On this episode of The Six Five – CXO, hosts Daniel Newman and Patrick Moorhead sit down with Sameer Vuyyuru, Head of Worldwide Biz Dev & Telco Industry for AWS, and Anand Chandrasekher, Co-Founder and CEO of Aira Technologies.
Their discussion covers:
- How their partnership is committed to solving complex challenges in the industry through collaboration with companies of all size and scale
- AWS’s leadership in sustainability
- How running workloads on AWS helps accelerate innovation, and deliver greater value to end customers
- Aira’s leadership in the application of machine learning to wireless
- How Aira is demonstrating that machine learning can deliver fundamental improvements in energy efficiency and performance in cellular networks
Be sure to subscribe to The Six Five Webcast so you never miss an episode.
You can watch the full video here:
You can listen to the conversation here:
Patrick Moorhead: Hi. This is Pat Moorhead and The Six Five is live in Barcelona, Spain at Mobile World Congress in the AWS Experience Center. It’s exciting, and I’ve been to, I don’t know, 15 Mobile World Congresses and the talk and attitudes are positive here, Dan.
Daniel Newman: Yeah. 2023 is a big inflection. A lot of the public perception is that the markets are tough. The macros are going to be more challenging, but there’s a lot of optimism here. After spending three days meeting with CEOs from some of the biggest companies in the world and having great conversations here on The Six Five, Pat, I got to say I’m going to be flying back on Friday, both tired, but with a bit of optimism about what’s ahead. This particular event is a bellwether, Pat. It’s a bellwether for the future. The telco mobile industry is so big for everyone’s life, everyone’s life. Do you look around? Everybody on their phones all day long.
Patrick Moorhead: Pretty much, and now we’re going to move to device to device to network, which is the future here. But one theme that you can’t escape here is how AI is making everything better. I know it’s easier to AI wash certain elements, but when it comes to network operations, our two guests here are going to talk about this. How are you, Anand? Great to see you again after so many years.
Anand Chandrasekher: Wonderful to see you again, Patrick.
Patrick Moorhead: Sameer, great to see you. Gosh, I think I was with you six or seven hours ago at breakfast.
Sameer Vuyyuru: It’s a good breakfast.
Patrick Moorhead: It was. Thank you very much. Anyways, welcome first-timers to The Six Five.
Sameer Vuyyuru: Thank you.
Anand Chandrasekher: Thanks for having us.
Daniel Newman: Yeah, appreciate you guys joining the show. Pat, you mentioned AI washing and AI and it has been thematic. I’ve asked quite a few questions myself in some of the different briefings because it’s like, are you going to generative AI that? But in serious, there’s a lot of really pragmatic and practical applications. Another thing though that’s been really thematic throughout our conversations this week has been sustainability. That has been a big topic. As we’ve moved towards companies talking about sustainability, I felt like early on, a lot to make sure they were appealing and appeasing that investor class and the potential employees they are trying to attract, we are seeing with this more austerity-focused market companies getting a little bit more analytical about it, saying, “We need to do this because it’s important to the world, but we also need to do this because it’s good for our business.”
Patrick Moorhead: That’s right.
Daniel Newman: We start out a lot talking about manufacturing or we talk about things like automotive industries, really heavy industries, but the telco industry is really affected by sustainability as well and that’s why I was so glad to see it spoken about so much here. Sameer, I’d like to get your take in this industry, what you’re hearing and seeing as it relates to sustainability.
Sameer Vuyyuru: Thank you, Dan. First of all, as I talk to my customers on a one-to-one basis across the world, the single biggest commonality across all of this beyond 5G is the cost of energy. It has, in some cases gone up five times, six times, seven times as a real expense item that is affecting the company’s bottom line. But more importantly, as the number of bits that get pumped through grows exponentially, it is starting to create a sustainability issue.
Patrick Moorhead: Yeah. Data density alone will likely dictate that increase in power consumption and that same phenomena is happening in hyperscale data centers too. When you carry that to the edge, the core network, it’s all impacted by this data density, this processing, and then feeding the data back for results. Anand, you had the insights to figure this out upfront that this was and could be a challenge, but also an opportunity out there, didn’t you?
Anand Chandrasekher: Correct. Yeah. Absolutely. Let me tell you a little bit about what we are doing right. Aira Technologies is a ML-based wireless company. We are founded by industry veterans. Myself, you know me, Pat. I’m ex-Intel, ex-Qualcomm. My co-founder is ex-Bell Labs, ex-Qualcomm. We are also joined in this adventure with two academic experts, Pramod Viswanath, who is the early inventor on Flash-OFDM, which is the basis for pretty much all wireless today, and Sreeram Kannan, who is a tenured professor at University of Washington. We weren’t a funded startup. What we simplistically do is we apply machine learning to all layers of the RAN stack and we improve spectrum efficiency by greater than 2X and we improve energy efficiency. What we’re here to talk about is how our ML approach can effectively attack and reduce the energy consumption that Sameer so eloquently talked about. We want to be able to control that and give some of the benefits back to the planet.
Daniel Newman: Together though, because obviously we’ve brought AWS, we’ve brought Aira. I understand there’s a demo that you’re doing alongside with Juniper. Sameer, can you talk a little bit about what you’re doing?
Sameer Vuyyuru: What we’re doing here at MWC ’23 is we’re actually showing errors algorithm, if I could call it that, model, whatever you want to call it, running on what’s called a Juniper RIC or Radio Intelligence Controller, which is part of the new architecture of O-RAN. Now, why the RAN? The RAN’s really, depending on the configuration, the band, how much throughput is between 50% to 75% of a network’s power consumption. That is the single biggest lever we have to reduce RAN power consumption. Historically, there’s always been this focus on not keeping the RAN on all the time when it’s not needed like at 2:00 AM in the morning. But that’s a very crude method and it has yielded results, but we believe that there’s a better way to do this, where you basically use models, such as Aira’s, to dynamically control the RAN, how much power it’s pumping out, what sectors it switches on and off based on multiple sources of data.
Historically, it was just the network data, but imagine you were able to also view traffic at the same time in real time, app usage and so on. So when you put all of these into a ML model, we have basically figured out that you can actually do maybe a 2X, a 3X for now, and over time, it will get better improvement on what’s best in class out there for RAN power consumption. So that’s what we are showing on a Juniper Intelligence Controller with the Aira’s model running and Viavi helping us test it and make sure that what we are reporting is real and templated against real life examples.
Patrick Moorhead: Anand, it’s funny, in the green room, I got a quick demo. In fact, one of your engineers gave the demo, which I was impressed with, but what’s going on behind the scenes there with KPIs and you explained?
Anand Chandrasekher: What we’ve developed is an application. It’s a non-real-time application that sits on the Juniper RAN Intelligent Controller. You can think of the RAN Intelligent Controller or RIC effectively as an operating system for the RAN. The demo that we’re showing is targeted at the Open RAN environment, but there’s nothing about the application that we’ve developed, all the environment that precludes it from running on legacy environments. So that’s the first thing. Now, how do we do it? We ingest standard KPIs that are made available to us in a legacy system. We ingest those KPIs coming out of an element management system or any EMS system. In an O-RAN environment, we ingest those KPIs, key performance indicators, coming out of the O1 Interface. We take about six or seven of those KPIs. What we’re able to intuit or learn using our machine learning is fingerprint what is going on in a particular region.
For example, if you’re in a rural area or in a city area, and we ingest that data, our machine learning model is able to tell how many of those individuals that are on that network are indoors, how many are in a car traveling around, how many are maybe a pedestrian. Turns out that actually matters, because if you have many low-mobility users, pedestrians, you can actually turn off some of the bands and basically save energy. If you had, say for example, a thresholding algorithm, where if you had more than a hundred users on the network, the power’s always left on or the bands are always left on. It may actually be the wrong thing because if those guys were moving around or ladies were moving around, then you might actually benefit yourself by turning the power off without impacting the throughput. At the end of the day, that’s what you want to do, save energy without impacting quality of service. So that’s in a very thumbnail way what Aira does. We use machine learning to fingerprint the environment, learn from that, and smartly turn energy bands on and off.
Patrick Moorhead: Yeah. One of the things I appreciate too is that actually, when you turn off some of the radios, your throughput can increase as well, which I thought was an unexpected benefit, but you’re like, no, no, we knew that going in there.
Daniel Newman: It just makes so much sense.
Patrick Moorhead: It does.
Daniel Newman: Like you said, it’s such a crude method to be so binary in a world where there is so much available to us.
Patrick Moorhead: Well, it reminds me the way we used to do data center power management 30 years ago, which was turn it off or take it down to 50%. One thing about this solution that’s interesting that you don’t see always is you can get benefit today and tomorrow. You have standard RAN configurations today, but also, I think pretty much everybody can agree the future is O-RAN. We can debate the year, the method, but what are you looking at as some of the benefits of O-RAN to your customers?
Sameer Vuyyuru: From an O-RAN perspective, I want to be very clear. We enable vRAN. We enable C-RAN. We enable O-RAN. Our view is that any alphabetical RAN-
Patrick Moorhead: You have to support all of them, right?
Sameer Vuyyuru: …yeah, runs best on AWS and you’re not going to use the same one all the time. That’s the benefit of the cloud, right?
Patrick Moorhead: Yes.
Sameer Vuyyuru: So if you want to switch from an O-RAN environment to a C-RAN in some geos, you can do that really easily on the cloud, which is where our value really comes in. But to answer your question about O-RAN, it is all about the data ingestion capability. It is so easy to do that with O-RAN, and it’s been augmented that way with the right APIs.
Patrick Moorhead: And with a standardized interface in as opposed, I think right now there’s probably 27 different ways to ingest data that you need to chase around and that’s not efficient.
Sameer Vuyyuru: Well articulated.
Patrick Moorhead: Yeah.
Anand Chandrasekher: If I could add to that.
Sameer Vuyyuru: Of course.
Anand Chandrasekher: First of all, I would add that the application we’ve developed when we’re demonstrating runs on both legacy and O-RAN.
Patrick Moorhead: Right.
Anand Chandrasekher: But back to O-RAN, you’re right, O-RAN will happen. You can question the timeframe when it actually happens, but O-RAN is huge for the RAN infrastructure. It’s as big a disruptor as what happened in the computing industry with PCs between 1980 and 2000. What actually happened in that timeframe was not just a big knock on cost structures, it unleashed innovation in ways that were previously not possible. O-RAN is going to do exactly the same thing for the RAN infrastructure. In fact, this demo that we’re showing collectively between Amazon, us, and Juniper and Viavi is a great example of that innovation. Now you take AI and you combine them with what’s going on in the cloud, you have really huge possibilities of where this can go.
Daniel Newman: Well, you definitely hit all the trends, the trend lines, mobility, AI, sustainability, 5G.
Anand Chandrasekher: I wasn’t trying to list the trends, but I guess I did.
Patrick Moorhead: It’s like a 3-for, or a 4-for.
Daniel Newman: Like one of my Forbes articles.
Patrick Moorhead: Yeah.
Daniel Newman: Sameer, let’s go and take this home. You mentioned a comment early on about a lot of the power consumption by telco’s done at the RAN, and so this is going to solve some of it. What do you see? Is this the biggest contribution that tech can make to solving the energy crisis and the sustainability issues for telco? What other methods and means do you see to help accomplish? What’s going to be a goal that’s never, by the way, going to be good enough? We’re always going to need to do more.
Sameer Vuyyuru: Thank you, Dan. First of all, at Amazon, we’re absolutely committed to this. You’ve seen the climate pledge.
Daniel Newman: Right.
Sameer Vuyyuru: You’ve seen that we want to be water-positive very soon. 80% of our infrastructure today runs on renewable energy. We expect it to be 100% in a few years. We’re taking this very seriously, not because it’s the cost necessarily, but just because it’s the right thing to do. In the telco domain, let me talk about three ways we’re addressing this issue. The first one is just moving from on-premises data centers to the cloud, by itself, means that you’re operating on ESG, eventually 100% compatible infrastructure. Number two, it’s all about the silicon innovation, which you would appreciate being a silicon veteran.
Patrick Moorhead: I do too.
Sameer Vuyyuru: Oh, you’re a silicon veteran too. We’ve been investing in our own silicon, which we call the Graviton series. If you look at MWC, what we’ve announced with NEC’s 5G core running at NTT is a 72% power reduction over historical microprocessor architectures. Now we expect to bring that same energy savings to the RAN over time. The third one is the power of AIML with partners such as Aira to basically take all of the data that’s available out there in a governed, controlled manner to create a real business outcome that Anand, more eloquently than I, talked about.
Anand Chandrasekher: If I might add to that on that third point…on AIML. We talked a little bit about how we are applying machine learning to reduce the energy consumption. Another way you can actually impact the power consumption of the RAN network is we use machine learning to improve spectral efficiency by greater than 2X. The implication of that is you can actually reduce the number of base stations quite dramatically without reducing coverage or quality in any meaningful way. The net effect of that is fewer G-node B’s, fewer base stations means lower power consumption of the RAN. In fact, that’ll be a bigger way to impact the power consumption in a positive way than the application we’re showing now. So that’s another example of how AIML can be a very positive contributor.
Patrick Moorhead: Yes. Sameer and Anand, both of our companies have written extensively on energy reduction and efficiency and the three that you hit on. I’m smiling just because we cover your silicon innovation too, and its Inferentia and its Trainium that’s coming up as well. I know that your latest version of Graviton’s not going to be the last one and we keep moving in. Fundamentally, when you have a hyperscale data center, it’s built to be the most highly efficient in leveraging its size.
Daniel Newman: Dave Brown’s going to love this part.
Patrick Moorhead: No, he is. I’m going to text to him. Exactly. But anyways, I think this is a good place to wrap. Sameer and Anand, thank you so much for coming on the show. We’d love to get an update to maybe in a year or so or sooner to see where the two companies have aligned and your march to improved energy efficiency out there. So thank you so much.
Anand Chandrasekher: Sounds good. Thank you.
Sameer Vuyyuru: Thanks for having us, Patrick, Dan. It’s just the start. It was day one.
Anand Chandrasekher: Absolutely. Thanks for having us.
Daniel Newman: Thanks so much, guys. All right, everybody, there you have it. We are here for The Six Five at AWS’ Inspiration Zone. This is MWC 2023 Barcelona. Pat, for you and me, it’s a wrap. It’s been a great show.
Patrick Moorhead: Been a great show.
Daniel Newman: Thanks everyone for tuning in. Hit that subscribe button, watch all our shows. We’ll see you later.