Newsfilter article unlocker

Hello, and welcome to today's webinar, Monetizing AI-RAN, Turning Next-Generation Networks into AI-Powered Revenue. Now, AI-RAN is usually presented as a way to make the radio network smarter, using AI to improve performance, automate operations, and squeeze more capacity out of existing infrastructure. But I think the more interesting question is whether that's actually the main business case, because once operators start deploying accelerated compute into the RAM, this becomes about much more than improving the network. It becomes a bet that telecom operators can turn their distributed infrastructure into an AI platform, hosting inference, supporting enterprise applications, and ultimately providing the compute layer for things like factories, robots, vehicles, and all sorts of other physical AI systems. And when that happens, the RAN stops being simply a cost center. It becomes one layer of a distributed AI infrastructure stack, which is providing an essential component of the entire new global AI economy. Of course, that raises some interesting questions, doesn't it? Who pays for the compute? Can operators and carriers keep utilization high enough to justify that investment? Are enterprises actually willing to buy these services from carriers? And do the economics work before we get to 6G? Well, that's what we're going to explore today. And joining me are top executives from three companies, all of which are going through an extraordinary period of AI-driven success. We have Pallavi Mahajan, CTO and Chief AI Officer of Nokia, one of the world's big five telecom leaders. Hi, Pallawi. Hi, Steve. Good to join you. It's great to have you here. We have Soma Velayutham, Vice President, AI and Telecoms from NVIDIA, which is currently the largest company in the world by market capitalization. And we have Ryuji Wakakawa, Senior Vice President and CTO, SoftBank Corporation, which is a global pioneer of AI RAN. Let's start with a question for all of you. Hi, welcome. Let's start with a question for all of you, which sounds like a simple question, but I suspect that the answer may be more nuanced. What is the first genuinely bankable AI-ran business case? Is it lower operating costs? Is it better use of spectrum? Or is it actually new revenue from some sort of new AI service? Pallavi, let's start. Let's talk with you. What's your thoughts on this? So, Steve, let me actually first step back and talk about, you know, how we at Nokia look at AI RAN. I think, you Know, the very first important pillar for us is that, you Know, we want to go about and have bring in AI to go out and make RAN more Efficient. And this is where, you Know, We are looking at a part to go About and get to 2x spectral efficiency. Because I think you know that's that's the most fundamental problem that we have to solve. If you look into the AI traffic, the unpredictable nature, the bursty nature of it, we need to make sure that our telco operators are ready to go about and meet that growing traffic demand. I think the second thing for us is AI RAN means that we are moving innovation at the speed of software. We don't want to wait for the next G2 happen for innovation to come in. In fact, like, you know, this is what I always keep on talking about, that for us, you know, RAN now becomes a platform, and 6G is just a software upgrade. And, you know, the third and the most important thing for us from an AI RAN point of view is to make this platform open, programmable, extensible, and secure. And I can tell you, and you know we can all go back and look into history, whenever you make solutions which are open, programmable, extensible, you know, there are a lot more smarter people in the ecosystem who can come in and contribute. There are use cases that, you Know, even we are not aware of, like, you Know, yes, we are talking of physical AI, but, you Know, what are the kind of use cases That will come in from physical AI? You Know, What are the different applications that need to run along with your baseband the possibility of that is not you know we don't know about it but making sure that the platform provides you the capability to make that happen that's that's a that's a fundamental thing in our thinking so you know just to summarize like you know make RAN more efficient the 2x spectral efficiency the second piece is like you know move at the speed of software make it a platform and the third is open programmable extensible that allows you to bring in different use cases that we can't even imagine today i mean that that's a sort of uh if we build it they will come uh philosophy and i think obviously that's that's very true particularly when we look at some of the use cases around 6g i mean it really enables uh things which we haven't even considered being possible before um soma from uh nvidia what's your take on this i mean what's the what's strongest business case right now for carriers looking at ai ran yeah i would i would say um similar to what was palavi said three things one is efficiency is very important rand continues needs to continue to be very very efficient second is utilization as you know systems rand systems are always designed for peak like electricity grid and by definition when you design something for p you know you have fellow capacity most of the time so utilization is very important as you move from a electricity utility kind of approach to a data center approach utility becomes very important and third is really delivering ai at a planet scale as we call it and being able to leverage this compute to deliver ai for consumers the last few years has been predominantly building frontier models the big open ais of the world but now we are seeing that ai is moving to a consumption model with a lot of inferencing and nobody can do this better than delivering ai to consumers and enterprises than the telecom networks yeah i mean that is just as an aside i mean nvidia has positioned itself uniquely amongst the hyperscale technology companies as the only one really that has grasped this truth, which is that AI doesn't live in an AI factory. AI is going to live out beyond even the traditional border that has been defined as a telecom edge. I mean, it's going to move outwards and that presents an extraordinary opportunity. And I give NVIDIA a lot of credit for not just trousering trillions of dollars from selling GPUs into data centers, but actually trying to push the boundary out into the real physical world of AI as well. That's the grid that we talk about, right? Yeah. I mean, that's absolutely the future. And in financial terms, it's a market which is so vast, it dwarfs anything that we think of as cloud or enterprise or anything like that at the moment. Because once you start attaching AI to the physical economy, you're talking about, you know, 85% of the global GDP being transformed. So it's a little, we're in the early stages, aren't we, of AI at the moment with LLMs and generative and things like that. But the future is a global economy running on AI from one end to the other. And that's why I think that this event today is so important, because AI RAN is where, you know, AI is not going anywhere without RAN. Ryuji, you are from SoftBank. You are actually doing this. I mean, you're a pioneer of AI R AN. Why? What are you using it for? What's the business case? Yes. Thanks, Steve. So first of all, like we have to really understand the mobile business is kind of commoditized these days. So revenue is saturated and our mobile revenue is declining. This is a fact. and also like we have to keep investing in the infrastructure because the mobile is getting essential for our daily life you know the disconnections of failure and coverage that's not the case for people today so we keep like investing in infrastructure but the revenue is declining so how we can sustain this infrastructure that's a like as a question we we share with SOMA like back in, you know, a few years ago. And that's how we start AIR-Run. So AIR Run is very disruptive approach for the infrastructure deployment from like a cost center to the revenue generator. As Steve said, like this is no longer cost center. And the creating revenue stream is very essential. New revenue stream is essential for our industry. And if like our infrastructure still being the cost center you know what the operator will do is a cost reductions you know tco reductions that's we did like a past 20 years but what we did is actually swapping the vendor you know switching the different technology but the costs still stay safe so and also like because of like a lack of the uh you know investment there's no strong innovations to be honest like after 4g um you know our industry like innovation speed is not really like uh um you knows amazing and it's a kind of stall the investment is still so that's why like you know we want to completely change this like situations using the arm so airline is not really like a technology aspect but also like economical and the business side of the you know phenomenon and so in order to get the ai revenue stream by ai mobile infrastructure this will be the game changer so you know this business case is still is is there because uh look at the ai uh you know there are a lot of like business case available in ai's ai uh business ai industry and this air run is air infrastructure that's where we try to capture that business opportunity by telecom so that's how like we see the air run i mean what you're really describing is a reinvention of telecom itself though isn't it i mean you're saying the old ways are not or not enough anymore we have to become part of this digital biome which is driven by ai and obviously with your partners here you have sort of the three the three legs of the stool in order to do that um do you worry that there uh that uh that ai might not create enough new traffic but also critically enough new revenue to materially change the ran investment equation quickly enough or or or do you think that the timing is going to be okay i mean that's obviously a question which softbank must have mulled But I think that's a good question. But look at the, you know, the AI business today. AI is happening. AI is still like being prepared. So, you Know, what Teleco has to do is preparing the infrastructure to serve the AI influence on the edge. So our priority is completing the AI run deployment immediately. Because otherwise, our revenue opportunity of AI is gone. So the timing is super important. So that's like, you know, we are very, very excited that Nokia, you know, released an AI-run product. Tier 1 operator, a Tier 1 vendor, finally come with the product, which is a very, muy strong, like, message to the market. And Steve, go ahead. So, you know, I just wanted to say that I get the question about, you know, is this an opportunity that is waiting to happen? The way I'll talk about it is, first of all, this is not an opportunity that is awaiting to happen. You know, you look at like what Generative AI did, right? It's completely changed the landscape from, you know, the uplink, downlink traffic. Traditionally, you know, it was more about downlink traffic, 12 is to 1, down link to uplink. Now it's like 4 is to one um and then you know with agentic you know we have like there's no concept of busy day you know typically you know all all these networks were planned from a busy day point of view uh there's no busy day it's all 24 7 because agentic is like 24 7 and the world that is knocking in front of us which is which is physical ai i think you know i think we need to be ready and the big thing that I feel is that this is an opportunity for us to go about and look at the network, not just as a cost center. I call it as the network not just has a dumb pipe, but as the Network that actually becomes the central nervous system of this grid that we are talking about. And and you know, for that, we have to be ready today. Because I feel like you know if we don't do that we'll be missing the opportunity? Well, you know, Pallavi, I mean, just to be clear, I couldn't agree with you more. I mean if we are not ready in the carrier industry particularly to embrace this new opportunity, somebody else will do it. But also not just in terms of industry segments or verticals, but also on a sort of a national level and an international level and a geopolitical level, There has to be investment. And what I'm seeing in the telecom industry at the moment, driven by your good selves, is this complete shift away from let's wait and see or how are we going to, when's the last moment we could possibly do this towards a much more aggressive posture. And I see this at Nokia. I mean, it's a complete transformation of the culture there and also, obviously, NVIDIA and SoftBank as well. And I think that that's I think you're exactly right. We have to build it. It's coming and we have to be ready. But, Soman, let me ask you, you know, the GPU. Wow. It's become sort of the unit of currency for the global digital economy. it's also become the building block and now it's moving out to the edge of the network but it requires you know as part of ai ran it it requires expensive compute there's additional power there's cooling i mean how much network improvement do operators need before the economics start to work for them? I mean, how should they gauge that in terms of rollout? Yeah, let me first start with the previous point. You know, this new economy is called the token economics, right? Everything's about tokens. And wireless networks, as Pallavi said, is no longer a connectivity platform, just connectivity. It's actually the best medium to deliver the tokens to 15 billion connected devices and probably another hundred billion going to come from physical devices and so on right there's going to be a mass massive growth of connected devices both from robots moving devices and IOTs and so on so telecom and wireless operators and wireless economy is sitting in this dawn of moving into delivering tokens right this is a unique opportunity and we need to, we are very excited about it. As we say, you know, there's no better time to be in telecoms right now. This is a good time. Absolutely. To your point in terms of power and economics and what does it make sense? I would say it already makes sense now. It's already made sense before. If you look at the way NVIDIA platforms are built, the AI platforms, we have, we actually write on what we call a CUDA. CUDA is the software that we write upon. And the software is actually abstracted from the underlying hardware. And we have many options for hardware depending on the workload, depending on, you know, the performance and power requirements. So we have GPUs that are as low as three watts, literally three watts tiny little three watts and, you now, $299. I think that's the lowest cost GPUs you can buy, system, full systems. So you have a GPU-accelerated AI computer as low as three watts, or you can have as high as, you know, thousands of watts, right? But you write on the same software. It's always writing on CUDA. So the benefit that Nokia gets is they write on CUTA once, and as Pallavi said, you now, this is backward and forward compatible. So we keep moving and they can choose the right platform. If they need to deploy in a small cell or they need to deploy on a large macro cell or if they need to deploy a large data center, all options apply. It's the same software. And this portability and flexibility is a huge gain. So the myth about NVIDIA GPUs being power-hungry or expensive is not true. It's, in fact, the most efficient way to deliver tokens. I think the RAN AI business case is kaleidoscopic. And I don't mean that as a negative. I mean, it's multi-threaded. It's diversified. There's lots of different ways to save and make money with AI RAN. And that's a good thing. One of those more existential opportunities, I think, is for carriers to monetize spare compute on GPUs for non-RAN AI workloads. Do you think, Soma, first of all, I'll ask you, is that a significant opportunity for them? Or if you're buying a GPU for $200, I mean, is that really just going to be serving that AI RAN function? Yeah. So at the foundation of it, what is NVIDIA doing? NVIDIA is building a platform that delivers the best token economics, the best tokens per watt and tokens per dollar, at the same time working with Nokia to deliver the best RAN performance on the same platform. So GBPS per watt and GBPS for dollar. So now we have one common platform, one common AI computing platform that can deliver the Best RAN economics and the Best Token economics now the and it has the agility to go back and forth you can operators such as softbank in fact they've built something called infernia and orchestrator hopefully ruiji will talk about it a bit you know they can actually shift they can do 100 rand and zero ai or they can do zero rand 100 ai and anything in between so this scalability and flexibility of the platform is is actually really groundbreaking and gives a huge opportunity to operators to scale up and down and monetize uh based on the the demand for you g you got to tell us about this is that is that a real thing now because that's amazing so you're really you're deploying something which is a like a movable feast of opportunity really you can use it you've built something which can be repurposed uh in any direction the business needs to move is that right yes so that's what we are building right now that uh thanks for gpu actually gpu is powerful enough to solve the radio software on gpu you know before gpu we need a hardware accelerations to run the you know software based like a radio software but the gpu is powerful enough it can just run on gpu without any acceleration car gpu that gpu use acceleration card by the way but the difference here is gpu is multi-purpose platform you know can support a multi-different purpose so you know we can repurpose gpu for radio platform for ai inference for other things so this is a huge And, you know, now, like, we can deploy GPU maybe for radio, maybe for AI. We don't need, you Know, strong reason to, You know, bring the GPU for radio only. Because if AI demands there, we just deploy GPU, and we can bring radio afterwards. So we have a lot of flexibility for infrastructure deployment and our CapEx, like a, You Know, plan. Wow. And also, as Soma said, we have Infineon. That's kind of like an AI platform, an operating system on the GPU. And it's a managed Kubernetes platform. So anyone can bring the AI workload onto any GPU we deploy in our infrastructure. And we also observed also like a GPU usage and where's the available GPU capacity. and our orchestrator can dynamically dispatch the AI workload to the available GPUs for. And one thing I want to highlight, the interesting things of the AI application is stateless. Some of the chat GPT, you still need your history and a RAG. That requires some of your personal data. But let's say video processing is statless. you can run this AI workload anywhere unless you have a video source. So it's easy for operator to dynamically allocate AI workload to anywhere in our GPU distributed. So this is one of the very unique technology characteristics of the AI application. That's like we can be well fit into our AI grid or AI run concept we have today. Hmm. I mean, and certainly as far as a business case, it makes it much easier because you're de-risking everything and, you know, allowing yourself lateral movement with the infrastructure which you're creating, which I think is absolutely, you know, unprecedented. Pallavi, I wanted to actually ask you, Nokia commercially launched your AI-ran platform this summer. I mean, can you tell me what does that change for operators, for carriers? What can they realistically do today that they couldn't do before? And, yeah, can can you make that live for us? Yeah. You know, let me first expand on what Ryuji was talking about in terms of distributed inference, because I think he was making a very, very unique and important point about distributed inference. a lot of people think about you know distributed inference as being driven because of latency and i think you know this is to my point about we'll see you know what use cases of physical ai need what latency and all of that stuff but it's not just latency which is which is the driver for distributed inference you look at like all the um you know the power and the cooling constraints that are there. So you now need to have like, you know, inference, which is not running in gigawatt data centers, but it has to be distributed. And then when you start looking into sovereignty, and specifically, you Know, from an agentic AI point of view, the round trip that it takes to go about and close like an agentically loop. These are all drivers for you know why we see that distributed inference is going to happen. And which is where you know any fallow capacity that you have right let's go about and utilize to that uh to your to your question steve about like you know how do we nokia see our operators using it i think you know first and foremost like you know the conversation that actually gets everyone interested and at the table is like you know when we talk about the spectral efficiency because you know for for for all our operators like ryuji was talking about like you knows capex is a big thing and you know spectral efficiency is like Like, you know, prime time real estate. And if we can go about and, you know, with the solution that we have, where we have a path of getting to 2x spectral efficiency by 2028, you know, that's absolutely like, you know, a conversation starter. The next thing that, you know, operators start to look at is, I think, you know, the, a lot of them are thinking about the fact that, you know. They have, you know, they have the real estate, they have like, you know they all own the MSOs. And the question in their mind is, we don't want to lose this opportunity by just being landlords, right? So help us go about and monetize that. And that is exactly where, if you look into it, it is not just about the real estate, the power cooling that already exists. It's not just About Connectivity. But when you look at it, the operators have the context. They understand everything about the users. And more importantly, you know, with latency-sensitive traffic, you can no longer go out and do workload placement like Ryuji was talking about, just looking at where the capacity is. But you also need to bring in network-aware application workload placement. And I think that's the opportunity that the telco operators have. So, you know, many of them, as they're looking at it, they're getting very excited about how this now gives them the flexibility of looking at the grid to start from the base station, from the MSO, all the way up to the AI factories. I think that that is an incredibly important point. And I also agree with you that that's why carriers should be very excited about this, both in terms of capabilities, but it also just comes down to geography. I mean, they're there. They've been there for 100 years. They're in place to serve this function. And without them, who? I mean unless we're talking about some huge leap forward in satellite technology or something like that i mean right now the it's not just the obvious choice it's the only choice for connecting the entire world into this new digital biome which we're all so excited about i just wonder why uh ryujin softbank are so excited about this and and and some other carriers aren't as excited is that a cultural thing ryuji um you know you're ahead of everybody at the moment is it was that just a personal decision on your part how did you convince your uh your peers at softbank to to that this that this was the way forward all right um so software is a very unique uh uh company i think unique culture we have so uh you know software started like from the dsl operations actually like originally like a master so our chairman started as a publication company and then like starting dsl and once the mobile coming you know we acquired the valise uh vodafone in japan and started the wireless business and so every time like there's a technical like a big wave coming we transform the business like core into that new business new kind of like a business now like it's ai so our Our CEO, all like a company, now try to transform entire software from mobile operator to the AI operators. That's a statement our CEO made a few years back. So in my company now, I don't have a big budget on the network infrastructure. If it's something related to AI, I can secure the budget. So I need a something AI plus network. That's AI run, actually. This is another reason to start AI run to secure the funding internally in the software. Need to be something AI related. But this culture is really like a very, you know, I appreciate because as you see, AI move is lighter in speed. and even you know we try to lead this air run initiative but you know last six months what's happening in ai is huge right so we we need to be like uh you know speed up in the entire uh you know industry mobile industry and uh hope like you know we can improve the power of the air run soon with Nokia. And so, you know, more like operator join us this journey because this ecosystem is so important. Because if the, you Know, AI grid is only available in software, none of the AI company will use this infrastructure. You know, they need a GPU every country and every operators. So hopefully, you know, we can, you know, go that level soon. Well, I think you're in the vanguard, and I think your point about AI moving quickly is very, very well made. But actually, I don't think that we've seen anything yet. I think that the wave of, the transformative wave, the impact of AI will be felt, you know, really outside of the periphery of the network that that we currently consider the edge network as as as we interconnect everything and you know you have probabilistic llms in in in ai factories in the core of the network and you have deterministic ml and ai out at the edge of the net and beyond and then everything the world just completely changes at that point and we're not even you know we're not even close to that yet i mean soma do you do you agree with that yeah video it seems like you've completely looked into the future and said hey you know we could just make this huge amounts of money from building ai factories but you've really actually targeted this expansion haven't you right so so let me give you some numbers So, I mean, the first wave of AI infrastructure has been predominantly building a centralized data center for training and maybe frontier models. The next wave is really about inferencing and distributed inferencing. And the numbers suggest that over the next five years, the largest growth is going to be in AI inferencing, 35% CAGR growth. And by 2030, the overall AI data center demand, 60% of that is going to be from distributed inferencing. So the largest growth, the largest absolute volume is by, you know, without a doubt, is going to be in distributed inferences. And as Pallavi suggested and Rishi talked about, telecom operators are in a very unique spot to move beyond connectivity to deliver this distributed inferencing at scale and there are a few reasons it makes sense for them not just an infrastructure that they compete on they compete because of proximity to the customer they compete because of sovereignty of the data there is a clear we call it the data gravity there is the cost of moving data so you know you have this distributed data centers at a planet scale. So every base station is now an AI inferencing data center. This is an amazing opportunity, you know, being able to connect, I don't know, maybe 3 million base stations out there and more with the 6G era to become a distributed inferencing data center is a huge opportunity and a very exciting opportunity. So the real prize is not Actually, I mean, optimizing the RAN, which is obviously important, but it's using distributed network compute to host inference, enterprise AI, physical AI workloads. I mean that's really where we're headed here. It's about more than just doubling the amount of available spectrum usage, which obviously is incredible as well. but there is a that's almost like tactic versus strategy which is let's change the world and i think this this ecosystem point uh that that you were making ryuji is incredibly important you know nobody can do this on their own um you have to identify kindred spirits within the technology ecosystem uh that are going to be interpolated and supported yes what were you Sorry, I didn't mean to interrupt, but I think both are important. And the reason is why I think, you know, as much as I spoke up more about the opportunity on monetization and distributed inferencing, I think the spectral efficiency is equally important. If we believe that the tokens are going to be exponentially growing and the tokens are going come from end devices being agentic tokens from computers and cell phones or physical tokens, think about the billions of devices that needs to be connected to this distributed data center. You can't connect these billions if you don't have spectral efficiency. So it's kind of like, you know, a virtuous cycle. you increase the capacity to connect, and then you increase, you know, the efficiency to compute on it. So it's kind of like you need to have both. Yeah, Pallavi, let me just come back to this, the spectral efficiency discussion, how much I mean, we're talking about 100% increase in available you know, a spectrum by 2028, which is an extraordinary figure. How much of that is actually derived from AI versus just improvements in chip technology or working on the software or massive MIMO and things like that? How does it break down? You know, I don't have a breakdown for it, but let me actually look at it, you know. Talk about three big categories that actually drive the spectral efficiency. I think, you know the first thing if you look into it, like, you A lot of what we do in RAN is L1 processing, which is very much of mathematical, compute, linear, and algebra kind of stuff, which is where when you bring in GPUs, and it's not about just bringing in GPUS. It's also with the work that NVIDIA has done in terms of bringing in the aerial stack that also provides you the real-time benefits for it. I think the first benefit automatically is that. The second is like, you know, if you look into it, like, you know traditional RAN algorithms have been very deterministic, right? They follow very fixed rules. Now, when you bring in AI models, you know, they are they are adaptive, right. They learn from real network conditions and then they can continuously go about and optimize and, you know, take decisions that were previously too complex. And, you You know, this is where, you know, we are bringing in models for link adaptation, multi-user MIMO, pairing, you know, carrier aggregation, channel estimation. The third is that, you know, this now provides us the ability, you know, now running on accelerated compute provides us the ability to bring in different algorithms. For example, you know, and I always take this example, RKHS was an algorithm that was probably written in 1931 the industry never used it because you know uh traditional custom silicon could never run it uh it has a lot of matrix computation in it but now that you start running it on gpus you know it gives you the efficiency so i always look at it like three-tiered thing um and uh you know to me it's i the thing that i'm loving right now steve is that this industry and, you know, to what Ruji was talking about, that, you know, we'd stopped innovating. Just with this whole conversation that we've started, right, like everyone is now looking at how do you go about and bring in more spectral efficiency, you know, whether it is custom silicon or it is using, you know, general purpose GPU. But I think, you know, that's the right push that we have made in this industry. Absolutely. And just in time, you You know, it's not just good timing. It's essential timing. I want to ask all of you a big picture question because telecom and communications itself is at a tipping point here. Networks no longer increase in value in proportion to capacity. I mean, that's a big, big shift, right? They increase in value in proportion to the value of the data services and applications, not just that they carry, but that they make possible, that they enable. And once AI moves into the physical economy, that value becomes essentially limitless compared to the four or eight billion or whatever it is that we make in this market at the moment. It becomes the biggest com market in the history of history. And that raises a question in my mind, who's in charge? Who's in control? Who owns the sovereignty? Is it you, Ryuji? Or is it the enterprise? Is het the industry? Is et the government? And is this a product that you can monetize, Ryuji? Because it seems like you have the trust which would enable you to be in charge of this, right? Yeah, Steve, you're right. I mean, AI gets into our daily life. AI comes to control the maybe traffic light or maybe traffic on the highway, You know, like maybe, you know, autonomous driving on the train and things, you know, then like this AI infrastructure is very, very critical. And that, you know, the sovereignty, the mobile operator is the right position, I believe, because the mobile operator business, our business is very domestic because of the spectrum. You know? we acquired a spectrum in each country and we do operation in the country. So most of the operator is a local and also like governed by someone like by government, you know, because the spectrum ownership and so on. And also like one thing like I want to highlight, you know, what one thing operator can still be part of is SLA. Everybody's laughing about like a high SLA, like a teleco is chasing. But this is going to be super, super important on the AI inference in the long run. And, you know, we have a technology. We have a know-how, knowledge, how to run the 100,000 base stations in nationwide with a higher SLI. AI inference is the same thing. You know, We have to keep all the GPU up and running, serving the AI influence, you know, almost 100% availability. Because otherwise, maybe, you know, our like infrastructure stop, maybe like a train stop or a traffic light stop. That's not the case. So I think that's where the, you Know, we have a technology, but the next question is operations. And that's somewhere the operator is very unique and strong positions. And we know how to manage like a lot of different equipment in distributed nationwide. That's our original business. So we are very, very convinced and we are very ready to serve these AI inferences in the future and hope this will limit the AI adoptions into different use cases. I think that's the future we are looking at And, you know, we are very excited on that, you know, current like AI and run AI run like deployment. Yeah. Well, with the help of your partners, it's not really the future. It's the you're making the future the present. I feel like you've reached into the future and kind of dragged it into the here and now. I mean, the conversation which I'm having with you is a conversation which i haven't yet had with a lot of carriers. But obviously, I think that that's coming. I mean, there is still a sort of feeling around, well, are we just going to be the data utility or are we going to do this? Are we goingto do that? Or, you know, how do we how dowe, youknow, create a sortof Google scale application success? And actually, it's a lot more. It's not simple, but obviously there is a huge commercial opportunity right now for those that are set up for it. I mean, Pallavi, can I just come back to you with another sort of existential topic, which is 6G? You know, we are now on the inevitable ramping up of interest in 6G, and 6G is not like anything that we've seen before in the telecom industry. Once you get 6G and AI in the same room or the same pop, that's going to be pretty extraordinary, I think. How much of the AI-ran deployment today should be viewed as a 5G business case versus, say, laying the architecture for AI-native 6G? I don't see a partitioning here at all, Steve. I mean, for us, everything that we are doing is like, you know, like I said, right, we don't want to be defining our innovation with generations. the platform that we are launching is going to be 5G ready as soon as 6G is ready the standards are ready which is like people are talking about early 2029 it's just a software upgrade for us so we don't want to draw we don' t want anymore the capability of a system to be identified with the generation on which you are running it's a platform and if you look into it Right. This is this isn't something unique that we are talking about. You know, the rest of the networking industry, especially if you look at like what happened in the data center, has already made that transition. And now we are looking at like, you know, to the wireless part of it to also start innovating at that same pace. Amazing. Unfortunately, we're almost out of time. um but simon let me let me finish with you what have you learned from this uh triumvirate this partnership that you have with uh nokia and softbank at nvidia what what has been the most exciting thing that you've uh that you're seeing or or established from this and what's the most exciting thing you're looking forward to now yeah um thanks steve for a great question so So, I mean, the concept of going beyond being a dump pipe provider into being part of the computing infrastructure is not a new concept. The operators did try this in the cloud computing era. And there are a couple of things that now, of course, it's the AI computing era and what's different. What's different is the technology maturity of the RAN, thanks to leadership in Nokia, they can deliver a highly efficient software-defined RAN in a system. So that softwareization of RAN has changed that, number one. Number two, the computing infrastructure, the AI computing infrastructure is much more democratic than it used to be, because the cloud computing was basically only a few global titans could actually manage it and scale it. But if you look at one thing that NVIDIA is doing in the AI competing side, again, it touches upon why regional operators can maintain sovereignty, but at the same time can also achieve that economies of scale for the technology. They couldn't do that in the cloud computing platform. Here, NVIDIA is doing a lot of work to make sure that it is democratic and it's at scale for every operator in the world. What are we doing? First of all, we are investing heavily in open models. We open the models as well as the weight and size and everything. So the availability of open models that operators can compete at the hyperscalers. The second thing is the maturity of something like a Kubernetes. And how much it is, you know, come a long way. So the technology maturity of RAN, softwareization of RAND, the technology majority of Kubernetes, the investments that NVIDIA is making to make AI open and sovereign, and then adding that the demand that's coming from agentic AI and physical AI. this combination like i said it's uh i've been in telecoms 30 years but this is i just feel like we are getting started this is the best time for us to be in telecons it's good to be you it's Good to be telecom now i'm really delighted because i've been doing this i've Been working in telecon since before there was an internet and uh it you know sometimes it's been dispiriting right you know everybody's disappointed with something or the other. And I think it's time for telecom to wake up and get really excited about its future. And obviously that's what the three of you and your companies are doing together. And I couldn't be happier about it. I think you guys are a credit to the industry. So I really- I have to- Yeah, go ahead, Balog. I just have to say this, Steve, that you never get so many opportunities in your lifetime to go disrupt an industry. And disruption does not happen by just one person or one company going about and making a pitch, I think, you know, this is where collaboration, partnership, all of that matters. And what you're seeing here in the room, you know, many of us have gotten a few gray hair in the whole process. But, you Know, that's the whole journey that we are on. Because I think you know like we all said, right? This is going to be networks are going to become the central nervous system. And we need to go about and make them open, programmable, extensible, and secure. Yeah. And this is happening right now. And as Ryuji, as you said, I mean, it's happening real fast. So I'm just glad to be able to be on the wall for all of this. Guys, thank you so much. I hope we get to talk. Yeah, it really is. Thank you, all of you. Thank you.