AI-Native RAN: From Innovation to Deployment
OBJECTIVE:
AI is moving from an emerging capability within the telecom network to a fundamental component of how future radio access networks will be designed, optimized, and operated. As the industry moves toward AI-native RAN, the critical question is no longer whether AI can improve network performance, but how operators, technology providers, and policymakers can move these innovations from trials and demonstrations into scalable, secure, and commercially viable deployments.
This session brings together industry and government perspectives to examine what it will take to accelerate the transition from AI-RAN innovation to real-world deployment. The discussion will explore how AI can be integrated directly into RAN architecture and infrastructure, the role of AI and accelerated computing, automation and intelligent network optimization, energy efficiency, network performance, and the evolving requirements for secure and interoperable AI-native telecommunications networks.
SPEAKERS:
Amanda Toman - Director, Public Wireless Innovation Fund, NTIA
Anand Chandrasekher - Founder, CEO, Aira Technologies
transcription
ABE: And with the NTIA's new solutions for the AI-native RAN funding opportunity, the U.S. government is placing a significant emphasis on advancing AI-native wireless technologies that to strengthen a domestic and commercially viable technology ecosystem.
This panel will examine the path from proof of concept to production and how public and private investment can really help accelerate the development and deployment of the next generation U.S. wireless infrastructure.
Joining this discussion are Amanda Thoman, she's director of the Wireless Innovation Fund that's at the NTIA's Department of Commerce. And we also have Anand Chandrasekar, he's founder and chief executive officer of Aira Technologies and panelists, welcome.
AMANDA: Thank you.
ABE: Anand, I want to start with you, if you don't mind. what is really fundamentally changing when AI becomes this really integral part of the RAN architecture, rather than simply another application layered onto an existing network? So what does a truly AI-native RAN look like?
ANAND: So I'll start with a little bit of a context of why AI-native RAN is important, and then I'll talk a little bit about how AI-native RAN actually helps fundamentally, right? So if you think about the networks today, and primarily I'm going to talk about cellular networks, right? There's a few billion connections worldwide that are occurring on a cellular network, right? If you branch predict a few years from today, that's likely to be a few trillion connections, right? Rather than a few billion connections. That's about an order of magnitude increase in just the number of connections, right? If you look at the technologies that these networks have to interface with, today already is 3G, 4G, 5G, soon to be 6G, and plus satellite, right? So complexity is also increasing in terms of the number of types of transports that are being dealt with in these networks, right? All of that is happening with the backdrop from an operation standpoint, where human headcount in these networks is actually flat to down, right? Largely because of financial pressures that every single operator around the globe is facing, right? Because the revenue per unit is going down, the traffic is going up, the complexity is going up, the number of connections are going up. So as traffic goes up, complexity goes up, and headcount goes down, you have to apply technology to be able to solve it. And this is where I think AI Native RAN comes in, right? So what does AI Native RAN fundamentally bring to bear? I think AI Native RAN enables a fully autonomous network. In the parlance of TM Forum, Level 4 Autonomous Networks, right? With a Level 4 Autonomous Network, what you effectively have is the equivalence of a fully automated network, which is always on from an operations standpoint, a network is always on, but the human component of management of the network is not always on. But when you get to a fully autonomous network, that component is also always on, and it's instantaneous, right? And it's correcting itself.
Those are fundamental breakthroughs in terms of how a network gets managed and operated, and it yields efficiencies at scale, which are extraordinarily important, given the increase in complexity that I talked about at the outstart.
ABE: Anand, I'm going to stay with you, and certainly I wanted to get Amanda's input on this, insight on this as well.
Let's talk about barriers, really barriers to entry, preventing today's AI RAN innovations from really moving from these trials that I mentioned in the aforementioned introduction, to these proof of concepts, into these large-scale commercial deployments. And certainly there's a public and private sector component to this, but Anand, I'll start with you.
ANAND: Sure. There may be several barriers. I'll talk about the one that I think is the biggest in terms of applying AI natively into the RAN, right? And that big barrier, in my view, is belief, right? Belief, and that stems from two standpoints, right? Belief in whether, can AI actually do the job in a network? Remember, these networks are five nines reliability, so they have to be up all the time. They're critical components of our infrastructure, right? And can you actually let AI run components of the network and still have it be five nines? All of that is bundling under the notion of belief needs to exist that the software, the AI, can actually do the job, right? There's two components to that in terms of how you can knock off that belief or knock that barrier and reduce it.
One is data. The more network data that becomes available, the more proof points that are delivered, whether they are through POCs, pilots, whatever, that increase based on real world data showcase that, yes, AI can do the job. There's an empirical element to addressing the concerns from a belief standpoint.
The second is psychological, and that psychological is just the fear of the unknown, right? It's human tendency, right? That second component, which is psychological, I think the way to address that is time. And the way to address that time component is actually put some of these solutions into the network. Once they've been proven out in production pilot environments, put them into the network with human governance around it, so supervised environments, right, until you're ready to fully embrace the autonomous environment. I think the biggest barrier that I see is really belief and approaches to address it, but I think that's, in my view, that's the biggest barrier.
ABE: That's interesting. Amanda, barriers to AI innovation from your perspective?
AMANDA: Yeah. I agree with Anand, but what I would add to that is I agree there needs to be belief. There needs to be demonstration capability to demonstrate the reliability to a potential operator, whoever that is, enterprise, mobile network operator, or just connectivity operator. But I think the business case is what's been missing, right? So I mean, we lived through this with Open RAN, and the push where we, from a US government policy perspective, really wanted to push Open RAN to try and drive that commercial adoption was really challenging. And I think the biggest challenge was around, which I think is going to be the same thing for this AI-native network architecture, is a demonstration of how it can make an entity, an enterprise, whatever it might be, how it can make them money. At the end of the day, if what we currently have isn't broken, I don't know if there's an impetus just to say, you know, there has to be a real demonstration. And I think CapEx or, sorry, OpEx savings is not necessarily going to be enough to do it. There's going to have to be a demonstration of new revenue streams, whatever it might be, but really facilitating that demonstration of belief to whoever that is, that in doing so, we'll actually get a return, a big return on their investment by adopting that new platform.
ABE: Interesting. I wanted to continue on the topic of ROI, certainly an overarching concern for industries sort of across the board. Anand, I'll go back to you. What use cases are really most likely to deliver this measurable near-term value for these operators? And I'll go through a short list here. Network optimization, energy efficiency, automation, capacity management, spectrum utilization, predictive maintenance or other. And what metrics should operators really be using to determine whether AI-RAN is delivering real ROI now or in the near future?
ANAND: Yeah, I think, so first of all, overall, all of these are great candidates for effectively having AI-Native-RAN kind of deliver substantial improvement. When I'm saying substantial improvement, I'm not talking single-digit percentages. I'm talking about double-digit percentages improvement, right? I'll stick in the operations domain since I started off in the operations domain. I'll stick in that same domain to kind of illustrate some of these a little bit, right? We've done, without taking any customer names, we've done a couple of large-scale pilots from a production standpoint in the operations domain. And what we have shown in two cases, two very different operators, one domestic, one overseas, what we've been able to show is easily a 35 percent plus improvement in efficiency or productivity, however you want to classify that, right? So, for example, takes a human, on average, a very competent human, a very competent RF engineer, takes a human somewhere between 30 minutes and maybe over an hour to root cause an issue associated with a network, right? A lot of time gets spent on these kind of root cause analyses in a network, right? What we've demonstrated with our software in network in a production environment is it takes our software two seconds and that's it, right? And it can be any node in the network. It'll give you an answer if something's wrong with it within two seconds, right? Now, that's a substantial change. I said 35 percent improvement. Well, half an hour to one hour to seconds is way more than that. But you're not spending your entire eight hour or 10 hour day doing root cause analysis. So that's the reason I temporized it a little bit and said it's 35 percent.
It could be higher than that, right?
You can also use this to effectively improve payload, right? And if you can do that, and that's the other pilot that we're doing with a customer, that's direct revenue in the pockets of the operator because we're improving the payload on their entire network, right? So you can use AI to both reduce costs substantially, but you can also use it to improve revenue, to Amanda's point, right? And both of these are actually production pilots. These are not experiments. These are not in lab environments. And as more of these kind of cases get accumulated, this goes directly to my belief point. I think it'll remove the speed bumps that operators feel they have to get over in order to put AI into the network, right? But then we can't be the only one. I think two operators are not enough. There needs to be more of a critical mass of this kind of capability. So you and then it becomes a avalanche on its own, right? Once enough and usually the something, the historical number is 16 percent. Once you kind of get 16 percent of operators kind of to the point where they're putting this into their network, the rest will kind of automatically get there, right?
ABE: Amanda, of course, the U.S. ecosystem is hugely important to the NTIA and to the Department of Commerce and the U.S. government, for that matter. So this is sort of a two part question. I'd like to have Amanda start and then Anand continue. What technologies, capabilities and really partnerships are still missing from the U.S. AI ecosystem? And again, I'd like to have Amanda really focus on the partnerships part of this and have Anand focus on the technologies and capabilities.
So from a partnership side, Amanda, where are we going? What do you see in the near future? And so where are the pitfalls and holes?
AMANDA: Yeah, so at the Innovation Fund, our name is the Public Wireless Supply Chain Innovation Fund, right, so we're trying to enhance the supply chain and directly the United States supply chain of telecommunications equipment that are open and interoperable. So, you know, the whole program is focused around doing that. I think a lot of my effort has been largely writing NOFOs and figuring out who gets award and such. But a great deal has also been focused on industry engagement and outreach. Right. And who can we talk to and who can we connect and really try to drive and facilitate partnerships between companies where they might not necessarily have talked or coordinated? You know, I think and I've heard it many times, and I suppose this is me tooting the program's horn, but I think particularly with NOFO 3 and hopefully with NOFO 4, just the nature of having it out and having the small companies have to work with the big companies to try to figure out how are they synergistic, how can they work together and how can they actually deliver capabilities has been hugely beneficial to the ecosystem. And I think it's been a huge benefit to the U.S. small companies and really trying to get them into the supply chain. Ultimately, we are trying to create a more resilient commercial ecosystem so that we do have options and that we can try and break into what is customarily been vendor lock. Right.
And that's that's been the biggest challenge.
So, you know, I think there's definitely work that still needs to be done. I think the Innovation Fund has had a good first cut and really trying to push the industry to collaborate and work together. You've got some groups, you know, working with partners where I don't think they would have necessarily worked together. So I'm really excited and very hopeful that NOFO 4, which is still out there on the streets, but I hope that we see partnerships from companies that, you know, are new or potentially not necessarily in the telecommunications space, but can be because of the open architecture, because we're incorporating AI. But we're really trying to facilitate those partnerships and that ecosystem to grow and to work together.
ABE: Yeah, that's interesting. I mean, startups are certainly a component of that ecosystem, if I may, and that's a good segue into Anand to focus on technologies and capabilities. So, Anand, operators, RAN vendors, semiconductor companies, cloud providers, AI companies like yours and startups as well, sort of the little guys out there, what's the commonality between all of them where they can better collaborate?
ANAND: I think the commonality across all of those is, I would say, standards-based innovation, right? I agree with everything Amanda said. I think the fostering innovation requires certainly a thriving ecosystem because it's not you don't innovate in isolation. You innovate together, usually as a team. And so that's super important, right? And the NTIA has been focused on fostering that. But the other leg of that innovation component is also the standards, right? And Open RAN, which NTIA has been fostering for some time and continues to foster, Open RAN has several standards that they've promulgated in the industry. Some have been taken up and some have yet to get taken up, right? But in part, the ability for Open RAN to kind of open up the network infrastructure northbound so the data can flow easily and then you can set commands back down southbound, right? I'm now referring to the O1R1 interface in colloquial terms, right? Those are super important interfaces, right? And adhering to those at scale, not in a lab environment where you can show that, hey, I'm kind of compliant, right? But at scale is a big deal. And pushing for that, which I think NTIA has been doing, is a very key component of enabling innovation because you would not have AI applications being able to run on these networks if we did not have that data, right? And one of my answers, I said the data's availability is increasing and part of that is the adherence to these standards. I think there needs to be more and there needs to be more consistency in the application of the standards and they need to be at scale, right? So that's the only thing I would add to the answer that Amanda gave.
ABE: Yeah, it's another good segue. Amanda, I wanted to go back to you. So with the NTIA solutions for this AI native RAN notice of funding opportunity, what types of projects and demonstrations, and Anand sort of alluded to this already, could have really the greatest impact on moving AI RAN towards commercialization? And sort of the second sort of B part of this is what should government really be prioritizing as far as funding? And what should industry also be expected to fund on its own?
AMANDA: Yeah, so hopefully the NOFO is clear that we're asking for really four key requirements, one of which is an AI native network fabric or architecture that is running everything. And this idea that this architecture will facilitate both telecommunications workloads and AI workloads at the same time is really central to what we're asking for. We're asking for a secure network. So the idea that whatever is being delivered can verify and validate security. And then the last two I see, I mean, they're all complimentary, but the last two, one being a business case, a very strong demonstration of a business case. That is really important, right? I talked about it earlier as to being one of the barriers for large scale adoption of, I think it could be a barrier for AI, AI RAN or AI native network architectures to be adopted.
So a really strong application will have a good business case proposition included in it. And then we're asking for a field demonstration, right? So something that is a science project or a laboratory project is not what we're looking for. We're looking for something that can demonstrate, show a path to actual commercialization of that capability.
So, you know, I'm not going to prescribe what that perfect business case is. I think industry is smarter than I am in that. And I'm most excited to see the types of responses that we get. But, you know, really recognizing that there has to be that clear path to actual revenue generation as part of what we're seeing as part of the submission.
ABE: Yeah. Anand, anything to add? I think that all makes a ton of sense. I will say two things and I will be selfishly biased towards the startup ecosystem.
One, I will say, I think where government can help in terms of their funding is reward innovation, right? And I think that is the bias of the government and the NTIA anyway, but I would say index on innovation, right? And why do I say that? Because, well, A, innovation is in the name, right, to some degree here. And number two, innovation comes from the startup ecosystem by and large, right? If you look at the larger firms that are in this telecommunications ecosystem, they're defined by one word, lethargy, right? And that's the one thing that I think we need to break in the US telco ecosystem. And there's a deep desire to shed off the lethargy of the past and be a lot more fast moving. And AI gives us that fundamental opportunity because these kind of disruptions in technology life cycles don't occur very often. This is a once in a lifetime opportunity to kind of shake off the past and fundamentally accelerate and leapfrog into where the future ought to be. For these ecosystems. So that's the first thing I would say, reward the innovation, reward the startups, whoever that might be. Right.
The second thing I would say is have a reward the standards based adoption at scale. Right. And I kind of I'm repeating myself from the previous answer, because that's, I think, super important, basically replicate AI solutions at scale. And if the standards based adoption is weak, I think it will stultify the adoption eventually of AI solutions at scale. So those are the two things I would probably add to what Amanda said. But I think her answer was comprehensive and does capture those two points.
ABE: Amanda, if you don't mind, could you give us for our audience, and I know I could probably tell him myself, but I like to hear from you, the final date or sort of the cutoff date for applications for this current round of NOFO funding, and then really maybe if you can sort of ballpark for us when we can expect announcements for the funding as well.
AMANDA: Yeah, so the application window closes September 9th, which is very fastly approaching at 11.59 p.m. So there's not much time left. So we will get those applications in. We will do our processing and evaluation. We're hoping to make awards in the fall. So, yeah, you know, stay tuned. We'll have the the heavy workload on our side sort of for the next couple of months, but looking forward to seeing the ideas that come in.
ABE: Well, Amanda, as always, we're very happy to have you on the program and you're very kind to give us your time. This is the third part or the third time we've spoken on a three-part series on AI RAN. The NOFO funding explains what we call it. So we appreciate you informing the industry about what what's happening now and what's to come. So, again, thanks for your time, Amanda.
AMANDA: Thank you for having me.
ABE: And Anand, as always, number one, we appreciate you supporting this session. And number two, certainly Aira Technologies is a big part of the AI RAND sort of ecosystem, if you will. So we appreciate your insight and also your time as well.
ANAND: Thanks for having us, Abe and Amanda. Always a pleasure to talk.
ABE: Thank you so much. And we'll talk soon.
Sign up for Daily CommTech Brief here.
For any inquiries, please email anejad@thenetworkmediagroup.com