere at the consumer site, yeah, 70% of the latency is basically trying to get the data and only 10% is at the edge. So it depends on where the actual application resides. If the application is moving from centralized data center to for example, on top of a Tesla, then obviously there is a difference in equation that you have to look at in terms of latency. Does it play a role? Wood a hundred millisecond is an acceptable latency for execution of the use case, which is sitting right next to where the car or the consumer is. So I suppose these use cases are only evolving now five years ago, 10 years ago as it was, I suppose a very good idea, but the economics, the numbers did not justify today. Also, we still have challenges, but I suppose we are on the way. I think we will see, I suppose when the robotics et cetera gets launched, the use cases that requires inferences and the lower latency near to where the consumption is coming. I suppose more and more real.
Mark Gilmour (42:28):
I could come back again, but I dunno if you want me to,
Atul Purohit, Nokia (42:31):
Yeah, let's see. Same forum in five years time, maybe we'll have a better answer, but
Mark Gilmour (42:36):
It almost, so putting latency to the side, then it becomes a question of transport versus compute, which is the most cost effective? If you can transport it for less than the cost of the compute IE you put the compute, if you get better compute costs centralized, then transport it then. And I think this is all the common problem with edge or the far edge. Edge in the US is Europe, each European country. But here in Europe it's kind of individual country. So really interesting. Sorry, I could go on that one. I
Ray Le Maistre, TelecomTV (43:16):
Stop. But it is a good point and it comes down to what kind of investment and infrastructure might be required to deliver, what kind of services to what kind of customers, which is kind of what we're trying to explore here and figure out, as somebody just said before, or I think Robert, you just said, is it just a subset of some services that could be offered anyway by other investments that are going to happen anyway over time? So yeah, it's an interesting one to weigh up and that incredibly is our time up these, can we stretch the day? Is there an application that we should run for
Kristian Toivo, Telecom Infra Project (43:57):
Another hour?
Ray Le Maistre, TelecomTV (44:00):
But there is another session. There are two more sessions to come and the next one is on supply chain. We'll be looking at energy and other things as well. But in the meantime it's another cake break. There is coffee and tea as well, but cake break. But in the meantime to our panel, thank you very much. Great.
Thanks everybody to come back for this afternoon's sessions. So we'll start off by taking a look at this afternoon. So we've got three more sessions for you. We've got this one on AI, R&D and sovereign. Then we've got securing the digital supply chain that's come up quite a few times already so far today. And then we're finishing off this afternoon with the cameras will be turned off and it'll be a Chatham House Rules unthinkable session on digital sovereignty covering well, whatever we decide we're going to cover in that final session, driven largely by Andrew and Dean, but we'll find that out a bit later on this afternoon. But this is the session on AI, R&D, and sovereign capabilities looking at the role of AI research, AI factories and service ecosystems in advancing sovereignty goals. So just like the other sessions, I'm going to ask the speakers to introduce themselves one by one so everybody knows who they are and the perspective they're coming from. So Atul, if we can start with you.
Atul Purohit, Nokia (01:18):
Hi everybody, my name is Atul Purohit. I work as the market CTO and head of solutions for Nokia's Cloud network services business for Europe based here in the UK. Come with a lot of background from telco's, worked in Vodafone and BT before. Happy to be here and contribute to the subject.
Diego R Lopez, Telefónica (01:33):
Thanks. Great, thank you Diego. Diego Lopez. I work for Telefonica and well I lead a team right now that is mainly focused on applied research in network infrastructures and well aspects related to the standardization of these results.
Kristian Toivo, Telecom Infra Project (01:54):
Okay, pressing the button. I'm Krristian Toivo, executive director of Telecom Infra Project or TIP as we call it. TIP is a nonprofit trade organization focused on driving open connectivity platforms and lately of course AI and I noticed pleasantly some of the previous speakers connected openness very much with the AI journey as well as the sovereignty, which I firmly believe in. Tip has about 200 member organizations, a lot of basically all the European large operators as well as some North Americans and Asian ones as well as lots of vendors and there's about close to 1000 engineers or experts working in our six project groups, different initiatives driving technology forward. So really nice to be here talking about the super exciting topic. Thank you.
Ray Le Maistre, TelecomTV (02:46):
Okay, thanks everybody. Right. As with the other sessions as well, each of our expert speakers is going to have three minutes to start off. So we're going to start with Diego, come to Atul and finish with Kristian. So Diego, your three minutes starts now.
Diego R Lopez, Telefónica (03:04):
Okay. So basically we're talking about AI and research and development firstly about the critical thing of how we apply research. I mean something that has been around in Europe, we are very good at research and not so good at the application of this or to real business. I think that there is something that has to do very much with some divorce between academia in the widest sense possible. And from time to time in inside my company, myself and my colleagues are considered part of the academia as well and how we breach this gap. We are trying to address this in Telefonica with something that is, well we are constituting what we call joints research units with several universities and research institutes, et cetera. Just to explore this. This is on the one hand, on the other hand is the importance of open interfaces and open implementations and to avoid something that is totally false, that is the dichotomy that you have to choose between standards and open source implementation.
(04:14)
The idea is that they compliment and sometime ago in ETSI presented a diagram that has been used several times. I'm very proud because this is kind of things that you see from 10 to 10 say, hey, that's mine. Which is also sort of an infinite loop between what you can achieve with open source and we can achieve with open standards. Let me just remark a couple of things. When coming to AI has been said before and let me exist when it comes to ai, the important thing, the real thing to guarantee sovereignty about the data, not the models. Models are something that are for sure they're important, they're complicated there. But think about it, it's about how you manage the data and how you change the data and how you warranty that. The data remains where it should be. And just to finish, and this is the kind of things in which my team is working right now related to this is on how we facilitate that kind of data infrastructures that support AI of whatever the flavor of ai you can think, agents, LLMs, whatever.
(05:18)
We can facilitate this in an open environment, what allows an open market for data and for the results. And second, with the dream of a multi-party secure computation, storage, communication. And for that what we have has started to work and during the break I was talking with someone about this cannot spot him right now where he's in the room about the promise of quantum technologies for that and we are exploring these results. They are quite promising for the moment we are confined to the lab but we hope to have something to show very soon. That's it. Thank you.
Ray Le Maistre, TelecomTV (05:58):
Excellent. Great. Thank you Diego. And exciting news on the quantum developments there as well. Okay, Atal, you have your three minutes starting now.
Atul Purohit, Nokia (06:09):
Alright, so yeah, I just wanted to take help of this picture to basically just convey my point about the subject and now Nokia as everybody knows is mostly a technology supplier working with telco partners to enable critical communications, enable network infrastructure across the globe. The role of AI is obviously as we understand there are models, there are enablers, there are technology building blocks, et cetera. But from our perspective, basically AI is basically a distributed AI. So we call this entire thing as a distributed AI continuum, which means that AI basically have models at the core, at the edge as well as near to where the service points are from a Nokia perspective. We basically enable if you like the infrastructure needed to build the connectivity where for example, if the governance suggests that the AI model should be broken down into a distributed pattern, then obviously the connectivity becomes important.
(07:05)
There are certain expectations from a networks perspective, which are basically not explicitly called out in this entire euphoria around the AI adoption over the last few years. Nokia plays as well as telcos play a very important role in terms of making sure that the connectivity is available. So I've got these two key call outs that I wanted to specifically touch on. So one is that from a networks perspective, from a technology perspective, there's a lot of expectation, unset expectation around the autonomous automation of AI control loops. And the idea is that AI would work, but it has to work with a lot of support from a network perspective. A fast, stable, reliable and secure control loop at every stage of the entire network economy is what is required as well as a continuous reasoning correlation and improvements near it to where the decisions are made or near to where the services are consumed is required as well. So there's a lot of work that has to go in making sure that the network fabric is in place to make it effective as well as the ai, the SOM of the AI for example, requires the data to be secure, to be available in a specific geography, to be available across distributed models within a certain geography. So all of these basically translate into this type of a distributed AI continuum and basically from our perspective basically play an important role in the entire network fabric available. So that's my
Ray Le Maistre, TelecomTV (08:27):
Okay, great, thank you. And Christian.
Kristian Toivo, Telecom Infra Project (08:30):
Okay, so in tip we work a lot with how to apply AI to make network operation and telecom operations more efficient. I'm not going to talk about that. I'm going to try to see as the task was if I understood it right, what does the telco sector and mainly the operators, what role can they play in enabling this sovereignty? And I think the previous panel before launch touched on many relevant things, so I hope I won't repeat any of that, but I started by trying to figure out what is sovereign AI and of course I use chat GT and a few other sources and there was a number of steps and for me it was very clear I'm not going to waste the time to read through those seven points, but I think it's very unlikely that hardly any nation except for a few examples could master even nationally all of those steps from national compute to data pipelines to foundation models to security, industrial capability.
(09:29)
But I think there are aspects of this which the telecom sector and the operators themselves are particularly well placed to play a role in. And we heard some examples in the previous panel in certain markets where their operators have taken that role, but be it whatever there is an opportunity and I think the short term one is really key is actually to get in there and start helping to build all that connectivity and infrastructure that is needed. I just read this morning in a report that there's expected almost 3000 new data centers to come online in the next three, four years. They all need transport, optical transport by the way, in between them there's all the aspects about the real estate and bringing in AI as part of the B2B or mission critical offerings to public sector or even defense. But I think we haven't talked about so far, maybe unless I missed that in the morning when I wasn't here, is the whole process flow.
(10:30)
There's a supply chain discussion later on. I think operators are very well placed to manage and bring those different partners together that are needed. And there's also a talent and internal governance aspect when you talk about sovereign because it will come with regulation or at least requirements for guardrails that I think our sector and telecom operators particularly are quite good in implementing. Nobody's perfect but at least it's ingrained in the DNA. And last but not least, there's a longer term perspective. Maybe there will be AI fabric, a continuum like my Nokia colleague said, or maybe there will be an edge cloud of some sorts remains to be seen and I think there's going to be quite a few things that have to happen before that is there. Maybe that's going to be the basis for six g, I don't know. But the key thing right now, operators are well placed to help both build out partner and guarantee that sovereignty as we go forward. So let's just get on with the work. That's my view.
Ray Le Maistre, TelecomTV (11:32):
Okay, thank you Kristian. Well I mean we've heard this a few times already I think this morning that there's a real opportunity for the telcos for the network operators to play a key role here and one that would be good for their business as well in terms of providing infrastructure connectivity and associated services around sovereign needs. But if we can maybe start with a question here for the panel about whether there are any specific AI use cases that the telecom network operators can enable that are considered so critical that the underlying AI models and training data must be developed and controlled by sovereign entities. So any specific use cases that telcos can get involved with in the sovereign domain. Diego, do you want to start there?
Diego R Lopez, Telefónica (12:33):
No, actually it's good that your question to ask that because we are now working and we are planning our first semi-public demo of a collaboration that we are having with Carol is with Red Hat and a small startup that is called Shanti on geofencing, on demonstrating the physical location of servers and hopefully in the future of any kind of device using the mobile network as the physical anchor because GPS the problem with GPS, I dunno if you have heard about this hacking that happened with the Swiss railroads application because the GPS depends on your terminal. If you can hack the terminal, you can hack the GPS. The idea is that you have a third party that is a trusted third party that can provide this information and right now we are trying to demonstrate that we can combine SPFI as defined the ITF with camera APIs and with all the whole thing to provide proof of location as part of the general extended way of dealing with the identity of our workload or a source of data on the 17th of the end of December will be our first public demo. We hope to have something to show you very soon to everyone and probably next year we can come here. I'm bringing something that is totally sobering to vision but I do believe it's a real quite interesting use case with many implications when we are talking about providing something that apart from fibers to some extent an antenna as we have that this is the physical widespread distribution that tel operators have now
Atul Purohit, Nokia (14:28):
Yeah, so if I understand the main question you're asking is the specific AI use cases from a controlled by the sovereign entities. Now from a telco perspective, I suppose we can divide this into a couple of areas. The first is what are the use cases required for survivability of the telco, the operational viability, criticalness as well as the security. So these are the use cases that a telco has to deliver related to these characteristics have to be sovereign in nature. Examples are many, for example the resource optimization on the radio side for example the spectrum optimization or anything to do with the critical communication for governments or for security agencies, et cetera. So a lot of such use cases have to be sovereign in nature, which telcos currently support and should continue to support with the sovereign envelope around it. Then you move up a little bit in terms of more on the management layer, more on the control layer if you like, the day-to-day running of the operations for example. So this is where also you can look at for example a bit of high level use cases for them. The sovereign requirement may or may not be as critical however, when it comes to some of the use cases that I mentioned earlier, I suppose that sovereignty is actually one of the key thing that there has to do. There has to be there in terms of use cases.
Ray Le Maistre, TelecomTV (15:47):
Okay, thank you Atul, and Kristian.
Kristian Toivo, Telecom Infra Project (15:50):
Yeah, so maybe somewhat controversially I would say I don't think there's particularly aspects on a telco that would differentiate in service use cases in general for AI unless we refer to things that partly Diego refer to as well. I mean if we connect it with the user with the identity location of a person, obviously telco has all that information so I would look at those. And then the other angle is of course they very often would be good partners to more public sector use cases and applications if you like, but I think I wanted to come back a little bit to the identity and the terminal or the user and now maybe having a bit too much of a mobile angle to it. I think it actually goes back to some of the things and challenges I sometimes see with the discussion around this edge cloud and the operator involvement in that.
(16:43)
I think these services, if they should be broadly applicable, they always have to be able to be accessed by any user, can't be connected to one particular operator and its subscribers, which means I think you need cooperation in a nation between operators for this to be feasible, applicable. Hence network APIs require full coverage in the market if you want to have any meaningful business developed on them. So I think those are the type of things where telcos are well placed, whether it's trust public sector, you use the location or user identity in some form. That's where I would go and start.
Ray Le Maistre, TelecomTV (17:23):
Okay. Any questions from the audience at this point?
Francis Haysom (17:30):
Okay, yeah Francis, just interested in what the panel think. When I look at these diagrams of what telecoms will do, how it will apply at the far edge, I see the same diagrams I saw about five years ago in terms of the edge use case for what does the panel see is the one that sovereignty brings to this and the differences that AI bring to this, is it truthfully different or is it just us taking a second dib on the same edge use cases that we didn't really take up five years ago?
Ray Le Maistre, TelecomTV (18:06):
He wants to go to that first.
Francis Haysom (18:08):
Hey
Kristian Toivo, Telecom Infra Project (18:08):
I dare to start. It's a loaded question or maybe a loaded answer and I think to some extent it might be a little bit that we are looking to solve the same problem another time. Now I think there are some real open questions on that. What inferencing is meaningful to do at the edge and what is the edge it's going to be the device. If it's the device, it doesn't really matter, you don't need the edge. You understand there are cases where it can't be the device because if it's VR glasses they probably don't have the form factor and the battery and all that so you have to go to the network. The question is do you have to do it in the edge or do you go up to a core site because latency wise there's no practical difference between that. At least that's what expertss tell me. So I think it's to be proven that there is a case for inferencing at large scale at the edge at the same time we don't know what those use cases are. Somebody said in the previous panel, it's very early days, so very often in this type of technological transformation things that we never thought about happen and they bring up new opportunities. But for me there is not an obvious case at this stage.
Diego R Lopez, Telefónica (19:24):
I would say that when it comes to ai, there is a difference that I foresee. I mean not in the architectures or where you put the dots and the lines et cetera because it's going to be a network that, but this in the way in which you're using data because data is used not only for being processed but as well to, I mean data is now becoming the way of programming the whole thing, but depending on the data you provide, the behavior mean that you provide for training the behavior of the thing that your training is going to be different. That's different because the sources of the data being assured that the sources of the data is what they're intended to et cetera becomes more important. That implies that telcos may not necessarily, that's the only ones that may have a role there when it comes to so sovereignty et cetera is the concern that that depends very much on the perception of the different actors there from the end users to the small companies to the big corporations to the governments, et cetera.
(20:31)
The concern of how important is for them to preserve a certain degree of control. We were talking this morning about control and resiliency. The important thing is that when it comes to the control, if you are dealing with your tech operator simply, and it's not because we are better than others for that, it's because we are obliged by local rules and not by the rules of the state of Delaware. That typically, I dunno why it's the state of Delaware, somebody could tell me why because it's simple achieve no I guess or because it's very, very, very expensive to litigate against them there. I dunno why this but yeah, but I don't know. But it is curious because
Audience member (21:16):
It's tax, it's tax and also you can incorporate very easily, quickly and it doesn't cost you anything.
Diego R Lopez, Telefónica (21:22):
So thank you. It's something I have learned at the end and it is quite important because you are obliged by local regulations and this is something that in the case if you are concerned about locally or local issues, it's quite natural for sure that's, that's nothing to do with against that someone else can appear and occupy that market niche. But is there and apart from that is, as I said with this use case that I was mentioning that by the way, forget to say that it's going to run in, well not so new, but we are very proud of it. Our Tesla lab in the Madrid area that is called next is that we can take advantage of these physical presence. There are cases for this I believe. So they're going to develop to become a real market. I'm not that sure because things change and in these days they change much faster than they used to.
Atul Purohit, Nokia (22:27):
There's actually something interesting that comes to my mind based on your question, Francis, A couple of years ago, I remember in London we had a Mac conference and one of the things I remember I made a point back then was that we have to look at this entire latency mumbo jumbo in three different numbers. So you have 70 20 10. Basically if you talk about the latency from the point of access to the point where the application is hosted, the overall latency, assuming it is about a hundred millisecond, you can roughly divide this into 70 20 and 10. Where at the edge the latency that you're dealing with is only about 10 out of the a hundred overall because the 20 comes from the edge to the central office and then from central office to the application hosting is where about 70 millisecond latency is what's happening now compared to that five, six years ago.
(23:17)
Is that the point where the application is hosted is shifting rather than an application setting only in a data center here are applications which are sitting closer to the edge. What Diego mentioned about Tesla, it's a good example you mentioned you can't think of a use case. I can because a connected car for example or physical ai, that's obviously going to come because of the evolution that happens in the hardware. There are certain requirements that for example, car has to do or a physical robot has to do, right? So Tesla opt whatever that robot is called, which means that you need essentially the workload to be sitting inference to be happening next to the workload which is nearer to the edge. So I suppose time is right now, five years ago it was an idea. Today it is an idea about to be realized with the advancements that we are seeing. So I suppose that's where we are. It's just one point I wanted to add.
Ray Le Maistre, TelecomTV (24:06):
Well, okay Dean, is it related to what we Okay, yeah, go ahead.
Audience member (24:14):
Inference currently probably accounts for about 10 gigawatts of power globally that's going to grow it over time. How many telcos have nuclear SMRs in their roadmap for sovereign energy source?
Diego R Lopez, Telefónica (24:32):
Well, among other things I do hope for the sake of the planet that we can reduce the energy cost of inference. This is something that you think that to make a school assignment for a 10 years old guy, boy you have spending that huge amount of energy and resources. I mean its tantalising when you think about it, I do hope that this is something that has to be reduced on the one hand. Second, the kind of inferences that are so much expensive has to do very much with natural language and not, probably not with tan language processing understanding but with natural language production to look like you are speaking, talking with a human being. In many cases I believe there are many uses for artificial intelligence that has nothing to do with LLMs, has nothing to do with this and has to do with an adaptive control of whatever I use. Very often the examples that for example, just when we have one here, dogs are intelligent, bees are intelligent, probably they're not as intelligent as some humans, but they can do things in an adaptive way and this is the kind of artificial intelligence I foresee that will be populating the network in the future. And my idea was it was going to be that, but things change and then someone came with LLMs and everyone became fascinated with LLMs. But I do believe that the times of LLMs will be over for different reasons. That's my take.
Ray Le Maistre, TelecomTV (26:21):
Okay. And just to point out, we're going to get to that kind of topic as well in the supply chain session coming next in terms of energy and so on, so forth. But great question, but just to come back to go a little bit further from the question that France has asked. I want to ask the panel whether you think that building investing in and running AI factories, whether that's a giga factory with partners involving billions of euros or smaller AI factories that still require a lot of investment but maybe aren't of that massive scale. Is this something that the telecom operators should be getting into? Now some are already there, Lenor was one of the early ones, Swisscom as well, Fastweb in Canada, we've seen Bell Canada and Telus making ways, but there's lots of other examples. But is this something that across the board, is this an investment worthwhile of the telcos to be making to position themselves with physical AI infrastructure to provide sovereign services? Who wants to start,
Atul Purohit, Nokia (27:39):
Who wants? Just one point I wanted to make here is that when you mention AI factory, it's not just physics part of it basically AI factory. I would rather call it more like an AI continuum because there is a lot of unspoken requirement that for example, the AI continuum brings on the telcos and I suppose a couple of them, for example, the edge computing, the distributed computing we already spoke of where the models are distributed and somebody needs to connect them. The secure connectivity backbone for example, if there is a research institute working with business, working with telecoms or somebody has to stitch that from a connectivity perspective. Is data residency sound?
Ray Le Maistre, TelecomTV (28:17):
Sure that's right in the telco's wheelhouse isn't it? I mean that's where they should be starting,
Atul Purohit, Nokia (28:22):
Which is where they have to start. But the point I suppose the challenge for telco right now is that ultimately it's basically just a bit pushing activity. How do you move up the value chain in terms of making it a bit more intelligent? So authentication of AI agents, imagine a future where there are like, I dunno, a billion, trillion agents who can authenticate them from a legit agent to a non-state or a state actor or whatever. So there are a couple of things I suppose that I could think of. Identity access management, et cetera. Telcos have to play a part in this, not just AI factory but more like building the entire AI continue.
Ray Le Maistre, TelecomTV (28:55):
But do you feel like they should be investing in their own facilities to be able to deliver this or just partner with the local branch of a hyperscaler?
Atul Purohit, Nokia (29:07):
No, I suppose this is again, again matter of choice, but for the lack of better word, I suppose it requires more like a framework based push from a local government perspective, I suppose what the overall sovety requirements for the nation is. And if for example hyperscalers are, I don't know, mostly North American based, they're also trying to build their own sovereign footprint. But I suppose equally if the data has to reside within a certain country, then telcos definitely have a part to play and that's what we are seeing like DT, et cetera. A lot of the operators have started to already start participating there.
Ray Le Maistre, TelecomTV (29:43):
Yes, with support of public finance as well of course, but to a certain extent who wants to go next? Kristian?
Kristian Toivo, Telecom Infra Project (29:52):
Yeah, I can comment. So I think you partly answered the question yourself because the examples you gave, I think
Ray Le Maistre, TelecomTV (29:57):
There's only a few examples.
Kristian Toivo, Telecom Infra Project (29:58):
Yeah, there was a couple of more telecom just announced
(30:01)
Germany yesterday. Anna learned a little bit more about that as well and we've seen some in Asia, Axiata, one of our leading members in TIP as well, IOH in DU has invested and beliefs and in Asia many operators will play that role and I think there's a strong link to the sovereign expectation there as well. So I think the smart answer is, it depends because I don't think it's for every operator given the vast sums and efforts and commitments included, I think it has to be partnership based or it makes sense to do a lot of partnerships and I think if you are in a position that you have a significant, what we call a B2B business as well, very often the telcos B2B business is focused on the public sector on top of it. Many of the leading telcos in Europe at 20 to 30% of the revenues is in the B2B side. For me the answer you can't not do it. I think you would exclude yourself from potential revenue growth and market positioning. You can't take the risk
Diego R Lopez, Telefónica (31:08):
Diego. Well if I were talking about putting my money on it, I wouldn't frankly. Okay. No seriously, I believe that there is a niche or is a market that is, I mean it is gone in the sense of the macro facilities for hosting any kind, kind of these big consumers, AI that requires gigabytes and things like that. It would be possible as a tool was mentioning for specialized services, very much connected with this actually with sovereignty or related with identity, with this extended identity was mentioning et cetera. Probably yes, but that we're talking about something that should not be that huge and far from what we now know because the other market is already there and is difficult to justify any kind of huge investments. Specifically right now that telcos are very much let's say in a very short cycles for whatever that is new. I mean the old business is totally understood and we follow deploying fiber. I'm building towers, I'm changing antennas, I'm occupying a spectrum, whatever. But when it comes to something new, and I have experienced that in my company, if in three, five years it's not providing revenue, high revenue, forget about it. So I don't see the point, but there are specialized kind of services related to this that probably we could occupy and telcos could occupy in a better, much better way than others. Okay.
Ray Le Maistre, TelecomTV (33:03):
Robert? Oh sorry. And another one up there.
Audience member (33:07):
Just following up the point you're mentioning here, I mean a lot of this stuff, while it sounds great in principle in a kind of overall market point of view, we seem to keep going back to niches essentially start from governments and work your way down. You probably won't get much further down than governments in lots of cases. I'm just wondering whether from a telco point of view, this just becomes a line of business, it's a feature on a service, not something at an infrastructure level. But what's your thought about that from conversations with customers? Are they looking to apply sovereignty wholesale or they saying, you know what, it's just a feature of our line of business that serves governments and pan national entities?
Diego R Lopez, Telefónica (33:40):
The point is that for certain features of those features, you need specific infrastructure. If you think about that, you are going to be rely much on mechanisms on with confidential computing right now some of the infrastructure we have is not supporting confidential computing. That's to give you an example, if you're talking about this geolocation right now, there is all these work that is ongoing for 6G around ISAC that you can, because with ISAC you can provide a match better position for example, and that would require additional infrastructure focus on that. That's the way I was talking about niches and not necessarily. And then there is the third one that is about the trust. This idea that we can certify at all was mentioning about certifying what this is legit or not alleged agents and things like that. Providing trust anchors to our customers or to customers in the European Union or in the marketing which we're present. That could make sense. And again, that would require additional infrastructure. Well they're not saying it's about gigantic infrastructure, it's about enhancing or updating infrastructure we have right now it's what they boot befall.
Audience member (34:58):
Is that more or less important than quantum safe networks, for example, just as an
Diego R Lopez, Telefónica (35:02):
No, no, it is the same. I mean at the end I was not going to, I didn't want to mention quantum safe networks and quantum compute et cetera because I was mentioning before and I but well if you want me say this is the quantum guy, I'm happy. So
Audience member (35:16):
Yes, go for it. It's always a question of calibration. A CFO is going to say, you want me to spend a billion dollars, am I going to make a quantum safe network that everybody gets benefit from or am I going to put a billion dollars into making a digital sovereignty? To your point, the minimum necessary infrastructure to support a hundred customers.
Diego R Lopez, Telefónica (35:31):
But for that you need quantum safe, otherwise, otherwise you cannot come. That's the point. So is that you can try to get an additional advantage that you're working on quantum safe or in other quantum capabilities to provide these services.
Kristian Toivo, Telecom Infra Project (35:48):
Okay. Can I add? Yeah, yeah, absolutely. And there's no question about the need for quantum safe and that was mentioned also in the previous panel. But back to your question, I mean, if you're an operator in either an European market or some of the Asian markets, you're kind of a trusted supplier to the public sector. Your 20 or even 30% of your revenues is business to business sales and who's going to pay for AI services? It's not you and I as private consumers, it's going to be embedded into our ARPU if you like. But if you want to maintain that 20 to 30% revenue and you have sovereignty requirements coming to your healthcare, to your tax authority, financial services, I think you probably need to think twice if you don't want to go into that and have a potential to offer those services. Spreading the risk, partnering with some players, but ultimately being able to front an offering of AI services, inferencing services that are safe produced in the country, have guardrails under regulation. I think there is a business case to be made. Of course it's going to compete with another things. It's going to compete with all the billions you have to do to upgrade your network. I don't think you can do either or you need to actually, as in any business, you need to manage both your opportunity for growing your business as well as taking care of your infrastructure and your operational safety or then you accept that you are just a connectivity provider and that's a business in itself,
Audience member (37:28):
A very good business if you can do it efficiently.
Kristian Toivo, Telecom Infra Project (37:30):
I think those choices are strategic choices that I pretty sure all the big operators and telcos are wrestling with all the time.
Ray Le Maistre, TelecomTV (37:38):
Yeah. Question up there. If you can introduce yourself before your question
Mark Gilmour (37:43):
Mark. Mark Gilmour, CTO at ConnectiviTree . So do you feel then that a consortium model that was used for submarine or things like that has no place for the telco in building these large sovereign AI factories or something like that? Is that a model that just really the telcos haven't got the opportunity to play in? I do have another question as well, but that refers, I
Kristian Toivo, Telecom Infra Project (38:13):
Just add to what I think there is partnership partnerships are necessary for that, whether it's the same partners and for subsea cables or some,
Mark Gilmour (38:21):
Well, there's a model that exists that kind of exists isn't there with the subsea cables, but really operators kind of got pushed to the side in most cases that was taken over by a different part of the ecosystem.
Kristian Toivo, Telecom Infra Project (38:35):
I mean the ones we've seen, the examples that were mentioned, the operators have a very solid key role in those partnerships. They're still partnerships in many cases, but they are fronted by the operators as far as I have understood some of the examples that Ray mentioned and some of the others you can read in the press. So I do think there's a role that might be different. I don't know the subsea cable consortium very well, but I think there's an operator play. That's what I'm
Mark Gilmour (39:01):
Thinking. Right. I did have another question, but it's not kind of related to this part, the discussion, but the other part earlier on at all, it's really kind of directed at you. You made a really interesting comment about the edge of far edge use cases. You mentioned a Tesla use case in there and connected cars and things like that. Just my question is are they being driven by latency or by something else that would bring a use case to the edge, far edge or that because if it's latency then it's kind of a similar discussion as we've previously had. And most of the latency requirements can be driven by, particularly here in Europe, we're already in an edge environment. If you look on a per country basis. Your comment intrigued me now and I just wanted to delve in on that particular point, whether there's something else other than latency that is driving it is sovereignty.
Atul Purohit, Nokia (40:02):
No, there is. So latency is just one of the examples, right? So sorry, one of the contributing factors, but for example, the inference if at all required nearer to where the execution, where the use cases, sorry, that's another thing that might be required. So latency is just a means to communicate to propagate the results of the inference.
Mark Gilmour (40:21):
Sorry, help. I need to be educated on that. Why does the inference need to be so close by that can't be solved by just a core data center. We're here in the uk, most of the UK can be served by two or three data centers from a latency want. So why does that inference need to move any closer?
Ray Le Maistre, TelecomTV (40:46):
So are you asking why would we need a hundred edge data centers in the UK rather than six regional data centers? Just the latency differences.
Mark Gilmour (40:56):
It's the same conversation as we had with MEC 10, 15 years ago. And I'm really trying to get into what's driving that closer.
Ray Le Maistre, TelecomTV (41:06):
And it's a discussion that the AI RAN conversation is bringing back into the fold as well.
Atul Purohit, Nokia (41:12):
The only thing I can say is, see, we have to go by the numbers I mentioned previously to Francis's point that it is 70 20 10. So just from a numbers perspective, the latency that each stage of where the workload is and where the consumption happens is distributed into So seventies, if the workload is somewhere in the central data center and then consumption is happening at the far edge somewhere at the consumer site, yeah, 70% of the latency is basically trying to get the data and only 10% is at the edge. So it depends on where the actual application resides. If the application is moving from centralized data center to for example, on top of a Tesla, then obviously there is a difference in equation that you have to look at in terms of latency. Does it play a role? Wood a hundred millisecond is an acceptable latency for execution of the use case, which is sitting right next to where the car or the consumer is. So I suppose these use cases are only evolving now five years ago, 10 years ago as it was, I suppose a very good idea, but the economics, the numbers did not justify today. Also, we still have challenges, but I suppose we are on the way. I think we will see, I suppose when the robotics et cetera gets launched, the use cases that requires inferences and the lower latency near to where the consumption is coming. I suppose more and more real.
Mark Gilmour (42:28):
I could come back again, but I dunno if you want me to,
Atul Purohit, Nokia (42:31):
Yeah, let's see. Same forum in five years time, maybe we'll have a better answer, but
Mark Gilmour (42:36):
It almost, so putting latency to the side, then it becomes a question of transport versus compute, which is the most cost effective? If you can transport it for less than the cost of the compute IE you put the compute, if you get better compute costs centralized, then transport it then. And I think this is all the common problem with edge or the far edge. Edge in the US is Europe, each European country. But here in Europe it's kind of individual country. So really interesting. Sorry, I could go on that one. I
Ray Le Maistre, TelecomTV (43:16):
Stop. But it is a good point and it comes down to what kind of investment and infrastructure might be required to deliver, what kind of services to what kind of customers, which is kind of what we're trying to explore here and figure out, as somebody just said before, or I think Robert, you just said, is it just a subset of some services that could be offered anyway by other investments that are going to happen anyway over time? So yeah, it's an interesting one to weigh up and that incredibly is our time up these, can we stretch the day? Is there an application that we should run for
Kristian Toivo, Telecom Infra Project (43:57):
Another hour?
Ray Le Maistre, TelecomTV (44:00):
But there is another session. There are two more sessions to come and the next one is on supply chain. We'll be looking at energy and other things as well. But in the meantime it's another cake break. There is coffee and tea as well, but cake break. But in the meantime to our panel, thank you very much. Great.
Please note that video transcripts are provided for reference only – content may vary from the published video or contain inaccuracies.
Panel Discussion
This session examines the roles of AI research, AI factories and service ecosystems in advancing sovereignty goals. The discussion covers opportunities for telcos to embed AI into networks and services, while ensuring responsible use, supply-chain assurance and alignment with sovereign priorities.
Recorded December 2025
Participants
Atul Purohit
Market CTO & Head of Solutions, Cloud and Network Services Europe, Nokia
Diego R Lopez
Senior Technology Expert, Telefónica, and ETSI Fellow
Kristian Toivo
Executive Director, Telecom Infra Project (TIP)
