Guy Daniels, TelecomTV (00:24):
Hello, you're watching the future of RAN Summit and our panel discussion on the emergence of AI RAN and its commercial viability. I'm Guy Daniels and welcome to the first program of this year's summit. AI RAN has certainly succeeded in dominating the mobile agenda and its supporters have made bold promises about its potential impact. But what constitutes a viable AI RAN roadmap? Which use cases will deliver value and how do operators architect production ready AI RAN deployments? Well, I'm delighted to say that joining me on the program to discuss the viability of AI RAN are Agi Ed, who is head of cloud and AI RAN at Nokia. Vihang Kamble, who is CTO ran business unit for Rakuten Symphony. Ignacio Gonzalez, Business Development Director, EMEA at Red Hat, and Francis Haysom, principal analyst with Appledore Research. Hello, everyone. It's really good to see you all. Thanks so much for taking part in our summit this year.

(01:37):
And the first question of the future of RAN Summit is what is the reality today of AI RAN? AG, let me come across to you first if I come for this one. What is the reality today? What are the early tests and trials evaluating and what do we know so far?

Aji Ed, Nokia (01:57):
Yeah, thank you, guys. So if you look back, so when 2G and 3G came up, it was all about voice. Then data came in with the 4G. Then when 5G came in, it's bit more data plus the video. So now we talk about AI being the new workload. So if you look at the current telco world, it's AI become the most dominant, the workload, what we are going to talk about. So AI is going to redefine how the networks are engineered, deployed, and managed. So at Nokia, when we announced the partnership collaboration in AIN in October 2025, that's with NVIDIA GPU accelerated computing. And since then, we have moved rapidly from concept to the real world network validation. So with MWC just happened recently, we demonstrated a lot of use cases on the AI RAN and also announced partnership with more than 10 operators around the world.

(03:08):
So it's all running with any RAN software from a Nokia standpoint. It's on the GPU accelerated computing. So what we really see is it's moving beyond purely concept into reality in terms of the network implementation, and that's exactly what we are preparing for in 2026. And this is redefining how the networks are going to be to build a AI native CHG network for the future.

Guy Daniels, TelecomTV (03:37):
Actually, that's great. So we're moving beyond concept. We're starting to see valuations come up there now. Francis, let me come across to you. And from your perspective, when you can oversee the whole industry, how do you see the current status of AI RAN?

Francis Haysom, Appledore Research (03:53):
I think it's worth setting this in the wider context of AI in Telco anyway. We've just recently done a survey of a number of vendors and operators. And our analysis at the moment is that we're still at a very early stage of the adoption of things like GenAI and Agentic AI internetworks in this situation. I think it's also important to say there is a lot of AI already in networks. Typically, the sort of machine learning models even embedded in RAND today, and that will form a basis for this. But it's really distressed. We're really at an early stage of AI RAN as an enabler of more than just bespoke proprietary solutions. Things are really at the early stage of proof of concept trials at this area, but there's a great opportunity to really build on that one, but we're really still at an early stage.

Guy Daniels, TelecomTV (05:02):
Thanks, Francis. Yes, we're at an early stage. Interest is off the chart though, isn't it? Vihang, what stage are you at? Where do you see us today?

Vihang Kamble, Rakuten Symphony (05:15):
Yes, thanks. Thanks, Karen. Yes, we are seeing that AIN is becoming the integral part of the telecom networks, and the early results are really showing positive trends. And in terms of the AIN for the non-real-time features, we believe that the RAN intelligence-based system, the rig-based system has matured. And we can see interesting our apps being developed by third party as well as us. And to give one real check in Rakuten Mobile, we showed that we could show energy efficiency gains of 25% by having AIN on the rig-based platform, and that's a significant gain. So that's reality. With respect to AIRAN on the edge compute node where essentially the real-time applications are getting developed, I think a clear picture is emerging. And of course there is a debate between CPU and GPU. And what we believe is even though most of the ARN problems need training on GPUs, but the inference could be done on the CPUs, and especially with the new Intel servers like the GNRD base, the Granite Rapid servers coming from Intel.

(06:31):
There are inference engines, there are advanced metrics, multiplication capability, there is floating point optimization. So those have to be tapped first before we move to the GPU. So that's our take so far.

Guy Daniels, TelecomTV (06:45):
Thanks very much, Fijon. And Ignavio, let's come across to you if I can. Are you seeing the telcos, your customers and partners, are you seeing them express more than just the passing interest? Are they starting to commit?

Ignacio Gonzalez, Red Hat (07:00):
We have seen in 2025, I mean, in 2024, some interest coming from Asia. If you remember, SoftBank was the first customer really great enough to trial GPUs for RAN and other workloads, AI application workloads altogether. So we already passed this time in terms of if it's going to work or not. But we have been busy in 2025, and we are currently now, it's doing the technical validation to make sure that no matter if you have GPUs or you have CPUs or other type of computing capacity underneath, everything is going to work through this cloud platform. That's what we're trying to bring here. And no matter what it's going to be these infrastructure requirements in the future, we have to make sure that it's going to work over this techno cloud platform. That's our main focus actually in 2026.

Guy Daniels, TelecomTV (07:55):
Great. And it's good to hear that technical validation is ongoing, which leads me quite nicely to my next question because I like to ask what measurable gains that operators should realistically expect with AI RAN. I'm thinking of spectrum efficiency, energy savings, automations, a whole range of potential gains that they could look for. Vihong, what is the realistic benefits for an operator to go and look at AI RAN?

Vihang Kamble, Rakuten Symphony (08:28):
Witnessing significant benefits if ARN is applied correctly. I think there's a caveat there. And in MWC, we showed some of the benefits. As I said, the energy savings of 25% was demonstrated in Rakuten mobile. And let me focus more on the ARN for the edge compute mode where the ARN natively decides on the DU software. And there, essentially, then the ARN could be applied in the uplink receiver or scheduler in the downlink or the rate adaptation features. And another prototype which we demonstrated in MWC was using ARN for the uplink receiver design. And what we showed is using the traditional receiver for users which were in really poor channel condition, 10% of the time we were not even able to detect the signal while when we applied ARN, that we reduced 1%. So essentially the error rate in uplink came down from 10% to 1%.

(09:30):
And for certain applications like VONR, that could be significant. And what we showed is by doing this optimization, the voice quality from very poor quality, it became very good. So that's something which can be easily measured. And the inference for this scheme was well within 100 microsecond, which makes it product quality. So again, the key is to understand what are the limitations of the traditional signal processing-based schemes and judiciously apply ARN there. Another use case was the downlink rate adaptation where in very fast-fading environment using the traditional scheme, we were seeing 20% error rate in the downlay. When we applied AI and there, we were able to meet the target error rate of 10%, which helps in user experience by reducing the throughput fluctuation. So that's another data point for you.

Guy Daniels, TelecomTV (10:26):
Thank you very much. And the more data points we get, the better it is for operators, obviously. And Francis, I'll come to you in a moment, but Aji, I want to come to you next if I can. Every week, I'm seeing dozens of AI ran research papers. There's so much work going on in this area. It's progressing daily, isn't it? But what should operators be looking for now in terms of or in realistic timeframes for benefits and gains?

Aji Ed, Nokia (10:54):
Well, I like Viha mentioned, so there are multiple improvements that we expect or that we see with the AIRN in the network. So one is about the speculal efficiency improvements. I call it as AI4RAN. AI4RUN is like it has two parts. One is for specular efficiency to how to improve RAN network performance using AI. And the second part is how to improve the operational efficiency of the RAN network, which is two parts to it. One is using the RAPs and edSAPs, which are inherently using the AI-enabled AI-enabled algorithms that we can really improve the network automation, which is, for instance, we have deployed it in multiple network congested networks, which improves the spell utilization and spectrum utilization by 10 to 15%. And at the same time, we are also able to really bring more than 15,000 autonomous operations using AI into highly congested network. So that is one part of the equation.

(12:06):
So when it comes to the spectal efficiency from the RAN perspective, that is where I see the next set of benefits coming in. And let's say if you talk about the GPU accelerated computing, that provides different set of benefits. When it talk about panel computing capabilities of the GPUs, because you can use it for ... Inherently, there is a lot of matters multiplications and advanced algorithms that you can run on the GPU. The second, you can use a lot of machine learning models. Of course, you can use the machine learning models on the custom silicon or the traditional systems, but the GPU accelerated computing provides much bigger benefits in terms of the compute possibilities for the machine learning models. And the third would be the advanced new algorithms that we can bring in for Beamforming. So channel estimation, BIM forming all of these different areas we can deploy machine learning models, and we demonstrated such use cases for MWC with the channel estimation, with the multi-user MIMO pairing, and all of these different use cases we have demonstrated using AIRN.

(13:18):
So all of these are improving the speculal efficiency, which provide compounding benefits to the operators from a network standpoint. So this is precisely where I see the real benefits coming in. And on top of it, with the AI RAN, you can innovate at the speed of the software so that at the pace of software, we can bring the innovation. You are no longer dependent on the hardware cycle refresh, but it is dependent on the software. So this is really another benefit which comes in. And last but not the least, of course, this provides new revenue opportunities or new service opportunities for the operators for the future, for the GPU as a service, or we talk about new services like ISAC or physical AI, which is in the future. So all of these are real benefits that will give to operators in the future.

Guy Daniels, TelecomTV (14:17):
Actually, thanks very much indeed. And we'll come onto the service opportunities a little bit later. But Francis, let's pick up on that because we do see an awful lot of innovation. There's a lot of innovative ideas and developments in this sector at the moment. But how is all this translation? Because operators are looking for their efficiencies, they are looking for the savings. How does it play into their requirements?

Francis Haysom, Appledore Research (14:43):
I think the first thing is to really say whether we're looking at spectral efficiency, whether we are looking at energy efficiency. We already have some fairly strong references for that one. You've got companies like Kahir, for example, that already saying you can get two times spectral efficiency in congested cells, for example. Okay. You've got AIRO with Mavenir late last year, which are talking about sort of spectral efficiency of 35% from their near real time Rick. In energy saving, an example actually I just picked up today actually was in a private network, a 64% saving in a Taiwanese private network. Now, I think all of those give you an example of the opportunity, and that is really with the current Open RAN and VRAN, cloud RAN initiatives, for example. But the key thing I think the AI brings in is that ability to innovate, but more importantly, fit to your existing organization.

(15:49):
It's one thing you can get good special efficiency for one particular private network in one particular case, but AI allows you to deliver that in multiple variants to deliver the benefit based on your organization. And similarly with automation, Raklam Mobile, for example, is making great leaps in terms of automating its platform, but AI has the opportunity to put on top of that to match it to existing organizations, existing networking structures as well. So I think the opportunity, the AI, the sort of targets, the opportunities there, we already know the AI makes them more applicable and more easily applicable to different organizations.

Guy Daniels, TelecomTV (16:34):
Great. Thanks very much for those insights, Francis. Ignacio, I'll come across to you next because I'd like to pick up on what Aji was saying about services. And I'd like to ask, should operators be looking beyond efficiency, for example, and view AI RAN as a means to new service opportunities?

Ignacio Gonzalez, Red Hat (16:54):
Indeed. I think tech operators, they should check AI RAN as a revenue platform and not just an efficient tool. So it's great all of the efficiencies in terms of a spectrum, in terms of energy savings. But at the end of the day, this is not about being the first able to do a new call as we have seen in the past, be the first in testing new technologies. I think this time, the goal is different. The goal is how you can leverage from a monetization point of view, the infrastructure you want to deploy. And I want to be very clear here. And this is a debate that we have seen in the mobile world Congress. What is first, deploying AIRM or deploying AI capabilities at the edge in order to enable other use cases that in the next year will be needed there. And it was mentioned before the physical AI or all of the extra capabilities that they're going to be needed at the edge in order to implement new use cases.

(18:01):
I think for me, the most important thing is Telcos, they need to be focused on this. They need to be more on the monetization of RAN AI or AI itself, rather than just deploy of AI RAN for the sake of AI RAN. But we have seen, for instance, with SoftBank, that's exactly the approach. They found first three use cases to deploy at the edge due to a latency, be some kind of level of sovereignty, let me call it, because they want to have some certain data privacy. And these three use cases, they leverage and they need to have brand connectivity there. So at the end, what SOPAN did, and to me, that's the main focus here is how they can combine both over the same platform to help each other, to enable, thanks to connectivity, 5G, 6G, new HAI, and how AI can help to manage and operate AI run.

(19:02):
To me, these two have to be together. And as I said, this is not about technology. From my point of view, it has to be about monetization of the edge. Telcos, they have a fantastic topology. They have a fantastic network beyond the data centers. I'm talking about the RAM sites, I'm talking about the concentration site, I'm talking about central offices that they can leverage in order to implement these AI capabilities that they're going to be needed. So answer your question with one answer, yes, indeed they're going to need that.

Guy Daniels, TelecomTV (19:35):
Very positive. Thanks very much. And it's an interesting break from what's happened before. Francis, I'll come back to you because I know this is an area that really interests you. Is this new? Is there a real opportunity now for operators to look at monetizing RAN and the Associated Edge, which I've talked about for a long time, but is this now a real opportunity?

Francis Haysom, Appledore Research (20:01):
Yes, absolutely. I think there's a huge opportunity, but in many senses, that opportunity comes with some caveats. The first is really that Telco learns from its history in terms of not making the same mistakes it's made before. If you look at AI RAN, for example, it's really a number of initiatives. There's AI for RAN, which is about making a better RAN, making better coverage. That gives opportunities for monetization, better coverage, better capacity means potential to make more money from existing services and value added services on the network that we have today. There's AI and RAN, which is really an infrastructure thing, which is really, if we look back at the history of Telco, we've gone into the sort of data center with connectivity before. In the 2000s, for example, a number of operators and a number of vendors invested quite heavily in data center. The challenge there was it was very easily disaggregated.

(21:07):
So if we're going to make money from AI and RAN, that needs to be built on more stickiness and the inability to disaggregate that capability because as we saw with the cloud profile, hyperscale cloud providers, they provided that capability much easier and you could buy the two things separately. So addressing that need is key for both vendors in this area and in terms of the network operators themselves. And finally, AI on RAN, we're talking a lot about placing AI workloads at the edge again, but in many senses, it's very much the similar argument to the MEC argument over a decade ago. And again, if we look at the far edge, the far edge is a very difficult place to win. We did a survey of this, and if you look at it, almost all the use cases for the far edge go either to the public cloud because they don't have a high latency requirement, or they go to on- premise.

(22:09):
And you need more than just simply about latency. You need something which is stickiness. You need to look at things like, for example, a cheaper field engineering. You're substituting for field engineers in the enterprise. You're making innovation at the edge, maybe geographically, socially connected networks, more easy for people to do. It's all those things that will make the difference for putting applications on a better development experience in that area so that you're not just simply relying on something. I could build it myself in the enterprise or I could put it on the public cloud.

Guy Daniels, TelecomTV (22:50):
Francis, thank you very much. And Vihang, I'll come across to you now because earlier in this conversation you did reference edge AI. Is this now a real opportunity with AI RAN for operators to realistic look at new service opportunities?

Vihang Kamble, Rakuten Symphony (23:06):
Yes. In my view, improving efficiency and new business opportunities that go hand in hand. And one example could be in 6G, if you see the sensing is going to be one prime application, and sensing is very heavy on upling processing. So if AIN could be used to handle that processing, then of course sensing could become a viable product. The other application could be in non-terrestrial networks, what we have seen is that the upling signal quality is really poor, mainly because of the power limitations of the UE, the path loss, and the frequency and the timing errors which creep into. But if ARN could be used to improve the uplink signal channel quality, then again, non-terrestrial network could become a commercially successful product in 6G. Now, so the key is we believe that the ARN should be part of the baseband software itself. So these spectral efficiency or these efficiency gains are just the early dividends, but the real benefit is you are ready for the next generation of the telecom networks where your RAN is programmable.

(24:16):
And I think that's the main advantage of integrating ARN within the base PAN.

Guy Daniels, TelecomTV (24:21):
Thank you. And Aji, we'll come across to you for your comments on this.

Aji Ed, Nokia (24:27):
Yeah, absolutely. I think 100% agree with the comments what was made earlier. So the RAN, we need to make it programmable and it is software defined, and that's the beauty of this AI and AT architecture, what we are defining. And when you talk about the specific applications, I would call upon one specific use case like ISAC, intelligence sensing and communication, which is we demonstrated this use case at MWC exactly what AIRN is capable of. That is the sensing computing because it requires a lot of compute functionalities from a sensing point of view using the AI algorithms to object detections and so on. So this was pretty much done inside the baseband using the AI RIN. So I think this use case is pretty much there. So I think this is going to really create new monetization opportunities for the future. And similarly, for the physical AI, if you listen to what T-Mobile US, they are referring to physical AI and kinetic tokens.

(25:35):
And these all are linked because it's all coming into the AI on- ran, how to build the AI capabilities within the distributed sell-side, distributed baseband. And this is where the physical AI comes into picture and the monetization of these tokens, that's going to play an important role for the future.

Guy Daniels, TelecomTV (25:54):
Actually, thank you very much indeed. Well, certainly a lot of capabilities being discussed for AI, Ryan. How this actually translates into business, we're going to have to wait for the operators to tell us, but we'll be there when they do. Francis, if I can come across to you, because I want to follow up this with what actual architectural or organizational changes do these Telcos need to make if they're going to support AI-driven RAN in production scale networks?

Francis Haysom, Appledore Research (26:26):
I think the organizational changes is the key one. And the keyword here is trust. How do I trust the AI making a decision? How do I allow an enterprise to trust that their AI workflow, their inferencing is better done on a telco? So I think one of the miscalculations in telco at the moment is a sort of view that at the moment all human decision making is kind of perfect and all AI and a mistrust, which is until proven otherwise all automated decision making is not to be trusted. And we need a kind of framework and organization for making comparisons between AI, AI in perfect human decision making and imperfect AI-based decision-making process. We need a framework for trust and risk management. We need a framework for tracking decisions. When did decisions go right for us? Were the consequences of those decisions bad or didn't matter?

(27:40):
All of that becomes important information in terms of us making adjustments to AI data, AI decision-making, but it also forms the basis on which AI can be trusted because it is making better decisions more of the time than human-making decisions. So for me, it's that ability to start tracking the decisions that are being made all of the time with an organization, the consequences of those decisions, and then making decisions as to whether they're better achieved by AI or they're better achieved by humans or things that we need to change in terms of algorithms or new models in AI terminology.

Guy Daniels, TelecomTV (28:24):
Great. Thanks so much, Francis. Lots of good advice there. Aji, what else do Turcos have to be aware of? How do they need to prepare?

Aji Ed, Nokia (28:33):
See, from an architecture point of view, we believe this has to be an evolution, not an evolution. So this is the key because we have to make it as seamless as possible from an AI native network introduction point of view. And the secondly, what we believe the networks are going to be hybrid for a longer period of time. When I say hybrid, what it means is it would be a midst of distributed network and centralized network, and also distributed network with the purpose-built network and the cloudified network. So it is a mix of different type of architecture, and that's going to be the case for next few years to come. So that's the reason when we announce the product for AIN, we announced two set of products. One is based on our purpose-built options based on RAN baseband, and we provide a capacity plugin card, which is AI native capacity plugin card for the seamless integration into the existing AI scale systems across the world.

(29:40):
All the network operators who have deployed AI scale systems that can easily integrate this capacity plugin card and bring AI-native capabilities from day one. Secondly, at the same time, we provide the cloud server infrastructure, which is based on standard servers, and which is providing the AI NATU RAN capabilities using the ArcPro GPU accelerated computing mechanism. So both these options provide a seamless integration into the existing network, which will make the network ready for AI native capabilities that is 5G advanced. And with a simple software upgrade on the baseband, you can support 6G.

Guy Daniels, TelecomTV (30:22):
Great, Agi, thank you very much. Indeed, evolutionary approach there. Ignatio, let's come to you. What else or other advice for operators who are going to be looking at deploying AI RAN? How should they architect? How should they sort themselves out internally?

Ignacio Gonzalez, Red Hat (30:38):
From an architecture point of view, I think we need to find and get some foundations. I think we need to make sure that we need to have this cloud-native platform where run workloads and AI applications that both has to be based on microservices. They need to be running simultaneously. We need to have what we call this single pane of glass. Then you're going to have to think about the operational framework, because as I said, it's great to have AI for the brand, but you're going to need to have maybe AI for other part of your network. So think about a wider architecture. You need to have this end-to-end operational framework where you're going to need to have an area for the inferencing capabilities, where you're going to have to do these runtime capabilities for inferencing. You need to have another area for the AI audience that these AEMs is going to help you in order to achieve higher level of automation and autonomy.

(31:37):
I think autonomy here is a strong word that we need to use and probably evolve towards this telco network orchestration. We talk about Agentic AI a lot in 2025 and could be the final step. And of course, probably certain operators are going to need to train models. So again, we're going to need to have this platform to do the, as I said, the inferencing capability to manage the AI agents and the training models. And most importantly, and we talk about this induced of last year data, the data platforms and the drinkable AI data that we need to have. It's still very crucial. We need to make sure, again, in this foundational approach, we need to have this element also included

Guy Daniels, TelecomTV (32:21):
There. Absolutely. And we did talk about that in usodorf. And this year's for AI Native Technoform, I bet you we're going to be talking about it again. It's so, so important. Thanks for bringing that one up. We're just going to turn to audience questions in a moment. But first, we've been asking Telecom TV viewers for questions on this topic. And we'll get to one in a moment, as I said, but I do want to also remind you if you're watching live to take part in our viewer poll, if you've not yet done so, because we run these for every summit and your responses do help us to shape our ongoing coverage. Okay. Then we do have an audience question that we received in the past couple of days has come off the back of MWC, this question, I think. And let me read this question out to you.

(33:10):
It's very specific. Do you think that the substantial commercial pressure and scale that NVIDIA brings will be the deciding factor behind the adoption of AI RAN? And will it make it the defacto RAN model? Well, that's the question. Is it going to be driven by one of our new entrant friends, or will it be something else entirely? And Ignacio, can I come back to you? Did you want to pick up on some thoughts around this topic?

Ignacio Gonzalez, Red Hat (33:42):
Yeah, for sure. I think it's great to have NVIDIA pushing, but this is not just about, if we talk about AI and we talk about AI rank capabilities, this is not only just about GPUs. I think we need to be very realistic that we need to also reuse the current infrastructure that operators are currently deploying. We just talk about the Intel CM six generations that we just start to deploy there. So we need to give some credibility to the current investment that Telcos are doing there. So also, if you think about what is coming next, we don't know exactly. Look at what NVIDIA did with the growth. They are now also focusing on inferiencing capability. We know that on top of GPUs, we are going to have NPUs, LPUs. So there are other processing elements there that probably will help in the future. Right now, of course, NVIDIA is pushing a lot.

(34:37):
I don't think that we ... It's only just because NVIDIA, I think NVIDIA has a wider view and a wider goal, and they believe that the connectivity coming from telcos could help them to achieve their goals. That's the reason why I guess NVIDIA is heavily involved right now.

Guy Daniels, TelecomTV (34:53):
Fantastic. Thanks very much for responding to that view question. And Francis, have you got any additional thoughts here?

Francis Haysom, Appledore Research (35:00):
I think NVIDIA's role is very important. It gives a push to the market, but I'll come back to my comment from one of the earlier questions, which is it comes down to the business case. You can put some very expensive GPUs at the edge. You can put GPUs in RAN as it is today, but it will ultimately come down to, is that NVIDIA GPU delivering new revenue to the operator, or is that NVIDIA GPU providing a better solution than custom silicon or anything like that? It will come down to the business case. So NVIDIA's involvement is important. It's a great push to the AI RAN, but it will be a variety of business cases that will actually in the end make the difference between AI RAN being a success or AI RAN being another MEC 2.0 as it were.

Guy Daniels, TelecomTV (35:57):
Yeah, the business case. Thanks very much, Francis. Absolutely. We'll come to other two guests as well. Vihong, I'll come to you first. What are your thoughts on this?

Vihang Kamble, Rakuten Symphony (36:06):
Yes, I think it's a great thing that mighty company like NVIDIA is joining RAN, RACE, and it builds per innovation for sure. But what we need to remember is in order to be successful in RAN, the reference L1 has to be product quality. And if you look at the Intel architecture, the FlexRAN, Intel, we started using FlexRAN in seven, eight years ago. And over all these seven, eight years, we have productized it. So it's really a commercial grade L1 software now. Of course, NVIDIA has area SDK. Question is how quickly NVIDIA and the partners can make it commercial grade? That's number one. And the second question to be answered is what is the power consumption of the GPUs? At least the new generation of NVIDIA servers are showing promising results. So what we see, the trend is the CPUs are upping their inference kind of capabilities by having these advanced capabilities, while the GPUs are trying to bring down the power and trying to make themselves as a viable option in DERA.

(37:18):
Yeah, that's what the trends we are seeing.

Guy Daniels, TelecomTV (37:20):
That's great to hear, Vihan. Thank you very much for that. And Aji, of course, I'm going to come across to you as well. What's your take?

Aji Ed, Nokia (37:28):
Yeah, of course. I mean, I should answer this question for sure because we announced this partnership with NVIDIA and Nokia in October 2025. When we looked into the partnership and collaboration, key topics were obviously NVIDIA is a leader in the AI, and of course Nokia is a leader in RAM. So these two leaders coming together to bring or build AI RAN. So that is number one. But at the same time, it's very important to really understand how the telco networks or telco business case work. It has to meet the three Ps. When I say three Ps, that's the power performance and the price characteristics. And this is exactly where we started building these solutions and products based on all the different three Ps, meeting those criteria and building this product together. And so we are positioned to really bring this competitive solutions with the higher capabilities and performance aspects with the GPU accelerated computing using the ArtPro platform and Nokia software, somebody mentioned earlier, which is on top of this ArtPro platform, which is a complete RAN software using the L1 algorithms, what we have built in for many years, more than 24 years.

(38:51):
And that is inbuilt into the GPS computing with the several more algorithms which is coming on top of it, which we believe that this is a strong foundation for bringing the full AI native capabilities when we move into 5G Advanced and 6G.

Guy Daniels, TelecomTV (39:09):
Fantastic. Well, it's a great question from our audience then. Thank you everyone for answering that. That is all the time we have for this discussion. Thank you so much for taking part in our program today. And if you are watching this live as part of our future of RAN Summit, then please do stay with us. Don't go away. There's plenty more to come. We have two further panels for you plus our live wrap-up show, and you can find the full agenda and speaker details on the telecom TV website. If you're watching this on demand though, you'll find links to all of the panels and programs from the summit on the homepage on telecom TV. But now though, thanks for watching and goodbye.