MWC 2025

Starting with Orange, I’m giving you the floor. Give us a little bit of the background, why you’re here and what you guys are doing.

Thank you. Thank you, Inbar. And hello everyone. I’m very pleased to be here with you today and with this great panel. So, yes, I am the head of the international network for Orange and I will briefly introduce our activity. I will take the mouse. What are we doing? So first we are managing infrastructure. We are managing 500,000km of cables, submarine and terrestrial cables. And on top of this next slide, we are managing a tier one Internet backbone. We have about 250 points of presence in 100 countries and we have also 50 telco clouds to manage the telco services. And if we move to the next slide, so we are a wholesale and B2B business. So we have roughly 1,800 customers, mainly operators, hyperscalers and multinational companies such as banks, factory and shops. And indirectly, we are serving 2 billion end users. We provide layer 1, layer 2 and layer 3 connectivity, SD WAN, CDN services as well as voice and mobile services. So if we move to the next slide, how do we see the market evolving? Quite basically we see our customers moving their application towards the cloud. We see our customers moving from dedicated MPLs towards SD WAN. Considering that the Internet is good enough, we see a huge traffic increase in our network. The traffic is doubling every three years. This is mainly driven by 5G. We see more and more requirements for very strong SLAs with IoT. And we see as well with GenAI, there is a need, of course, for training and for optimization. We see the need for what I call bursty traffic. We have some customers that request for instance a 100 gig or a 400 gig only for a short amount of time. And so at the end of the day, our customers are looking for very high speed connectivity in a flexible way. They need some programmable connectivity. So having said that, if we look at our network, it used to take months to upgrade the network to set up a new connectivity. While in the cloud it takes only, as you know, a few minutes to add a cloud instance. So what we have to do is very simple. We have to move from months to minutes to update our network. And this is what we call network as a service. The ability to provide on demand connectivity in a pay as you grow model. In order to do so, we have been launching a network transformation that is quite simple with five main pillars. The first pillar, we are still investing massively in the infrastructure. We are investing in order to support 30% traffic increase every year. On top of this, on the IP layer, we are currently softwarizing this IP layer. And the softerization of the IP layer relies on two main building blocks. First, the disaggregation and I will come back a bit later. And second, the segment routing. Why segment routing? Because it allows you to slice the network and to provide differentiated SLAs depending on your requirements. Then we have been introducing the Telco cloud layer that allows for a decoupling between the underlay and and the overlay. And on top of this telco cloud layer, we are virtualizing all what can be virtualized. We have been virtualizing SD WAN, CDN, we are virtualizing our voice services and we are virtualizing as well our mobile services. So in a nutshell, we are transforming the network from a rigid box to a kind of Lego box. And then we need to play with the Lego. And in order to play with the LEGO pieces, we have this orchestration layer on the left hand side that allows to combine Lego pieces and set up end to end service for our customers in a zero touch operation way. Relying on automation and artificial intelligence when it makes sense. And last slide before to move the floor. If we move. So now if we deep dive on the disaggregation of our IP MPLS network. This is a journey we started three years ago. We started in 2023 disaggregating the access of the network. It was an easy the layer two switching really at the access of the network. We started by this part because it was easy and removed to learn step by step. So we did this relying on the Sonic Open Source network operating system with our own developers to support this. And so we are the first box in 2023. And now end of 2024 we have more than 200 points of presence relying on those disaggregated access switches. As we are happy with disaggregation, we decided to move forward to the core of the network. Disaggregating the core. This is why we decided to launch an experimentation with DriveNets. We did a live experiment in the core and the results are very, very promising. What we are expecting from the desegregation this is mainly four benefits. The first one is to gain some flexibility, elasticity, ability to upgrade easily with fine granularity, the hardware ability to easily upgrade the software independently. The second benefit this is we are looking to reduce our cost to gain competitiveness. And here when you separate the hardware and the software, it’s really easier to optimize your cost. The third pillar this is, we wanted to simplify, really our operation, simplifying the ability to upgrade, to add some capacity without reinventing our design. And we can do this thanks to disaggregation. And last but not least, we are looking at our carbon footprint. And thanks to disaggregation, you can really fine tune adapt your hardware exactly to your needs, so you can really save some carbon. You can also better manage the obsolescence of your equipment. So this is very promising.

Okay, thank you, thank you.

Thank you for having me today. I’m Kenji Kumaki and actually I appreciate the opportunity to join this panel discussion today. Okay, so here’s our KDD company overview here. And actually we are reading on the Japanese telecommunication providers and actually we provide it on the mobile and the fixed enterprises customers services as actual core business and market cap is like that. And actually we have over 50 million subscribers here and we are one of the Japan top three carriers and the others in the SoftBank and NTT Docomo like that. And also the most important thing is we are focusing on open innovation like that. And actually we have already deployed DDBR architecture as a peering router in production networks for about almost two years ago. Okay. And sorry and also that we have collaborated with the global operators like US Tier 1 carriers and the European Tier 1 carriers as well. And also we are focusing on that open innovation standards like TIP and actually I’m TIP technical members and also that reading the Distributed aggregated Open Routers DOR and here is our network vision and actually you know AI traffic volume is growing and that growth is much higher than we expected. So that’s why we are focusing on open disaggregated model and in order to prepare the AI traffic on growth we want to focusing on open disaggregate model to navigate through the AI era. And actually right now we have challenges on on regarding the traditional routers like Cisco and Juniper or something like that. Actually we have a limitation of the number of line cards and also that we have the dedicated routers such as Peering and the Backbone Core and also the Edge router as well. And also we had have the dedicated knowledge and spare parts. But if we move on open disaggregation model we can solve this issue. So that’s why we are focusing on this model and here is our network on roadmap here KDDI have already deployed DDBR standalone as a peering router in about two years ago and as the next steps we are focusing on the backbone core and two weeks ago we KDDI have announced we are moving on the backbone networks and we are using the DDVR cluster as well. And after that we are moving on the edge routers and the deployment schedule is 2028.

Thank you. So Kumaki-san, I’m going to ask you so Orange is very, I would say they have aspiration very bold plans for 2025 but you have experience with deploying that. So from your experience what did you learn from the project? You already did that, you know Orange and Cox can can learn from that in their journey.

Okay so I think we have the three major factors and actually one is the software stability and the second is vendor support and collaboration. And finally I think the router flexibility and router capacity extension. I think the three major factors So I remember actually we KDDI have deployed on DDBR standalone architecture as a peering locus in about two, about two years ago and actually we haven’t on experience bags and service outage during two years. So meaning that the software itself is very very stable right now and also you know that the DriveNets is a very reliable software vendors for KDDI and also I think I have discussed DDBR architecture with DriveNets and Broadcom since 2018 and actually they are really really supportive and also I think they accepted our feature set we KDDI required and also finally that the DDBR architecture itself is really really on flexible. I mean that we can make a router itself very easy and also that we can change that cluster size very easily. That is also very important for KDDI and also that the traffic extension, sorry the capacity extension is about right now our DDBR architecture is about over 9 terabyte, 900 terabyte per second but compared to the existing routers it’s mostly one and a second. So meaning that its DDBR architecture is really really high capacity. So that’s why we will deploy a DDBR architecture in our backbone networks. So yeah, thank you.

So Cox Communications, I don’t know how many people really know Cox Communications. You do. Fantastic. So we’re the third largest cable company in, in America. We come with all the challenges of what we are going to do and how we are going to be relevant in the future. So there’s a lot of things that we’ll. I’ll just give you an introduction what we are. So if you look at Cox family of businesses, it’s a private company. So Cox Communication is just one of the what I would call as vertical arms of Cox Enterprises. So we are all into business, we are into residential services. We have actually started mobile services. We are, a huge bag is Cox Auto. People don’t understand that. We are the biggest Cox Auto auction house in the world, right. And then we have the Cox Enterprises. So one of the focuses of the family is how do we make tomorrow better. So if you look at some of the things that we do, OpenGov, Cox Farms, Grows Edge, Nexus, all these guys, all these things are actually towards being making tomorrow better. So that’s basically what it is now. What is the future vision? I didn’t go into the network diagrams and everything else because I wanted to give a huge a highlight view. Number one, by 2034 we are going to increase our residential homes. We manage residential homes. We actually Cox Communities is more around MDUs and how do we actually work with MDUs and offer service out there. And then the biggest one is small medium business. So Cox business is all about going B2B. Right. And then the one thing that’s missing out here which I should have put in is Cox Wireless. We started Cox Wireless about a year and a half, two years ago and seems to be doing pretty well. I can’t go into numbers, anything else, but I would say that we have been, we’re more than happy as to how it’s expanding and how we are growing. And then at the end of the day on the other side is all about business technologies. So if you see a problem, it’s multifaceted. We have a lot of different things, some things that were done, you know, like 20 years ago and we should be knowing that too because there are things that are, that exist as part of the legacy network. How do we evolve that legacy network to actually remove the costs but more importantly align with how the business and business models are evolving. So there’s something that I actually hear a lot is that we need to know the use case before we can start designing something. Those days are over in my mind because use cases back in the day used to take a year, two years to actually come out. Now it comes out in weeks and months. So we need a network that can be what I call as a programmable data driven network. So you can actually reprogram your network to align with the latest business model, and that’s what we are really looking at doing.

So Igal, you’re the most experienced here in disaggregation in the room. I think most carriers and I think the word convergence and disaggregation came up in your analyst day in December quite a lot. So tell us what’s AT&T?

Good afternoon everyone. Igal Elbaz with AT&T. Before I even start talking about AT&T, when you think about an industry and you think about disruption, what you see on the stage is what you need, right? You need customers with scale, you need merchant, doesn’t matter which space you are, you need merchant silicon and you need software that do the network, in this case network operating system in the audience we also have the hardware piece. So these are the pieces that you need in order to create disruption in certain industry. Now, at AT&T, our mission is, as you said, Inbar, to be the best converged broadband company. We are doing this because our customers are telling us that this is what they want. They want one provider for all of their connectivity needs. To support that, we are building a converged network. It’s made out of two scale platform, a 5G wireless network and a multi gig fiber network. A 5G network already cover more than 300 million POPs in the US with our new mid band that we’re deploying, we’re already covering 270 million pops. At the end of 2023 we announced the modernization or the acceleration of the modernization of our wireless network as well as introducing openness and open ran into our wireless network. At the same time we’re continuing to migrate our customer to our standalone core on the fiber side. In 2025, as we said publicly, we’re already going to cover 30 million living units. And in our last analyst day in December we announced that we are going to expand this into 2029 and by then we’re going to cover 50 million living units. 45 million we’re going to build ourselves and 5 million is going to be out of partnership that we build over the years to such as Giga power or open access platform. Now you ask about converged network, how do we do that? Well, in the past, the first, even before that, our convergence starts with an integrated network planning. A lot of what we do comes out of the same AI based network planning platform. We used to build fiber to run different use cases. Now we are trying to make sure that the same fiber routes that we design are supporting three distinct use cases. Mobile, backhaul, consumer broadband and enterprise services. In the past, the first point of convergence in our network was our core backbone in difference than Orange. We started our open disaggregation years ago in the core backbone. Just to give you a sense, this is where we carry close to 840 petabyte a day. Okay, this is where we started our open disaggregation and I’m happy to report that over 80% of our traffic is already carried over this open disaggregated architecture with DriveNets and Broadcom. That’s quite significant if you check the recording from two years ago. I can’t remember what the number I reported back then, but I guess it was 660 or something. What we are doing is now we are pushing the same architecture into the edge. We’ve been on that journey for the last several years. It’s the same hardware platform and we can push it to converge all of the use cases at the edge of the network. On top of all of this, we are building a software layer to help us to take all of this and make it a platform for innovation. This is where we’re putting software on top of our network to help us expose APIs and monetize our network in different ways than standard monetization in our industry.

Hi, my name is Ram Velega.
I run the switching and routing team in Broadcom.
In this discussion about disaggregation, a couple of things I just want to bring to your attention.
Not long ago, if you tried to procure about 128 ports of 100Gig, 128 ports of 100Gig, you’d be probably paying about a million dollars in capex to an OEM. Okay. And the amount of power that box would consume was about 10 kilowatts. Okay. And it pretty much took up half a rack of space. Today with what’s happening in silicon, you can do the same. 128 ports of 100 gig. And we would be lucky if we got 10 to $20,000 per that switch hardware in the marketplace. So think about it. You’re going from an equipment that not long ago cost about a million dollars now in the range of 10 to 20 thousand dollars. Right? And then when you think about it in the context of power, you’re going from something that consumed 10 kilowatts of power to something now that consumes less than, you know, thousand watts of power. So the role that Broadcom plays in this idea of disaggregation is that we are a very, very firm believer that that bandwidth is continuing to grow significantly. Right? And especially when you think about it. I’ll give you another context for this. If you try to buy a traditional server, you would have 50 gigs of a NIC coming out of it.

Today, if you buy a GPU, the backend of that GPU will have at least 400 gigs of bandwidth coming out of it and soon shipping with 800 gigs. But there’s also something called a scale up that is attached to the GPU that’s going to be about 10 terabytes of bandwidth. So think about this. You know, if you bought a traditional server, it’s 50 gigs of bandwidth. If you connect the backend of a GPU, it could be anywhere from 400 gigs to 10 terabits per GPU. Right? So when we are actually building hardware for that kind of scale and we take the key IP from there and bring it into the service provider network, what we are able to provide is completely change the economics of this business. And that’s where the role that we play is, being able to provide the hardware. That absolutely changes the economics of this business. Now obviously this hardware on its own cannot do much. And that’s where I will give it to Hillel to talk about how their software really brings all of this together. And one last Point on this. Back in 2015, we were already at least five years into this journey of disaggregation with the cloud guys, the big cloud guys, whether, you know, I won’t take their names here. But then AT&T came to us in 2015 and said, hey, look, what you’re doing with the cloud is very different economics than what today AT&T is used to in terms of the equipment that they’re procuring from the OEMs. And they said, look, we want to be on the same price per bandwidth curve that the cloud guys are on, because otherwise, if you’re competing with them, we don’t have the cost structure to be able to play in that space. So we said, okay, we can build the silicon, but we need AT&T to tell us exactly what features you need.

So over multiple years, they sat down with us and they actually told us exactly what features they need, and we built the silicon to the features that they asked for. Okay. But then they also had the foresight to say, hey, look, just the silicon’s not sufficient. You need somebody who builds the hardware for it. And then they had partners like UFISpace that come in and build the hardware to it. But they also realized eventually you need the software. There was some discussion about using something like Sonic. Yes, you can, but obviously that’s something that you can use in use cases towards the edge. But they wanted to do it from the core of their network and you needed somebody to be able to actually write this software from scratch that is able to scale. And they went and partnered very closely with DriveNets back in 2015, 2016 to build what it is possible today. And I can tell you, back then, most people would actually would have told us this is not going to be successful. Right? But to be where we are today, where 80% plus of an 800 plus petabytes of traffic per year is happening. Sorry, my apologies. Day per day is happening. That’s because of the vision at AT&T, but more. And along with it, the execution from DriveNets. Right. You know, Broadcom is doing what they were doing. We are okay with it. But then obviously the help from, you know, Vincent at UfiSpace. So we’re very thankful for you guys and I actually think the rest of the telco industry has a lot to be thankful for. These guys who led this effort here and were trailblazing, a lot of these guys would have been fired if this thing did not happen back then. So thank you guys. We very much appreciate you taking the first step forward.

You know, I’m curious about DriveNets. because that could bring your costs down. You know, disaggregation as well. You know one of the things we always talk about the show where we’re talking about step income and is the ability to create a new revenue generating services. Right. Because you’re bringing your bottom line down, but you have to bring the top line up. And I’m curious with the concept of convergence and disaggregation, what are some of those services? You think new services, you think you can monetize.

Anyone? So I can take it. That’s not it. So I wish I had a crystal ball because if I had I would have given you a right answer. But I don’t. The only thing that I know is whatever we have today is not going to be there tomorrow. And how do I actually think about a network architecture that can evolve? I can push it, I can pull it, I can actually get it to fit what the use case of tomorrow is going to be. The other thing that I look, I absolutely think is the use cases of tomorrow are going to come in faster and faster which basically means that we will not have the time to re architect the network and rethink it. So yes, we are talking about disaggregation, we’re talking about core routing and everything else. But we need to think about it from an end to end perspective. So for example, in my company we have Docsis, we have PON, we have wireless, we have, you know, like commercial business, we have Segra, we have dark fiber. In fact we do provide a lot of dark fiber to my friends in AT&T for their backhaul. So but again that is what we know today. The difference is tomorrow’s gonna be a little bit different. So we have to think about it from a complete end to end perspective.

Okay, I can add a perspective. At the end of the day, our network is our product, that’s what we do. And if we are moving from a point on which from an idea until we can put something in the market, if we are moving from months and years to something that we can do way faster, we have achieved a lot. I know that in many cases when we talk about open and disaggregation, the first question that comes to mind is how did you drive your cost down? It’s important. There’s no doubt there’s like that’s a major driver for those transformation. But at the same time, as you said, if we’re not going to use it for converging security into the network like we’re doing with dynamic defense, like doing things like AT&T Turbo on our wireless network, the ability to iterate quickly, take advantage of the fact that the network is more configurable, programmable, the data is way more exposed, intelligence is way more exposed and it helps us to move way faster in creating new product and services.

I think it’s a good point on where the revenue opportunity was. I think couple of years ago, no, we thought, you know, some of these silicon companies are going to have a trillion dollar market cap. Even the best analysts out there won’t have forecasted revenue growth. Right. I would say we are living in an era where for the first time distributed computing is taking off at a massive scale. Now most people, when you go and ask them the difference between cloud computing and distributed computing, they can’t explain it to you. Okay, but think about it very simply. In the world of cloud computing, any one application never fully pushed a CPU. So you took a CPU, used a hypervisor, virtualized it. The role the network played was not that significant. You provided some segmentation and so on and so forth. But today we are living in a world of distributed computing where the network is the computer. Okay, this is what Sun had the foresight of saying it 20, 30 years ago. I couldn’t understand what they said then. But now when you think about it, in distributed computing, network is the computer. In machine learning AI, it is a distributed computing problem. Any one GPU cannot do the workload. It’s a distributed computing problem. When it is a distributed computing problem, first thing you can say is look, I can try to fit it inside the rack. Then you’ll say, well I’ll try to fit it inside the data center. Then they’ll say look, none of these data centers have been built to have multiple gigawatts of know power out there. I’ll give you an interesting statistic today. My understanding is all of one of the large cloud, you know, guys have about 3 to 4 gigawatts of power available to their cloud data centers. They’re going to be building 4 to 8 gigawatts going forward. What this means is people are going where there is fiber, where there’s power in space. They’re not looking to wait to build these huge nuclear reactors and so and so forth. So network is the computer is going to play an extremely important role. So if you are a service provider who owns the fiber already, has the right of ways already and you can now throw away your junk equipment, that’s not only not giving you the bandwidth, but actually give you bandwidth at scale, at the cost structure that is needed. You can actually be the core of this network being the computer. So I think the possibilities are very different than the world that we used to live in, the network. I think there’s something to keep in mind. Now, can any one of us imagine what the next app is going to be? I don’t know. But what I can tell you is the network is the computer. You need the fiber. You need to be able to move this data very, very quickly from point A to point B.

Jean Louis, it’s so great to be here with you on the MWC show floor. So we’re here at the DriveNets booth. We just had a panel and I wanted to talk to you a little bit more about some of the stuff you mentioned. So what is the main challenge in your network today?

Yeah, so basically our customer requirements are evolving very fast. The traffic is doubling every three years and customers are moving their application to the cloud. They are moving from MPLS to Internet. They also need for very strong SLAs for IoT, for instance, and sometimes they need for bursty traffic, in particular for GenAI. So at the end of the day, they need for programmable networks. So we need to move from a rigid network to a flexible network and to propose network as a service.

So talk to me a little bit about strategy. How are you addressing these problems?

So in order to move from these rigid networks to this flexible network, we have a simple strategy with five pillars. The first one, we invest in our infrastructure in order to absorb the traffic increase. Then we are ‘softerizing’ the IP layer with disaggregation and with segment routing. And we are decoupling the underlay and the overlay thanks to a telco cloud architecture. And on top of this telco cloud, we are virtualizing our network services such as SD wan, such as voice or mobile services. So we move the network from a rigid box to a Lego box. And to play with those Lego pieces, we have the orchestration layer that is going to instantiate the Lego pieces and set up end to end services for our customers.

So I hear you talking about an orchestration layer and also about playing with Lego. Where does DriveNets fit into this plan?

So DriveNets clearly fits at our IP layer. We are ‘softerizing’ the IP layer and it comes with disaggregation. And by the way, two years ago we disaggregated the access of the network and now we think about disaggregating the core. This is why we did this experimentation with DriveNet that was successful.

Talk to me a little bit more about the work you did with DriveNets. So what was your key takeaway from the proof of concept that you did with them?

So we did first, we did a lot of testing and we were quite happy. It was quite easy to understand the product, our engineers and to learn how to configure the product. Then we did some interworking. Basically we were using the basic routing protocols such as ISIS, BGP, LDP, etc. So we did some interworking. And the interworking with other suppliers went very well. So then we deployed in our live network and it was working very well. And it took only 45 days between the start of the project and the deployment, which is quite a record for us.

That sounds pretty fast for a proof of concept. So we’ve talked about trends, we’ve talked about the proof of concept you’ve done. What’s next?

Yeah, what’s next? So we are still experimenting on DriveNets. We are currently doing sustainability, some tests on the cluster mode, how to combine several pieces in parallel. We are doing a study to compare several clustering models and we are launching an RFP in order to generalize the evolution of our core backbone.

Excellent. Thank you so much for your time.

You are welcome.