Ray Le Maistre, TelecomTV (00:07):
So it's MWC 25 in Barcelona. I'm here with Stéphane Téral he's the founder of Téral Research. Stéphane, thanks so much for taking time in your schedule to join us here today to talk about one of the, probably the hottest topics in mobile networking at this show, which is AI ran. And you've been looking into this, I know you had a report out recently. What's your view of AI ran as a concept? Do you think this is something that the mobile operator community will latch onto and see as something of interest for the future? Well, it's only

Stéphane Téral, Téral Research (00:50):
Very interesting, however, this is leveraging the entire evolution of RAN. So on the one hand you have D RAN moving into cran, which was adopted by the Koreans and the Japanese back in 2010 with LTE. Then you go to virtualization V RAN and eventually you open the front pole and you end up with open RAN. Well, let's say open if you come from the vRAN world. But then when we think about the intelligence, so Deran, you have to look at this. This is where the intelligence of the deran was. So this evolved into centralized csun, which is morphing into aaps. Well, CSUN is really aaps and sun there. That's self organizing networks. So you have the rig with the aaps Xap and all of this is converging within the open RAN domain. And then all of a sudden you leverage that and you say, okay, let's kill it and do something radically different.

(02:05):
And that is

(02:06):
on the one hand, you take the intelligence, you bring AI workloads, and on the other hand you take the RAN functionality and you developed RAN workloads, you combine the two and you have AI RAN. So this is what it is all about. At the same time, it's interesting to look at who is behind this. This is the Sun, sun Empire, okay. Southbank arm, Nvidia, this is the entire family. They have plenty of money. So you throw hundreds of millions of dollars at it and you can break the RAN world and propose something totally different.

(02:48):
So yeah, I mean people are looking at it because when you are Nokia, Ericsson and Samsung in particular, you'd better be on board. They are, thanks God, because you may lose your software side of the RAN business and end up only sending radio units,

Ray Le Maistre, TelecomTV (03:08):
Right? So if this takes off and becomes a popular architecture choice with the mobile operators, that's going to accelerate the shift away from where we are now with traditional ran towards virtualized ran

Stéphane Téral, Téral Research (03:27):
Well, I mean where we are now, we are actually in a transition mode from transitional RAN to Open vRAN, open vRAN 2025. We have been saying this since 2021 is the trigger. And we see the ramp up at and t, then Vodafone is coming on board and now we are, I mean DEU Telecom already is gradually ramping up its deployment in Germany and in other properties in Europe you have the Japanese, they have been there from day one. There was a big slowdown last year. But when you look things are moving, all of a sudden you have the open Vern market that is really taking off. Verizon is about to coming on board as well. So that's where we are. And all of a sudden you drop that AI RAN bomb in this momentum. So is this going to change things? I don't think it's going to change things in the near future because you need to prove that it's viable to, I mean this AI RAN is viable working and delivering the promise of, on the one hand, bringing more efficiency and on the other hand, turning a cost center into a moneymaking machine.

Ray Le Maistre, TelecomTV (04:53):
Well, this is some of the messaging that we've already been hearing from the supporters and you can see why, because where it's still in this situation where 5G return on investment is still a really big deal for the industry. So nobody wants to even consider another way of doing things without some kind of ROI model to go with it. But it seems like a bit of a tough sell at the moment. I mean the AI Ran Alliance has been going for a year now and it might have more than 70 members, but only that is correct. Seven, I think of those are network operators. So everybody, there seems to be some who are really, like we mentioned SoftBank and T-Mobile is fully behind this as well and a few others. But a lot of operators seem to be stepping back and just letting things develop and have a watching brief if you like. Do you think it's going to be like that for a while?

Stéphane Téral, Téral Research (05:51):
Yes. Because again, we are just in this transition mode of migrating from traditional RAN to open their ran. So it took a while to actually have the Brownfield service providers to sort out whether or not it would work in their network. I mean, I'm talking about open now. They know that it can. So they decided to introduce that novel architecture in their network. And this was led by the pioneers like Rakuten Mobile on the one hand and dish on the other hand in the us. But I mean during this time and because this started in 2019, look, I mean we're in 2025. So it took a while to solve this out because of what Vodafone has been saying. The integration process takes some time. This is very complicated. And you need to make sure that when you reassemble the RAN with all the open RAN components, everything works together.

(07:06):
But so for a minute, forget about AI RAN. Nobody's ready to go to AI RAN. And by the way, in this AI RAN concept, so you have three components and the AI four ran is already there. So my question to the AI ran alliance is what else are you going to bring to the table? Because AI four ran is CSUN plus non-real time rig, plus near realtime rig. And when you look at the non-real time R, you already have a large ecosystem. We have least 50, 53rd party suppliers or app developers. And if you look at the Ericson Rick platform that they are deploying at and t, they're onboarding already dozens of our app suppliers. And this is AI four RAM already. So when I hear, oh, we are going to work on bringing more efficiencies and energy savings, I am hearing and I'm like, well, you'd better think about spectral efficiencies because the pressure is on the spectrum.

(08:17):
I'm not saying that there is no pressure on bringing more energy savings to the table. No, no. I mean energy savings is a never ending story, but you already have very powerful tools to do that. Now, on the spectrum efficiency, you need to do more, you need to squeeze more bit per second per TZ from your spectrum such that you don't need to add new spectrum or you can postpone the need for new spectrum. And you may not have actually new spectrum in the pipeline depending on which nation you are operating. So there are few spectral efficiency, but Xap is going to be very difficult to crack because this is the proprietary domain of Ericsson, Samsung and Nokia. It's a world garden and they're not ready to open the door.

Ray Le Maistre, TelecomTV (09:20):
They're being pressured, aren't they? So it'll be a slow process. And in terms of AI around, I mean a lot of the talk at the moment is around the membership and the concepts of the AI ran alliance, but there are other alternative looks at this, aren't there? I mean, you were telling me yesterday about is it Microsoft?

Stéphane Téral, Téral Research (09:41):
Yeah, so that is correct. What's interesting is that at the time SoftBank published, its not only research papers, but white papers in the fourth quarter, and I've seen things in December, 2024. At the same time, Microsoft also published a very interesting research paper about their vision of using combination of GPUs and CPUs and also the different deployment scenarios where you can have a full blown AI RAN or a hybrid RAN, which is bringing the best of traditional world and traditional RAN world and AI RAN as they propose it, or you stay traditional RAN and gradually you bring AI RAN functionality until at some point it becomes AI RAN. So this is a very interesting concept, which is different from, because they're using probes to collect data in the network in different parts of the network. The issue they have is that there is no, it's very difficult to get real RAN data to train your model, but everybody has the same problem except SoftBank, right? But at least Microsoft is proposing a very compelling alternative.

Ray Le Maistre, TelecomTV (11:18):
Okay, interesting. Well, I know that you are going to be tracking this particular development in the market. Absolutely. And I'm sure there's going to be lots of talk about here at this show and in the months ahead because everybody is keeping an eye on it because there isn't an aspect of AI that nobody wants to miss out on something they should know about the impact of AI in telecom. So Stéphane, thanks very much for joining us. Very great to talk to you.

Stéphane Téral, Téral Research (11:45):
Thank you very much, Ray. Thanks for having me.