Ray Le Maistre, TelecomTV (00:07):
So it's MWC26. We are here in Barcelona and Bruno Zerbib, who is Chief Technology and Innovation Officer at the Orange Group, has joined us. Bruno, thanks very much for taking time out of your busy schedule. Really appreciate it. Now, everybody here at the show is talking about Agentic AI, but where is the industry in terms of actual real-world use cases and potential deployments? And what's the reality for Orange at the moment in terms of Agentic AI?

Bruno Zerbib, Orange (00:39):
So I would say you have three distinct areas that are progressing at a different pace. The first one where we made a lot of inroads, I would say in the last 12 to 18 months, has been with regards to customer relationship. The fact that we could evolve from the old-style chatbots to something that will be much more proactive, much more hyper-personalised in the way we interact with our customers, that will be also grounded in very advanced understanding of data. It's personalised data. So I think this is where the biggest breakthrough has become visible. And now you're seeing the fact that we are either using our stores and we are augmenting the interaction between the people that are essentially talking to our customers over there with Agentic AI, or we are using next-generation chatbots, but it's really Agentic as well. And the reason why it's Agentic AI is not just LLMs.

(01:39):
There's a notion of routing to the right part of the, I would say, the interaction based on the needs and all of those things. So I would say in this area, it has become predictive. We know and deterministic. We know how quickly we're progressing. We know what we're going to achieve. We understand the economics and it's going to be great. It's going to be great and it's going to be great for the customer. And so that's number one.

(02:09):
The second one that is exciting and terrifying is co-generation. And we are a software company and we have legacy IT that we need to fix. We want to reduce the tech debt. It's fantastic to use Agentic AI for solving this. Plus we are working on the next new capabilities. It could be the new app on your mobile for customer care, all of those things. A lot of that development now is going to accelerate with AI. The economics are clear. The trajectory is a bit more unclear because there's a lot of training, but we know where we're going. The third one, I'm sure that's the one you would expect me to talk about first, but it's the third one that is the most challenging one. And the one that is the most core to the way we operate is obviously how we transform the network with Agentic AI.

(03:14):
That's essentially if you look at MWC, what everybody's talking about is Agentic AI changing the way you are managing your network, right? Automation, all of those things. And this one, we really have to divide and conquer into smaller pieces to have a more, I would say, comprehensive answer. Digital twins are going to be powered by Agentic AI for sure. The way we think about automation is Agentic AI as well. The way we think about field ops, our ability to understand not just the physical topology, but also the logical configuration is being managed by Agentic AI. And I would say for this third bucket, we are really working closely with the vendors and we're seeing that they are making progress. What is challenging is the end-to-end coming together.

Ray Le Maistre, TelecomTV (04:09):
You're

Bruno Zerbib, Orange (04:10):
Right. So it's still fragmented. I would say it's a bottom-up construct at this point in time. We are seeing islands of Agentic AIs automating things that used to be manual, but can I tell you I have the Space IDC experience where the whole thing is automated end-to-end? No, I don't have that. Can I tell you when I'm going to be able to have that full end-to-end experience fully automated with Agentic AI with this one brain? I can't tell you that. And so when you work for a company like Orange, you need to risk that because I cannot go through a tunnel of two, three years over-promising, overselling this Nirvana thing. And so what we do is divide and conquer. We look at smaller parts of the network, be functional, could be different kinds of functional areas we want to address.

(04:56):
And we are solving that with Agentic AI, one functional area at a time. And then at some point in time, we'll bring all of that together.

Ray Le Maistre, TelecomTV (05:04):
Okay. And I guess key to that then is planning that in a way so it doesn't become siloed, I guess. It's important that these things do come together in one homogenous plan at the end of the day.

Bruno Zerbib, Orange (05:18):
Yeah. And the number one problem that we have with any large organisation is that organisations are siloed. So to really tap into the full potential of Agentic AI, to unlock Agentic AI, you have to change the culture because if humans are siloed and if we don't want to share data, if we don't want to work together, then it's going to be very hard for organisations to allow another team to be part of that end-to-end process. That is the reason why part of my job, which might be surprising as a tech guy, is really changing the way we work. We call that the new operating model at Orange. People might say we are becoming more like a hyperscaler, all those things, but it's much more solid than that. It's changing the way we work together, the way we trust each other, the way we learn what are the other problem areas and how one plus one equals three, because it cannot be a zero-sum game.

(06:11):
And so that's the hardest part.

Ray Le Maistre, TelecomTV (06:13):
Okay. Excellent. Now still vaguely on the topic of AI, AI RAN is also showing up in more and more strategies and announcements, but it can mean a lot of different things to different companies. What does it mean for Orange?

Bruno Zerbib, Orange (06:33):
So without being super, super, super technical, when you look at a very high-level topology, right, you have the radio unit that really does super low latency modulation, and then you have what's going on at the basement within the baseband, right? You have all the MIMO optimisation, the fact that you can do more by combining multiple signals together, and then obviously you have what can be done closer to the core and then in the core. The radio unit right now, I don't believe that it can be disrupted by a GPU because it's extremely low latency, it's really ASIC-based, but the DU or the baseband, this is where it could be powered by a GPU.

(07:21):
So if that happens, then it means you see ASIC being replaced by a GPU. And then it means that within a cycle of the hardware being deployed, you might have software updates that improve the kind of spectrum efficiency you can get out of that GPU for a given amount of spectrum allocation, right? So we are going through the proof of concept. Obviously, I mean, there is the Nokia NVIDIA announcement, so we're working with all of them. We're going to deploy that. We're going to test that. It's possible that it's going to work out, but we need to go through the motions. We need to test it. And then the question is, is that a zero-sum game again? I'm using that term. Are we just shifting costs from ASIC to GPUs or is it going to give us more? And for me, the most important thing is, is that going to give me higher spectral efficiency and is it going to lower my CapEx?

(08:23):
And if that's the case, that will have a very bright future. If it's not the case, then it's a gimmick. We don't know yet, but for sure, I'm not dismissive of it.

Ray Le Maistre, TelecomTV (08:32):
Okay. Maybe come back to that next year, because at the moment, it used to be a case that a year things would move quite slowly, but now in a year, so much changes in the telecom sector, that's definitely something we're seeing. Now, there are some amazing advances to be made as well with the data that telcos have at their disposal and coupling that with the increasing sophistication of AI systems that you mentioned earlier on digital twins. To what extent do digital twins play a role in your strategy and what you're doing right now? Because this sounds like something that's really being put to good use in the industry now and enabling that trust of the telcos in what AI can achieve.

Bruno Zerbib, Orange (09:18):
So you're combining multiple concepts which are very core to us because you just used the word trust, which is at the centre of our strategy. Well done. So listen, it's very interesting because when we think about digital twins, one of the key discussions is should it run on the public cloud? Should it run on the private cloud? And part of the discussion is trust and big data. So we have kind of conflicting expectations. I want to have the most advanced big data engine, could be Fabric, it could be BigQuery from Google or Fabric from Microsoft, but at the same time, I need complete control and it needs to be completely safe and obviously we have regulation. So those are the things we always kind of have to balance out. But for us, you're exactly right. For us, digital twin truly emphasises the power of big data coupled with AI.

(10:09):
So we need speed of processing, we need very advanced processing and we need AI. And all of those things will lead into very significant reduction, for instance, of mean time to repair. Our prediction is, this is one of the things we shared during the capital market day two weeks ago. We are evaluating the impact of using digital twins that will turn, for instance, into a 30% reduction of mean time to repair. It is significant because it's 30% of a network that is becoming even more complex and even more powerful. So it's a mix of big data, of trust and AI for sure.

Ray Le Maistre, TelecomTV (10:47):
Okay, excellent. Now we are at Mobile World Congress, so we should turn to a mobile-specific topic for a moment. And this is something I know you've been looking at for a while. How do you see mobile uplink data flows developing as new devices and applications come online? And how do you prepare for the potential data flows that this might bring about?

Bruno Zerbib, Orange (11:13):
So we're working closely with the GSMA. We actually, we have a task force for monitoring the overall evolution of the AI traffic. It's one of the key initiatives. We're doing this obviously across the group. And if you look at the starting point today, I would say the baseline is an 80/20 mix, 80% download 20% upload, maybe even 90/10. That has changed the last few years with the increase of video conferencing, which essentially is pushing towards a bit more of a symmetrical ratio, but still we're very far from symmetry. So with AI, we think the ratio is going to change and the way we have designed our networks. Actually, if you look at the antennas themselves, they've been optimised for an 80% download, 20% upload. The challenge that we're going to have, which is very obvious when you think about it, is uploading is much more energy consuming. This is actually the ... I didn't think about it, but that's the elephant in the room.

Ray Le Maistre, TelecomTV (12:19):
When

Bruno Zerbib, Orange (12:19):
I was trying to think about that, I was thinking about the network itself, but the device actually is maybe the main culprit because if you have your phone and then your phone needs to start emitting much more than it needs to start receiving, I can tell you that will have a significant impact on the battery. So I don't know what the bottleneck is going to be, the primary bottleneck is going to be. I thought it was maybe going to be more on the RAN side, but it might end up being the device because of that. But to your point, we're going to have to rethink progressively. We don't know yet the speed. The way we think about emission and reception and change and that optimisation ratio over the next few years, for sure.

Ray Le Maistre, TelecomTV (12:57):
Okay. So that's an interesting one to watch and now everyone's going to be thinking about that handset issue as well. And finally, Bruno, we're here at MWC. It's still the first quarter of the year. What would you like to see throughout the rest of 2024 from the ecosystem that would help with your plans?

Bruno Zerbib, Orange (13:22):
So it's a good one. I would say the most important thing for me is the demystification of new technology. And I think what we've been struggling with over the last 20 plus years in the segmentation space is we have those new big things that show up and then they are not within our grasp. We don't really know how to turn that into something that's operationally sound. So I want to land, for instance, the technology, I want to see how AI Agentic is going to land. Really, I want to be able to use all of those great technologies we've been talking about into our job lines. I want to see how they are transforming ourselves in very concrete ways and having a great vision with a lot of buzzwords is good, but for me, my obsession right now is to make it real and to make it real very quickly.

(14:10):
And so we are going to go into a very practical set of discussions, job line by job line, function by function, to see how we can have agile milestones quarter after quarter to see how this is making us a better company.

Ray Le Maistre, TelecomTV (14:24):
Okay. Well, before we know it, we'll be back at MWC27 and we'll be able to examine if those real use cases, what kind of impact they're having on companies like Orange, but also the whole ecosystem. So Bruno, thanks so much for joining us. It's great to see you. Great to see you.