To embed our video on your website copy and paste the code below:
<iframe src="https://www.youtube.com/embed/jFgCXJKtHkQ?modestbranding=1&rel=0" width="970" height="546" frameborder="0" scrolling="auto" allowfullscreen></iframe>
Guy Daniels, TelecomTV (00:14):
Time I think to start day two and we have a big focus this morning on AI at the edge. So first of all, if I could invite the presenters for the first session onto stage. We've got Chetan, Nico, and Dario, so if the three of you could come and join me on stage and then we'll start our presentations. So give them a round of applause while I'll make their way to the stage. So Chetan is going to start us off. So Chetan, if I can ask you to go towards a lectern.
Chetan Narang, Colt Technologies (00:46):
Good morning everyone. I hope you guys had a great day yesterday and it's a pleasure to open the event today. So Colt edge. Just a quick introduction. Chetn Narang, I lead product innovation at Colt, focus on AI product strategy and I'm here to talk about Colt Edge AI. Just a quick thing, I think there's a short agenda. I think I just go through a quick Colt overview, talk about the Colt edge propositions, what we have and then just going to give kind of what we are doing in the Edge AI space in terms of prototyping as part of the new services. So just a quick word on Colt, Colt is a global digital infrastructure company. We are mainly a fixed line service provider focused on I think having a global network. We one of the largest B2B digital infrastructure platform. We mainly serve enterprises. We have around 20,000 plus customers around presence in 40 countries and around 230 cities out there.
(01:55):
Our network is 400 gig plus in terms of the core infrastructure. We have also trialing 800 gig kind of a capacity. That's basically an evolution of the network we see. I think also from a strength of our network we have are one of the most connected network in terms of connecting to 1100 plus data centers, 31,000 buildings and around two 50 plus on-ramp capabilities kind of it. We also have an award-winning NaaS platform which is about digitization of how do you consume network services, kind of it going towards the edge side. I think we have two aspects of edge propositions currently which are quite focused on network functions. The first one on the right-hand side is the network edge, which is mainly focused on multi-services telco cloud. We have a deployment footprint of around 200 plus Colt owned VNF I want to highlight. It's called owned VNF.
(02:59):
We haven't opened to the partners or customers yet, so it's pretty much Colt estate right now deployed across 12 strategic geo cities which aligns in terms of where our network presence is. It's a multi-tenant architecture and we run thousands of services across. I think it's a very dynamic scalable platform and it's optimized for TCO and accelerated ROI. I think we want to have a practical approach to it but also want to ensure that it's scalable and high performing. On the left-hand side we see the on-premise edge. The on-premise edge is about a hundred plus enterprise site deployments and I think this shows in terms of what we have in motion from a distributed scale and field services kind of it at the same point I would like to also highlight I think on-premise edge is not for every use case, it's for critical enterprise use case and I think one of the main use case that drives this is mainly the security aspect of it and I think we see the strategy growth enablers from this, which is part of a zero trust access in partnership with Zscaler.
(04:11):
We also see Quantum Safe networks as the other basis. And then lastly about Edge AI, Edge AI also is one of the main perspective now just coming back to the topic, what we have over here, the edge AI aspect of it. I think we've been also going through and hypothesis in terms of how do we evolve these capabilities from an AI perspective and as we've seen in terms of what's happening in the industry, we see that, I think there are like three ways of monetizing this. I think the first one is basically you bundle the hardware and the software go in the space of GPU as a service kind of a construct and I think you try to position more like network edge is where do you do local data handling and you can also have latency wins from a perspective of network and compute when you compare to the cloud side of the business.
(05:02):
The other aspect is also from an on-premise edge is basically you can have on-premise deployments where you feel like the response time is critical and you need a quick response from a decision-making perspective. So you try to position on the on-premise head side of a business, that's one. The second one is basically you evolve from the hardware and the software towards more towards the platform space. This is basically doing the AI lifecycle aspect of it, starting from data ingestion to training and taking to the inferencing at a scale and maybe going also to a model as a service, kind of a construct over there. And the last one is mainly towards outcome, deliver outcomes to the customers and it also goes back to the debate what we have in industry telcoms going from a telco to a TechCo kind of a concept over there. I think on that approach we have a twist. I think we also see there's an opportunity from a partnership perspective you work with a wider ISV ecosystems to deliver outcomes to the customers and as part of that in the subsequent slides I'm going to talk about two of the couple of engagements. What I've been doing with partners to develop kind of an Edge AI proposition. This is mainly focused on the on-premise aspect of it. I think the first one is basically the Colt Edge AI for smart buildings.
(06:26):
It's an Edge-first AI, AI IoT. What it means is basically the data processing and the AI models in this case are running at the edge with cloud mainly used for data gathering, running analytics and also mainly providing training of an AI ML model either way. So how it's been positioned is basically deploy a Colt edge on the on-prem, you connect to A BMS system, you try to put sensors, gather that census, get it co-location into the Colt edge and then you push that data into the cloud and we've been partnering over here with the cloud company with the smart building solution company, which is nuca over here and once that data has been pushed into the cloud, the data analysis and the models are trained on the cloud and then they're pushed into the edge to provide predictive maintenance, remediation and optimization kind of an aspect to it.
(07:24):
It's also an example of a good AI because there's always a debate about good versus bad AI and it's a good example of innovation for sustainability. The interesting aspect is also this is one of the use case where AI ML doesn't need GPUs. It's a very compute intensive, it's just running on cpu so it provides that innovation for sustainability aspect. We also have one, the code, I think it aligns with the Colt ESG values. We are quite strong, we're quite vocal on the ESG perspective. It's one of our main pillars. It also gives us, I think we've also won an eco innovation award with a partner with a customer partner Bertelsmann there, so that's one use case. The second use case which I'm going to touch on is basically the edge AI kind of an aspect which is more towards the street talk. It's all about building edge on the street.
(08:16):
This example shows of a rack or an edge being deployed in London in show just next to Colt House. We've been working with partnership over here with CIN, which is a startup working on this. I think the basic principle in this case is basically to have Colt edge AI for public square and this public square can be anywhere. It can be in the middle of a street, it can be in a campus, it can be in an airport, it can be anywhere where you have free space to give that capabilities from a concept. This is all about building infrastructure stack, which is rack coming up from CIN connectivity comes from Colt and then the Colt edge also part of the stack and then it's all about how do you enable those use cases on top of it. And this is where it comes to having an outcome led approach and having a wider part ecosystems.
(09:13):
We have been working with few of the partners in terms of evaluating this. I think some of the use cases, what we have looked is like a private 5G fixed wireless access with Weaver Labs, Edge gaming with EdgeGap, Canadian based company. It's all about how do you give the gamers and access to an edge based upon geo proximity kind of aspect. Wifi off flowed from a global reach perspective, from an AI perspective, what we have done in this case, we have worked on AI video analytics with a partner called AICuda. There's also a value from an AI perspective to run analytics over there and that analytics can be across demographics of what people walking in, how people are moving in. It can be on the basis of traffic. It can also be based on the environment and once you run this analytics, the idea is that you can also monetize it by providing insight for example to the retailers which they can use the data to tailor their services in that way. That is just one example of how we can do it over there. Yeah, pretty much. I think we are still on a journey. It's a new subject for us. It's evolutionary, but I'm happy to take any questions as part of the panel discussions. Great, thank you for listening. Thank you
Guy Daniels, TelecomTV (10:29):
Very much indeed. Chetan, come and sit down here over this side come in fact pass that over. Great, thank you very much. That was great. We're going to get to see if there's any questions yet. I'm sure there's a question or two you'd like to put to Chatan, so let's have a look at hands and also looking for the roving microphone which is, oh here comes Tony. Tony's going to have to run the length of our ballroom across to Francis here. Well done Tony intercepted. Dean, nicely caught.
Dean Bubley, Disruptive Analysis (11:01):
Sorry about that. I really like that presentation. I've said for a while that I think there's more synergy of edge compute with fixed operators than there is with mobile. Are you pushing this towards other industry verticals? You mentioned the smart building, but are you also looking at this for industry, maybe transportation sites, other things?
Chetan Narang, Colt Technologies (11:30):
Yeah, so I think definitely from an edge perspective there's a vertical play I think I agree with over there. So I think the couple of industries which we are trying to focus is one is manufacturing, which comes from a perspective of computer vision being the main use case from a predictive maintenance kind of an aspect in that way. And also I think retail, that's another thing, how do you do shelf analytics, smart retailing, kind of a point of sale and wind pre analysis. I think that's another use case which we are trying to target from perspective.
Dean Bubley, Disruptive Analysis (12:01):
One quick follow up. How much power does that...
Guy Daniels, TelecomTV (12:03):
In case anyone didn't hear that? How much power?
Chetan Narang, Colt Technologies (12:08):
I think so at this point I think this, I don't exactly know what is it, but it is a CP intensive. We haven't deployed GPUs in that way, but the current analysis, what we have is that you don't have to go with massive GPUs. Also kind of if you think from an on-premise edge and if you say it's dedicated to the customers, you can even go with Nvidia RTX two thousands which are competitive from a pricing perspective and is also enough like 70 watts. I think if I can, don't quote me over here, but maybe just top of my head, I think it's low from a power utilization perspective it should be enough. I think when you go towards the network edge you talk about massive GPUs like the 6,000 pros and all and that's where you're looking about 15 kilowatts per rack and that's a total game in that sense anyway.
Guy Daniels, TelecomTV (12:55):
Great, great question. Thank you Francis.
Francis Haysom, Appledore Research (12:57):
Very quick question. You showed two really great use cases there. Have you at all looked at what I would term is what you're going to need to do to scale them, make the at scale business case for them because each of those has quite a lot of investment in IT and payback and are you looking at all in terms of that sort of ROI model in terms of what you're doing with these cases?
Chetan Narang, Colt Technologies (13:24):
Yeah, I think it's an interesting one because so for smart building I think we are trying to prioritize that and take it back to get our different markets which are quite sustainable conscious like Singapore is one of the markets which we want to target and from a perspective of green buildings kind of a concept I think in this, I think it, it's a journey for us because this is not an easy, if it's a connectivity we controlled asset and it can position it in that way. This is more of an ecosystem play. I think the real win will be if we can evolve an ecosystem, basically this is one use case but maybe if you can have certified applications or certified aligned to the use case so then you can say these are the outcomes you can deliver. That's where I think we can get the scale in a way because one solution or one partner will not fit the fifth answer or an outcome to the customer.
Guy Daniels, TelecomTV (14:19):
Great. Thanks very much and great question. Thank you very much. Do we have another question from our audience? We can probably slide another question in if we're fast. Well I've got a quick We do, we do. Terry, thank you.
Terry Riley, Spectro Cloud (14:32):
Hi Chetan, Terry Riley. We are seeing edge being deployed in lots of restaurants. Is that a use case that you are coming across?
Chetan Narang, Colt Technologies (14:44):
Yeah, I think I've heard about it. There's a lot of restaurants also from a perspective of AI. I think I've seen it out there. It's just like I think our focus is mainly B2B, which is more towards enterprises and professional services. I think we haven't gone into that kind of a use case, but I certainly think there's a value over there. Yeah.
Terry Riley, Spectro Cloud (15:06):
You say B2B, the customer I'm talking about is a large global coffee retailer with 53,249 stores. That's a business in my opinion.
Chetan Narang, Colt Technologies (15:17):
Yes, I know about who you're talking about. Yes, so I think are think in fact I know two names come across in that vertical out there. I think it's an interesting phase because that's another thing of positioning out there because if you go with, they're like DIY approach in a way. Basically they want to build this up and they want to have it. I think we would like to position themselves in where we can support the journey aspect of it. But having said that, I think that's more about the market segmentation, but I agree restaurant is a business and if you talk about that kind of a scale, yes, and I'm aware there are use cases, a couple of use cases on that way we can do it.
Terry Riley, Spectro Cloud (16:00):
Yeah, the main driver is compute vision you say, but it's looking for length of the queue and how quickly you start getting cofee.
Chetan Narang, Colt Technologies (16:08):
Yes. Yeah. Yeah, that's one. I think the other one is also about the worker safety in the kitchen and I think I've also heard about they want to elevate the experience by saying when you walk in they know you from the profile in terms of what you would like from the menu out there. So it's all about enhancing the experience kind of it. I agree. Yeah.
Guy Daniels, TelecomTV (16:32):
Great, great. Another good use case and thanks for going through those two use cases. So we're going to move on to our next presenter, but round of applause for Chetan first. Thank you. Right now our next presentation is from Nico. Nico, I'd like to go to the lectern. Nico Marziliano is regional VP of Telco Sales at Wind River.
Nico Marziliano, Wind River (16:50):
Thank you very much. Good morning everyone for me is interesting moment to present after an intense day yesterday, a lot of intelligent people and more expert people than me definitely on the AI topic and I feel very honored to be invited here to give my vision on the AI chat and start talking about the potential few services that we can launch from the B2B. I'm going to present a few other use cases, but before I jump to the possible new services, let me spend a little bit of time on the, going back to the memory, I remember yesterday a friend from Dell, Eric was mentioning the calculator. Everybody remembered the old days having a calculator that was fantastic tools and now it's like an internal things. Now if you look at this picture for example, this is a skyline that looks like modern and all these connection points looks like the modern way, but in reality, if you remember for those a little bit experience like me running maybe a modem in a PC card, a PCMI card, whatever, 56 kilobytes, maximum speed, downloading a PowerPoint presentation for one meg from where friend which were killing you for an hour standard at the airport or the train station just to wait the mail to come.
(18:21):
The picture doesn't change from that days from the nineties and early 2000 up to now. The only difference is that probably the compute capacity of the device at the edge that is operated is probably equal to the same amount of capacity that has a data center at that time. So every single phones now has two fifty five hundred gig. That was probably the capacity of mini data center at that time. So there is a lot of change that happened since then and that is bringing to what we call it, I call it, and I like to define as the intelligent edge. So we are leaving and again chat and make a very good example of, and also the question from the audience about the restaurant or the coffee shop. So you were really pushing where is the edge ending? The edge is ending almost at the end device.
(19:18):
So if you think where the compute and the execution of the operation is happening is not anymore. As we used to design big large data center where everything was executed remotely but then computed centrally somewhere you sent to the cloud in a nowhere location. Now everything has to be executed where you do the activity. So cameras but also phones or whatever, any radio unit or anything that is there, it'll require to execute at a very edge or far edge. That is exactly why we call it the intelligent edge. And that is not only in term of technology, but it's also having an implication in term of what is the consumer expectation, what is the consumer that is expecting to receive not anymore to wait an hour to download an email. So they want to have a boom, boom, boom. So that is exactly driving also the technological requirement which is low latency compute capability, immediate reaction, trusted reaction.
(20:27):
And we will touch about this point later on. So if you think of that today we have probably already a kind of 30% of the traffic executed at the edge. We expect or the analyst report that by 2030 the compute will be 70% at the edge and everything just 30% remaining in the central cloud, which is an incredible change off. And if you imagine that every single camera is producing one tera byte, it's a massive number. So you cannot execute everything just going back has to be executed local. But that has a lot of implication and let me go back next week there will be a GTC event in DC so the guys over there on the picture will present again another fantastic GPU. Of course I'm not here to talk about Nvidia, but I'm here to say that if I think about, again back to the calculator example or to the modern transmission model, I mean everyone would say, oh that is impossible.
(21:39):
I was talking with a friend yesterday about someone having an opinion that the GPU cannot be run multiple tenant at the far edge. At the end of the day it's just a question of time, but technology is evolving and is evolving rapidly. So we are now in a situation where the technology is evolving so fast and so quick that finally you can realize a lot of the desired opportunity that we have been thinking before. Now yesterday we spent a lot of time on agentic AI, someone is demonstrating Mr. Elon Musk, even the physical AI and I'm expecting probably maybe tomorrow someone shows something similar.
(22:22):
We are moving toward that direction where the endpoint device at the intelligent edge will be, how can I say, being aware of the context where it's operating. I'm not saying that my phones will be waking up and taking the autonomous decision, but the end device will be more and more context aware and that's as a huge implication on all the use cases that we are going to see shortly. So if I move a little bit on the final point, oh well I forget to say one point in the previous one question from the audience before, I did not answer on your, but the energy is clearly one off. So there is a kind of paradox now with the acceleration of AI and especially with the agent AI introduction. So the cost transaction per inferior or per operation will be reduced significantly, but the amount of inference and the operation will increase massively.
(23:26):
So on one side we are reducing the cost per, but then there will be a massive implementation and utilization of the energy. So the energy will be clearly one of the challenge going forward and that's why the increase of new GPU NPU CPUs, all these new central unit will require capacity to save cost of energy at the verified edge. So that is a little bit one of the things that we need to be super conscious of that. Now back to the use case, which is the topic of today, listening to the colleagues yesterday, a lot of people ask it about is it about cost reduction, is it about revenue generation? What are the use cases? Someone was saying, oh, I've seen these use cases before probably yes. Honestly, even if I look at this picture on the screen, the use cases are very similar to the one that we're seeing probably 10 years ago, 15 years ago. It is nothing new again, the only difference is that technology now enable us to do it faster, quicker, better.
(24:42):
Someone yesterday was commenting about this all RAN AI, oh there is a cell phone organizing network talking about the TMF in 2005, probably 2004, so 20 years ago. So a lot of those teams has been already exploited and discussed and probably even implemented with different technology. Now we have a technology that is capable to do it faster, quicker, better, and eventually cheaper. As Danielle was saying yesterday in the BSS area, and I would say of course the consequence is not only economical but probably also from a user experience because what you see here, what we promise as industry that the customer will have a faster reaction, better control better. That was a kind of hypothesis. Now it's real and it is real because, and that's why I spend a little bit of time about technology before because now finally we have the technology that is supporting that.
(25:40):
So we have the underlying cloudification that is now enabling with a Kubernetes, with an orchestration layer with the agentic AI, with the possibility to have the silicone at our hands which will be accelerating and implemented that in the way that we expect if anyone say what is the upsell, what is the new use cases For me, these are the use cases we're working already with, especially with the aerospace and defense, which are very much on top of this transformation apti. And when river is working with the automotive, which is by the way, it's our domain and in the automotive, everybody's expecting the next, I call it the A mobile on wheels. So the cars will be the next mobile phones or the edge traveling around the cities. So everything will be really moving around the city as the new next AI device but not limited to, it can be flying drones as well recently.
(26:44):
There is a lot of talk about drones, but many others, everything will be interconnected. And of course last but not least, I think that yesterday there were few guys mentioned about sovereignty cloud. There are different opinion but definitely the way the data are executed at the edge and they're shared among, and I think about drones or especially in their space and defense has to be secured, private and with the right control layer. So otherwise there is no trust between the information that are shared. And so I would say use cases are not new. So that is a bad news for you guys, but the good news is that we can do it now finally, better, quicker, faster. Thank you very much.
Guy Daniels, TelecomTV (27:30):
Thank you very much indeed Nico, please join us again. Great. Fascinating. Thank you Nico. Any questions from our audience here in Dusseldorf? Any questions for Nico? Anyone want to raise hand? We've got one in the middle. Tony, can you see? Microphone is coming. Thanks George.
George Glass, TM Forum (27:53):
Thank you. So you're talking about AI at the edge, but everybody knows that AI consumes a massive amount of computing par. We've got the concept of a cloud continuum from the device to the edge to the cloud to give you compute and storage. Are you seeing the same then being able to leverage that capability but maybe having restricted AI on the device, enhanced AI at the edge and full AI in the cloud or some concept like that?
Nico Marziliano, Wind River (28:25):
Interesting question. I was thinking myself the same question to me, I was making the same question to me. I mean is it already available? I mean if you think about your phones, everyone, every major brand has AI including your phone. Is it this AI at the device level that can be in a car tomorrow or anywhere? It's how that AI works with the other AI. I think that definitely your point is valid. That should be a kind of desegregation of the high level, but that again goes back to some of the technological layer that are available now but not yet fully implemented, which is data lakes and federated data and the way that the AI can interfere by proximity and within a group of logic audience or a limited audience. It depend on the use case that you want to and then you may distribute the AI capability limited to the far edge device. And of course if you talk about the cars or mobile phones has to be for a user that has a certain expectation in term of use case, if you talk about the robotic robot harms, probably you don't need that sophisticated AI or complex AI.
Guy Daniels, TelecomTV (29:39):
Great, thanks Nico. Thanks George for the question. We got another one here. Dean?
Dean Bubley, Disruptive Analysis (29:43):
What's your view on connectivity in AI being able to work offline because a lot of the use cases you talked about have unpredictable connectivity. This room is unusual. We're in an underground conference room that actually has decent in-building
Nico Marziliano, Wind River (30:02):
Indeed.
Dean Bubley, Disruptive Analysis (30:02):
Wireless
Nico Marziliano, Wind River (30:03):
indeed.
Dean Bubley, Disruptive Analysis (30:03):
But most basements don't. Lots of things that 400 meters out for drone, it's a bit variable. Anything with virtual reality is going to be indoors not on the macro 5G network. How do you design for unpredictable connectivity?
Nico Marziliano, Wind River (30:21):
That's a very good question and thanks for giving me the opportunity to talk a little bit about windriver as I omitted that definitely the Kubernetes layer. So the underlying layer that is creating the connection between the different cloud and sub cloud and the far edge require stability. Now you need to have an architecture to your point that is able to be self-healing and independent from any control layer that is organized in the cloud. Otherwise, if you lose the endpoint then there is no more connection and especially if that is a mission critical, then the use case is closed and then you are out of control. So you need to secure that Your far, far edge, sorry, is always able to self feel and to be independent from the controller and when the connection is established is reconnecting back, but guarantee zero downtown and that is exactly what our real time operating system is typically doing.
Guy Daniels, TelecomTV (31:22):
Great, thank you.
Chetan Narang, Colt Technologies (31:24):
No, was just about to say that. I think just to add to that, I think there are air gap deployments. Basically you have to do it because use cases like virtual reality and from a privacy perspective you need to have that kind of it. But yeah, I think as Nico said, I think it's all about connecting back when you need from a control plane perspective. Yeah,
Guy Daniels, TelecomTV (31:42):
Great. Thanks for the question. We've got to move on unfortunately. You okay? Yeah, we'll pick up later. Fantastic. Thank you very much. Round applause for Nico. Thank you very much Nico. Thank you. Right. Well our third and final presentation of this session is going to be given by Dario, Dario Sabella is chairman of the ETSI MEC, ETSI MEC and also the VP of xFlow Research. So Dario, over to you.
Dario Sabella, ETSI MEC & xFlow Research (32:09):
Thank you so much. Good morning everybody. I'm very happy to be here and also very relaxed because the previous speakers made my life easier. So maybe there was the presentation as an example for operators using the edge and also the presentation regarding the architecture view of the edge computing, different level of deepness. So I can relax because some of the aspects are tough questions that may be made to me. Actually we have already the answers, right? So here today I'm representing a standard group, an international standard body in the umbrella of ETSI, standardizing the edge computing. Multi X edge computing means of course that we are targeting different kind of access technologies. So not only mobile but also fixer networks and also wifi and also satellite if you believe. So I'm representing as a chairman of this standard group and I will give you my preference sometimes to save time is just to present you and reveal the conclusions so that also you can be relaxed and you start not sleeping but you are relaxed because if we are in lack of time, maybe you already captured my messages.
(33:29):
Okay, my main messages are we need to understand better also through the real fundamental values of the edge. It's not just shortening the path between a client and server. So let me call it a real estate benefit of the edge, which is physiological by deployment but also the nature, exploring the nature of the edge, capturing the data where this data is generated and then capturing the value for new service creation. And of course as a matter of fact, edge is reporting multiple use cases. We are talking about the standardization. We are of course very careful in standardizing what is really needed for the edge to open the market because the temptation would be to over standardize, over define everything, but actually we need also to open the market and leave the possibility to have appropriate implementation also from many companies. So it was a very delicate work done by many Tors in this group, which I'm pleased to be the chairman.
(34:37):
And also with a clear focus on the developer needs, I would call it developer-centric approach, having APIs in mind, software development activities in parallel to the practical standard production. So not just only paper and PDF file readable by humans, but also software readable by machines and automated test suites for conformance and compliance and also hackathon competitions to exploit these infrastructure and this standardization to help them to use and adopt this technology and practice via technical competitions in hackathons. So also Mac as I said, is an access agnostic technology and of course the standard should be made in this perspective in an access technology way, agnostic way. So also the very important part is that our critical work is on API exposure and data and privacy and security and data governance is a critical aspect also in the future for the regulation for example. And last month the least AI at the edge is coming because of course we are all seeing AI in the big clouds also with a little bit smaller footprint.
(35:55):
It's also very much affordable, started to be achievable also the usage of the edge for enterprises in a so called continuum between cloud edge and iot. Okay, so just a quick overview starting from the edge. What I'm talking about when we talk about the edge computing, actually there are many flavors of the edge, many level of deepness as we have seen with Nico. And actually we are from application developer point of view. What matters is the application and point. So we are adding in a treaty approach are sort of intermediate place where there can be some workload running at the edge and consuming services from the edge platform and also producing services. We will see. So actually there are a lot of use cases. We are not the experts of the use cases in et cmec. We are the experts on the edge. We have to listen at the vertical markets, understand their needs, their pain points, and start a dialogue with those guys.
(36:59):
So that's why we started a sort of events which we call it the edge discovery events, where we wanted to start a dialogue with the various experts on a certain market domain. So for example, drones, we organized a workshop with the experts for this kind of industry to understand why and where is appropriate to use the edge, what is the value for the edge from their market perspective to listen at their needs and understand what is pertinent to be standardized in the framework of ETSI MEC. Again, special computing and gaming is also call it as a metaverse but also cyber security aspects. So very vertical segments, very much important where we want to listen at them to understand what is pertinent to be standardized in ETSI MEC. So what is MEC, ETSI MEC? ETSI is a big umbrella of standardization where edge computing is being introduced by MEC, M-E-C and ISG, which is a group counting more than 200 members and participants, right?
(38:03):
It's a big group, very heterogeneous, well very well populated and growing in membership, not just because I'm a beautiful guy. Well not only that but because this is a sign that of course the edge computing market is growing and then also it is very important to operate. It's a complex value chain. The standard is critical when you need interoperability and global adoption. So that's why ETSI MEC was created like 10 years ago and starting from the definitions of the virtualization of the network, those of you who are experts on NFV knows what I'm talking about. But it's going beyond much more beyond because it's actually upper layer at application level we target the make application development communities developers, but the open standard approach is a must for us. We need to keep open to various kind of deployments, various verticals and not only just the mobile network.
(39:07):
Well let me clarify. 5G is a mainstream. I can tell you I'm a bit proud of my delegates. It was a more than three years work in alignment between ETSI, me three GPP SA six and CT three to align three GPP release 18 with MEC phase three, you maybe are not so technical you cannot appreciate the detail of that. But I can tell you that if you have a 5G network with release 18 equipment and network, you have also the possibility to interoperate with the MEC platform. There is a MEC profile for the cap for the common API framework. And this is actually very much impactful achievement that we have done is a joint effort between multiple standardization bodies. So ETSI MEC is seriously alignable with three GPP and 5G networks. But not just only that. Also wifi networks and fixed networks. That's why it's edge computing.
(40:05):
Cloud computing at the edges standardized and also we are focusing on, well this is the history of what we have done in the various phases of the standardization you can forget, but important thing what we are introducing is a set of APIs talking the commonly universally accepted language, which is restful HTTP comments. We see in all our browsers in the internet we are standardizing some APIs. You can see here some examples requested by the industry. But also it's important that also companies can add their own APIs. They don't have to knock at the door of ETSI MEC to make them available to the platform. There is an automatic mechanism to automatically recognize new services present from the application. And this is advertised to all authorized applications in the platform. So it's an open approach in our standard that permits to increase the portfolio of services.
(41:07):
So you don't have to standardize everything. We are very much open also to open source communities. I will tell you about CAMARA collaboration. We have the Linux Foundation in CAMARA project. So actually we are doing also a lot of work because we recognize that developers, they don't spend too much time reading specifications, but they need also to play with software and they need to practice. So a set of complimentary activities. We created a working group years ago dealing with MEC deployment and ecosystem engagement activities ranging from production of software and open AI representation in open source testing and conformance proof of concepts, trials with live networks from multiple operators. Also in a Mac Federation, Mac Sandbox, which is a sort of reference platform for software developers that can practice about the usage of the APIs and also collaboration with open source communities, hackathons, plug test for interoperability and also some sort of podcast tutorials to explain the difficult parts of the standards.
(42:18):
So this is just an example of a few APIs that we've published, but I would like also to give you some pointer if you are interested about the pragmatic approach that we have. Really we wanted to clarify this huge space of APIs. Everybody today is talking about APIs, especially now that we are talking about AI. Consumers of APIs will be also AI agents and they will be AI to AI communication via APIs. APIs are important today. They will become even more important in the future through MCP and A2A, any kind of protocol and gateway that serves to communicate in terms of that exchange between entities, consumer and producer of services. And then there is a lot of confusion in terms of APIs. Should I use a standardized API? Should I use an open source a p? I can produce my own proprietary API. If I'm a developer, I am really confused about that.
(43:13):
So maybe if you want to learn more, there is this kind of white paper written by delegates from ETSI MEC and CAMARA and the TM Forum delegates together to shed some light, possibly give some guidance humbly to developers to let's say deep dive into this kind of universe of APIs with different flavors, how they can fit in the bigger picture, how they can complement to each other and how they can be consumed with some commonalities and some possibility to expose the services in an interoperable way. So moving forward, we are working also seriously in the space of AI, which is actually many domains. In some domains, we are seeing a lot of potential for the future and for example, abstraction of APIs sensor sharing, but also for the security part, for the security monitoring management, the cm, the API gateway. And also there is a working item, especially focused on artificial intelligence at the MEC and also other, there was someone maybe Nico talking about let's say constrained devices. Also possibility to have in the future more and more workloads that are also in unpredictable connectivity or connected remotely at the far edge. So distributed networks and IoT deployments. This is a very important flagship project that we have. And yeah, with that I believe I conclude my presentation, we are looking forward to see more advancement in the standards and with more work on AI and edge native design support for application developers. Thank you so much. Thank you very much.
Guy Daniels, TelecomTV (45:02):
Your day sit down. Great, thanks so much Dar. Lot of information there. Thank you very much. Do we have a quick question for Dario? I neglected you earlier, Francis, so let's quickly get Francis and I think there's another question over here. Let's get to him. Yep. Thank you Francis. The gent over in the second row there Tony. Yeah, we'll find someone for you eventually.
Audience Member (45:23):
Thank you. Thank you very much for the presentation. Sorry for my voice. So basically my question is regarding ETSI effort in term of future A2A and MCP standardizations. If there are any effort on that, let me give a little bit of context for that. So basically if we see yesterday for the presentation we have this agent AI layer, which is very important and if we look to what we have seen that we have multiple AA, A2A implementation from Google, IBM and others and so on. So how to ensure interopability and open ecosystem within agentic AI layer and what are its effort in this regard beyond API.
Guy Daniels, TelecomTV (46:12):
Great. Great question.
Dario Sabella, ETSI MEC & xFlow Research (46:12):
Thank you. Thank you for the question. Of course, it's difficult for me to give you an answer because there are some decisions which are ongoing. But I can tell you that the group in Mecca and driving, I'm pushing them to evolve to what I call Friendly Mac 2.0 strategy to evolve the group. And I've also sensed the temperature from companies. I see also many inputs for new ideas for the topics, for the future standardization. And also I see a lot of interest on edge AI and these topics are really on the radar. I truly agree with you. We need for interoperability in this kind of new set of technologies and the standards will play a key role. I'm driving this group to migrate into a technical committee to announce the footprint and have more impact also for European regulation when there is a lot of issue of course for companies to be compliant with and a lot of unclear situations. So the standard may help on that and I believe we are considering this really, it's on the radar, I cannot tell you now, but maybe next conference I can give you some updates on the standards where we are going.
Guy Daniels, TelecomTV (47:18):
Fantastic. Thanks Dario. That's great. We're out of time, so any more questions, maybe we can lead them for our coffee break, which is coming up in about 30 minutes or so. So I'm going to start our next session now, but please a round of applause for our three guests now.
Time I think to start day two and we have a big focus this morning on AI at the edge. So first of all, if I could invite the presenters for the first session onto stage. We've got Chetan, Nico, and Dario, so if the three of you could come and join me on stage and then we'll start our presentations. So give them a round of applause while I'll make their way to the stage. So Chetan is going to start us off. So Chetan, if I can ask you to go towards a lectern.
Chetan Narang, Colt Technologies (00:46):
Good morning everyone. I hope you guys had a great day yesterday and it's a pleasure to open the event today. So Colt edge. Just a quick introduction. Chetn Narang, I lead product innovation at Colt, focus on AI product strategy and I'm here to talk about Colt Edge AI. Just a quick thing, I think there's a short agenda. I think I just go through a quick Colt overview, talk about the Colt edge propositions, what we have and then just going to give kind of what we are doing in the Edge AI space in terms of prototyping as part of the new services. So just a quick word on Colt, Colt is a global digital infrastructure company. We are mainly a fixed line service provider focused on I think having a global network. We one of the largest B2B digital infrastructure platform. We mainly serve enterprises. We have around 20,000 plus customers around presence in 40 countries and around 230 cities out there.
(01:55):
Our network is 400 gig plus in terms of the core infrastructure. We have also trialing 800 gig kind of a capacity. That's basically an evolution of the network we see. I think also from a strength of our network we have are one of the most connected network in terms of connecting to 1100 plus data centers, 31,000 buildings and around two 50 plus on-ramp capabilities kind of it. We also have an award-winning NaaS platform which is about digitization of how do you consume network services, kind of it going towards the edge side. I think we have two aspects of edge propositions currently which are quite focused on network functions. The first one on the right-hand side is the network edge, which is mainly focused on multi-services telco cloud. We have a deployment footprint of around 200 plus Colt owned VNF I want to highlight. It's called owned VNF.
(02:59):
We haven't opened to the partners or customers yet, so it's pretty much Colt estate right now deployed across 12 strategic geo cities which aligns in terms of where our network presence is. It's a multi-tenant architecture and we run thousands of services across. I think it's a very dynamic scalable platform and it's optimized for TCO and accelerated ROI. I think we want to have a practical approach to it but also want to ensure that it's scalable and high performing. On the left-hand side we see the on-premise edge. The on-premise edge is about a hundred plus enterprise site deployments and I think this shows in terms of what we have in motion from a distributed scale and field services kind of it at the same point I would like to also highlight I think on-premise edge is not for every use case, it's for critical enterprise use case and I think one of the main use case that drives this is mainly the security aspect of it and I think we see the strategy growth enablers from this, which is part of a zero trust access in partnership with Zscaler.
(04:11):
We also see Quantum Safe networks as the other basis. And then lastly about Edge AI, Edge AI also is one of the main perspective now just coming back to the topic, what we have over here, the edge AI aspect of it. I think we've been also going through and hypothesis in terms of how do we evolve these capabilities from an AI perspective and as we've seen in terms of what's happening in the industry, we see that, I think there are like three ways of monetizing this. I think the first one is basically you bundle the hardware and the software go in the space of GPU as a service kind of a construct and I think you try to position more like network edge is where do you do local data handling and you can also have latency wins from a perspective of network and compute when you compare to the cloud side of the business.
(05:02):
The other aspect is also from an on-premise edge is basically you can have on-premise deployments where you feel like the response time is critical and you need a quick response from a decision-making perspective. So you try to position on the on-premise head side of a business, that's one. The second one is basically you evolve from the hardware and the software towards more towards the platform space. This is basically doing the AI lifecycle aspect of it, starting from data ingestion to training and taking to the inferencing at a scale and maybe going also to a model as a service, kind of a construct over there. And the last one is mainly towards outcome, deliver outcomes to the customers and it also goes back to the debate what we have in industry telcoms going from a telco to a TechCo kind of a concept over there. I think on that approach we have a twist. I think we also see there's an opportunity from a partnership perspective you work with a wider ISV ecosystems to deliver outcomes to the customers and as part of that in the subsequent slides I'm going to talk about two of the couple of engagements. What I've been doing with partners to develop kind of an Edge AI proposition. This is mainly focused on the on-premise aspect of it. I think the first one is basically the Colt Edge AI for smart buildings.
(06:26):
It's an Edge-first AI, AI IoT. What it means is basically the data processing and the AI models in this case are running at the edge with cloud mainly used for data gathering, running analytics and also mainly providing training of an AI ML model either way. So how it's been positioned is basically deploy a Colt edge on the on-prem, you connect to A BMS system, you try to put sensors, gather that census, get it co-location into the Colt edge and then you push that data into the cloud and we've been partnering over here with the cloud company with the smart building solution company, which is nuca over here and once that data has been pushed into the cloud, the data analysis and the models are trained on the cloud and then they're pushed into the edge to provide predictive maintenance, remediation and optimization kind of an aspect to it.
(07:24):
It's also an example of a good AI because there's always a debate about good versus bad AI and it's a good example of innovation for sustainability. The interesting aspect is also this is one of the use case where AI ML doesn't need GPUs. It's a very compute intensive, it's just running on cpu so it provides that innovation for sustainability aspect. We also have one, the code, I think it aligns with the Colt ESG values. We are quite strong, we're quite vocal on the ESG perspective. It's one of our main pillars. It also gives us, I think we've also won an eco innovation award with a partner with a customer partner Bertelsmann there, so that's one use case. The second use case which I'm going to touch on is basically the edge AI kind of an aspect which is more towards the street talk. It's all about building edge on the street.
(08:16):
This example shows of a rack or an edge being deployed in London in show just next to Colt House. We've been working with partnership over here with CIN, which is a startup working on this. I think the basic principle in this case is basically to have Colt edge AI for public square and this public square can be anywhere. It can be in the middle of a street, it can be in a campus, it can be in an airport, it can be anywhere where you have free space to give that capabilities from a concept. This is all about building infrastructure stack, which is rack coming up from CIN connectivity comes from Colt and then the Colt edge also part of the stack and then it's all about how do you enable those use cases on top of it. And this is where it comes to having an outcome led approach and having a wider part ecosystems.
(09:13):
We have been working with few of the partners in terms of evaluating this. I think some of the use cases, what we have looked is like a private 5G fixed wireless access with Weaver Labs, Edge gaming with EdgeGap, Canadian based company. It's all about how do you give the gamers and access to an edge based upon geo proximity kind of aspect. Wifi off flowed from a global reach perspective, from an AI perspective, what we have done in this case, we have worked on AI video analytics with a partner called AICuda. There's also a value from an AI perspective to run analytics over there and that analytics can be across demographics of what people walking in, how people are moving in. It can be on the basis of traffic. It can also be based on the environment and once you run this analytics, the idea is that you can also monetize it by providing insight for example to the retailers which they can use the data to tailor their services in that way. That is just one example of how we can do it over there. Yeah, pretty much. I think we are still on a journey. It's a new subject for us. It's evolutionary, but I'm happy to take any questions as part of the panel discussions. Great, thank you for listening. Thank you
Guy Daniels, TelecomTV (10:29):
Very much indeed. Chetan, come and sit down here over this side come in fact pass that over. Great, thank you very much. That was great. We're going to get to see if there's any questions yet. I'm sure there's a question or two you'd like to put to Chatan, so let's have a look at hands and also looking for the roving microphone which is, oh here comes Tony. Tony's going to have to run the length of our ballroom across to Francis here. Well done Tony intercepted. Dean, nicely caught.
Dean Bubley, Disruptive Analysis (11:01):
Sorry about that. I really like that presentation. I've said for a while that I think there's more synergy of edge compute with fixed operators than there is with mobile. Are you pushing this towards other industry verticals? You mentioned the smart building, but are you also looking at this for industry, maybe transportation sites, other things?
Chetan Narang, Colt Technologies (11:30):
Yeah, so I think definitely from an edge perspective there's a vertical play I think I agree with over there. So I think the couple of industries which we are trying to focus is one is manufacturing, which comes from a perspective of computer vision being the main use case from a predictive maintenance kind of an aspect in that way. And also I think retail, that's another thing, how do you do shelf analytics, smart retailing, kind of a point of sale and wind pre analysis. I think that's another use case which we are trying to target from perspective.
Dean Bubley, Disruptive Analysis (12:01):
One quick follow up. How much power does that...
Guy Daniels, TelecomTV (12:03):
In case anyone didn't hear that? How much power?
Chetan Narang, Colt Technologies (12:08):
I think so at this point I think this, I don't exactly know what is it, but it is a CP intensive. We haven't deployed GPUs in that way, but the current analysis, what we have is that you don't have to go with massive GPUs. Also kind of if you think from an on-premise edge and if you say it's dedicated to the customers, you can even go with Nvidia RTX two thousands which are competitive from a pricing perspective and is also enough like 70 watts. I think if I can, don't quote me over here, but maybe just top of my head, I think it's low from a power utilization perspective it should be enough. I think when you go towards the network edge you talk about massive GPUs like the 6,000 pros and all and that's where you're looking about 15 kilowatts per rack and that's a total game in that sense anyway.
Guy Daniels, TelecomTV (12:55):
Great, great question. Thank you Francis.
Francis Haysom, Appledore Research (12:57):
Very quick question. You showed two really great use cases there. Have you at all looked at what I would term is what you're going to need to do to scale them, make the at scale business case for them because each of those has quite a lot of investment in IT and payback and are you looking at all in terms of that sort of ROI model in terms of what you're doing with these cases?
Chetan Narang, Colt Technologies (13:24):
Yeah, I think it's an interesting one because so for smart building I think we are trying to prioritize that and take it back to get our different markets which are quite sustainable conscious like Singapore is one of the markets which we want to target and from a perspective of green buildings kind of a concept I think in this, I think it, it's a journey for us because this is not an easy, if it's a connectivity we controlled asset and it can position it in that way. This is more of an ecosystem play. I think the real win will be if we can evolve an ecosystem, basically this is one use case but maybe if you can have certified applications or certified aligned to the use case so then you can say these are the outcomes you can deliver. That's where I think we can get the scale in a way because one solution or one partner will not fit the fifth answer or an outcome to the customer.
Guy Daniels, TelecomTV (14:19):
Great. Thanks very much and great question. Thank you very much. Do we have another question from our audience? We can probably slide another question in if we're fast. Well I've got a quick We do, we do. Terry, thank you.
Terry Riley, Spectro Cloud (14:32):
Hi Chetan, Terry Riley. We are seeing edge being deployed in lots of restaurants. Is that a use case that you are coming across?
Chetan Narang, Colt Technologies (14:44):
Yeah, I think I've heard about it. There's a lot of restaurants also from a perspective of AI. I think I've seen it out there. It's just like I think our focus is mainly B2B, which is more towards enterprises and professional services. I think we haven't gone into that kind of a use case, but I certainly think there's a value over there. Yeah.
Terry Riley, Spectro Cloud (15:06):
You say B2B, the customer I'm talking about is a large global coffee retailer with 53,249 stores. That's a business in my opinion.
Chetan Narang, Colt Technologies (15:17):
Yes, I know about who you're talking about. Yes, so I think are think in fact I know two names come across in that vertical out there. I think it's an interesting phase because that's another thing of positioning out there because if you go with, they're like DIY approach in a way. Basically they want to build this up and they want to have it. I think we would like to position themselves in where we can support the journey aspect of it. But having said that, I think that's more about the market segmentation, but I agree restaurant is a business and if you talk about that kind of a scale, yes, and I'm aware there are use cases, a couple of use cases on that way we can do it.
Terry Riley, Spectro Cloud (16:00):
Yeah, the main driver is compute vision you say, but it's looking for length of the queue and how quickly you start getting cofee.
Chetan Narang, Colt Technologies (16:08):
Yes. Yeah. Yeah, that's one. I think the other one is also about the worker safety in the kitchen and I think I've also heard about they want to elevate the experience by saying when you walk in they know you from the profile in terms of what you would like from the menu out there. So it's all about enhancing the experience kind of it. I agree. Yeah.
Guy Daniels, TelecomTV (16:32):
Great, great. Another good use case and thanks for going through those two use cases. So we're going to move on to our next presenter, but round of applause for Chetan first. Thank you. Right now our next presentation is from Nico. Nico, I'd like to go to the lectern. Nico Marziliano is regional VP of Telco Sales at Wind River.
Nico Marziliano, Wind River (16:50):
Thank you very much. Good morning everyone for me is interesting moment to present after an intense day yesterday, a lot of intelligent people and more expert people than me definitely on the AI topic and I feel very honored to be invited here to give my vision on the AI chat and start talking about the potential few services that we can launch from the B2B. I'm going to present a few other use cases, but before I jump to the possible new services, let me spend a little bit of time on the, going back to the memory, I remember yesterday a friend from Dell, Eric was mentioning the calculator. Everybody remembered the old days having a calculator that was fantastic tools and now it's like an internal things. Now if you look at this picture for example, this is a skyline that looks like modern and all these connection points looks like the modern way, but in reality, if you remember for those a little bit experience like me running maybe a modem in a PC card, a PCMI card, whatever, 56 kilobytes, maximum speed, downloading a PowerPoint presentation for one meg from where friend which were killing you for an hour standard at the airport or the train station just to wait the mail to come.
(18:21):
The picture doesn't change from that days from the nineties and early 2000 up to now. The only difference is that probably the compute capacity of the device at the edge that is operated is probably equal to the same amount of capacity that has a data center at that time. So every single phones now has two fifty five hundred gig. That was probably the capacity of mini data center at that time. So there is a lot of change that happened since then and that is bringing to what we call it, I call it, and I like to define as the intelligent edge. So we are leaving and again chat and make a very good example of, and also the question from the audience about the restaurant or the coffee shop. So you were really pushing where is the edge ending? The edge is ending almost at the end device.
(19:18):
So if you think where the compute and the execution of the operation is happening is not anymore. As we used to design big large data center where everything was executed remotely but then computed centrally somewhere you sent to the cloud in a nowhere location. Now everything has to be executed where you do the activity. So cameras but also phones or whatever, any radio unit or anything that is there, it'll require to execute at a very edge or far edge. That is exactly why we call it the intelligent edge. And that is not only in term of technology, but it's also having an implication in term of what is the consumer expectation, what is the consumer that is expecting to receive not anymore to wait an hour to download an email. So they want to have a boom, boom, boom. So that is exactly driving also the technological requirement which is low latency compute capability, immediate reaction, trusted reaction.
(20:27):
And we will touch about this point later on. So if you think of that today we have probably already a kind of 30% of the traffic executed at the edge. We expect or the analyst report that by 2030 the compute will be 70% at the edge and everything just 30% remaining in the central cloud, which is an incredible change off. And if you imagine that every single camera is producing one tera byte, it's a massive number. So you cannot execute everything just going back has to be executed local. But that has a lot of implication and let me go back next week there will be a GTC event in DC so the guys over there on the picture will present again another fantastic GPU. Of course I'm not here to talk about Nvidia, but I'm here to say that if I think about, again back to the calculator example or to the modern transmission model, I mean everyone would say, oh that is impossible.
(21:39):
I was talking with a friend yesterday about someone having an opinion that the GPU cannot be run multiple tenant at the far edge. At the end of the day it's just a question of time, but technology is evolving and is evolving rapidly. So we are now in a situation where the technology is evolving so fast and so quick that finally you can realize a lot of the desired opportunity that we have been thinking before. Now yesterday we spent a lot of time on agentic AI, someone is demonstrating Mr. Elon Musk, even the physical AI and I'm expecting probably maybe tomorrow someone shows something similar.
(22:22):
We are moving toward that direction where the endpoint device at the intelligent edge will be, how can I say, being aware of the context where it's operating. I'm not saying that my phones will be waking up and taking the autonomous decision, but the end device will be more and more context aware and that's as a huge implication on all the use cases that we are going to see shortly. So if I move a little bit on the final point, oh well I forget to say one point in the previous one question from the audience before, I did not answer on your, but the energy is clearly one off. So there is a kind of paradox now with the acceleration of AI and especially with the agent AI introduction. So the cost transaction per inferior or per operation will be reduced significantly, but the amount of inference and the operation will increase massively.
(23:26):
So on one side we are reducing the cost per, but then there will be a massive implementation and utilization of the energy. So the energy will be clearly one of the challenge going forward and that's why the increase of new GPU NPU CPUs, all these new central unit will require capacity to save cost of energy at the verified edge. So that is a little bit one of the things that we need to be super conscious of that. Now back to the use case, which is the topic of today, listening to the colleagues yesterday, a lot of people ask it about is it about cost reduction, is it about revenue generation? What are the use cases? Someone was saying, oh, I've seen these use cases before probably yes. Honestly, even if I look at this picture on the screen, the use cases are very similar to the one that we're seeing probably 10 years ago, 15 years ago. It is nothing new again, the only difference is that technology now enable us to do it faster, quicker, better.
(24:42):
Someone yesterday was commenting about this all RAN AI, oh there is a cell phone organizing network talking about the TMF in 2005, probably 2004, so 20 years ago. So a lot of those teams has been already exploited and discussed and probably even implemented with different technology. Now we have a technology that is capable to do it faster, quicker, better, and eventually cheaper. As Danielle was saying yesterday in the BSS area, and I would say of course the consequence is not only economical but probably also from a user experience because what you see here, what we promise as industry that the customer will have a faster reaction, better control better. That was a kind of hypothesis. Now it's real and it is real because, and that's why I spend a little bit of time about technology before because now finally we have the technology that is supporting that.
(25:40):
So we have the underlying cloudification that is now enabling with a Kubernetes, with an orchestration layer with the agentic AI, with the possibility to have the silicone at our hands which will be accelerating and implemented that in the way that we expect if anyone say what is the upsell, what is the new use cases For me, these are the use cases we're working already with, especially with the aerospace and defense, which are very much on top of this transformation apti. And when river is working with the automotive, which is by the way, it's our domain and in the automotive, everybody's expecting the next, I call it the A mobile on wheels. So the cars will be the next mobile phones or the edge traveling around the cities. So everything will be really moving around the city as the new next AI device but not limited to, it can be flying drones as well recently.
(26:44):
There is a lot of talk about drones, but many others, everything will be interconnected. And of course last but not least, I think that yesterday there were few guys mentioned about sovereignty cloud. There are different opinion but definitely the way the data are executed at the edge and they're shared among, and I think about drones or especially in their space and defense has to be secured, private and with the right control layer. So otherwise there is no trust between the information that are shared. And so I would say use cases are not new. So that is a bad news for you guys, but the good news is that we can do it now finally, better, quicker, faster. Thank you very much.
Guy Daniels, TelecomTV (27:30):
Thank you very much indeed Nico, please join us again. Great. Fascinating. Thank you Nico. Any questions from our audience here in Dusseldorf? Any questions for Nico? Anyone want to raise hand? We've got one in the middle. Tony, can you see? Microphone is coming. Thanks George.
George Glass, TM Forum (27:53):
Thank you. So you're talking about AI at the edge, but everybody knows that AI consumes a massive amount of computing par. We've got the concept of a cloud continuum from the device to the edge to the cloud to give you compute and storage. Are you seeing the same then being able to leverage that capability but maybe having restricted AI on the device, enhanced AI at the edge and full AI in the cloud or some concept like that?
Nico Marziliano, Wind River (28:25):
Interesting question. I was thinking myself the same question to me, I was making the same question to me. I mean is it already available? I mean if you think about your phones, everyone, every major brand has AI including your phone. Is it this AI at the device level that can be in a car tomorrow or anywhere? It's how that AI works with the other AI. I think that definitely your point is valid. That should be a kind of desegregation of the high level, but that again goes back to some of the technological layer that are available now but not yet fully implemented, which is data lakes and federated data and the way that the AI can interfere by proximity and within a group of logic audience or a limited audience. It depend on the use case that you want to and then you may distribute the AI capability limited to the far edge device. And of course if you talk about the cars or mobile phones has to be for a user that has a certain expectation in term of use case, if you talk about the robotic robot harms, probably you don't need that sophisticated AI or complex AI.
Guy Daniels, TelecomTV (29:39):
Great, thanks Nico. Thanks George for the question. We got another one here. Dean?
Dean Bubley, Disruptive Analysis (29:43):
What's your view on connectivity in AI being able to work offline because a lot of the use cases you talked about have unpredictable connectivity. This room is unusual. We're in an underground conference room that actually has decent in-building
Nico Marziliano, Wind River (30:02):
Indeed.
Dean Bubley, Disruptive Analysis (30:02):
Wireless
Nico Marziliano, Wind River (30:03):
indeed.
Dean Bubley, Disruptive Analysis (30:03):
But most basements don't. Lots of things that 400 meters out for drone, it's a bit variable. Anything with virtual reality is going to be indoors not on the macro 5G network. How do you design for unpredictable connectivity?
Nico Marziliano, Wind River (30:21):
That's a very good question and thanks for giving me the opportunity to talk a little bit about windriver as I omitted that definitely the Kubernetes layer. So the underlying layer that is creating the connection between the different cloud and sub cloud and the far edge require stability. Now you need to have an architecture to your point that is able to be self-healing and independent from any control layer that is organized in the cloud. Otherwise, if you lose the endpoint then there is no more connection and especially if that is a mission critical, then the use case is closed and then you are out of control. So you need to secure that Your far, far edge, sorry, is always able to self feel and to be independent from the controller and when the connection is established is reconnecting back, but guarantee zero downtown and that is exactly what our real time operating system is typically doing.
Guy Daniels, TelecomTV (31:22):
Great, thank you.
Chetan Narang, Colt Technologies (31:24):
No, was just about to say that. I think just to add to that, I think there are air gap deployments. Basically you have to do it because use cases like virtual reality and from a privacy perspective you need to have that kind of it. But yeah, I think as Nico said, I think it's all about connecting back when you need from a control plane perspective. Yeah,
Guy Daniels, TelecomTV (31:42):
Great. Thanks for the question. We've got to move on unfortunately. You okay? Yeah, we'll pick up later. Fantastic. Thank you very much. Round applause for Nico. Thank you very much Nico. Thank you. Right. Well our third and final presentation of this session is going to be given by Dario, Dario Sabella is chairman of the ETSI MEC, ETSI MEC and also the VP of xFlow Research. So Dario, over to you.
Dario Sabella, ETSI MEC & xFlow Research (32:09):
Thank you so much. Good morning everybody. I'm very happy to be here and also very relaxed because the previous speakers made my life easier. So maybe there was the presentation as an example for operators using the edge and also the presentation regarding the architecture view of the edge computing, different level of deepness. So I can relax because some of the aspects are tough questions that may be made to me. Actually we have already the answers, right? So here today I'm representing a standard group, an international standard body in the umbrella of ETSI, standardizing the edge computing. Multi X edge computing means of course that we are targeting different kind of access technologies. So not only mobile but also fixer networks and also wifi and also satellite if you believe. So I'm representing as a chairman of this standard group and I will give you my preference sometimes to save time is just to present you and reveal the conclusions so that also you can be relaxed and you start not sleeping but you are relaxed because if we are in lack of time, maybe you already captured my messages.
(33:29):
Okay, my main messages are we need to understand better also through the real fundamental values of the edge. It's not just shortening the path between a client and server. So let me call it a real estate benefit of the edge, which is physiological by deployment but also the nature, exploring the nature of the edge, capturing the data where this data is generated and then capturing the value for new service creation. And of course as a matter of fact, edge is reporting multiple use cases. We are talking about the standardization. We are of course very careful in standardizing what is really needed for the edge to open the market because the temptation would be to over standardize, over define everything, but actually we need also to open the market and leave the possibility to have appropriate implementation also from many companies. So it was a very delicate work done by many Tors in this group, which I'm pleased to be the chairman.
(34:37):
And also with a clear focus on the developer needs, I would call it developer-centric approach, having APIs in mind, software development activities in parallel to the practical standard production. So not just only paper and PDF file readable by humans, but also software readable by machines and automated test suites for conformance and compliance and also hackathon competitions to exploit these infrastructure and this standardization to help them to use and adopt this technology and practice via technical competitions in hackathons. So also Mac as I said, is an access agnostic technology and of course the standard should be made in this perspective in an access technology way, agnostic way. So also the very important part is that our critical work is on API exposure and data and privacy and security and data governance is a critical aspect also in the future for the regulation for example. And last month the least AI at the edge is coming because of course we are all seeing AI in the big clouds also with a little bit smaller footprint.
(35:55):
It's also very much affordable, started to be achievable also the usage of the edge for enterprises in a so called continuum between cloud edge and iot. Okay, so just a quick overview starting from the edge. What I'm talking about when we talk about the edge computing, actually there are many flavors of the edge, many level of deepness as we have seen with Nico. And actually we are from application developer point of view. What matters is the application and point. So we are adding in a treaty approach are sort of intermediate place where there can be some workload running at the edge and consuming services from the edge platform and also producing services. We will see. So actually there are a lot of use cases. We are not the experts of the use cases in et cmec. We are the experts on the edge. We have to listen at the vertical markets, understand their needs, their pain points, and start a dialogue with those guys.
(36:59):
So that's why we started a sort of events which we call it the edge discovery events, where we wanted to start a dialogue with the various experts on a certain market domain. So for example, drones, we organized a workshop with the experts for this kind of industry to understand why and where is appropriate to use the edge, what is the value for the edge from their market perspective to listen at their needs and understand what is pertinent to be standardized in the framework of ETSI MEC. Again, special computing and gaming is also call it as a metaverse but also cyber security aspects. So very vertical segments, very much important where we want to listen at them to understand what is pertinent to be standardized in ETSI MEC. So what is MEC, ETSI MEC? ETSI is a big umbrella of standardization where edge computing is being introduced by MEC, M-E-C and ISG, which is a group counting more than 200 members and participants, right?
(38:03):
It's a big group, very heterogeneous, well very well populated and growing in membership, not just because I'm a beautiful guy. Well not only that but because this is a sign that of course the edge computing market is growing and then also it is very important to operate. It's a complex value chain. The standard is critical when you need interoperability and global adoption. So that's why ETSI MEC was created like 10 years ago and starting from the definitions of the virtualization of the network, those of you who are experts on NFV knows what I'm talking about. But it's going beyond much more beyond because it's actually upper layer at application level we target the make application development communities developers, but the open standard approach is a must for us. We need to keep open to various kind of deployments, various verticals and not only just the mobile network.
(39:07):
Well let me clarify. 5G is a mainstream. I can tell you I'm a bit proud of my delegates. It was a more than three years work in alignment between ETSI, me three GPP SA six and CT three to align three GPP release 18 with MEC phase three, you maybe are not so technical you cannot appreciate the detail of that. But I can tell you that if you have a 5G network with release 18 equipment and network, you have also the possibility to interoperate with the MEC platform. There is a MEC profile for the cap for the common API framework. And this is actually very much impactful achievement that we have done is a joint effort between multiple standardization bodies. So ETSI MEC is seriously alignable with three GPP and 5G networks. But not just only that. Also wifi networks and fixed networks. That's why it's edge computing.
(40:05):
Cloud computing at the edges standardized and also we are focusing on, well this is the history of what we have done in the various phases of the standardization you can forget, but important thing what we are introducing is a set of APIs talking the commonly universally accepted language, which is restful HTTP comments. We see in all our browsers in the internet we are standardizing some APIs. You can see here some examples requested by the industry. But also it's important that also companies can add their own APIs. They don't have to knock at the door of ETSI MEC to make them available to the platform. There is an automatic mechanism to automatically recognize new services present from the application. And this is advertised to all authorized applications in the platform. So it's an open approach in our standard that permits to increase the portfolio of services.
(41:07):
So you don't have to standardize everything. We are very much open also to open source communities. I will tell you about CAMARA collaboration. We have the Linux Foundation in CAMARA project. So actually we are doing also a lot of work because we recognize that developers, they don't spend too much time reading specifications, but they need also to play with software and they need to practice. So a set of complimentary activities. We created a working group years ago dealing with MEC deployment and ecosystem engagement activities ranging from production of software and open AI representation in open source testing and conformance proof of concepts, trials with live networks from multiple operators. Also in a Mac Federation, Mac Sandbox, which is a sort of reference platform for software developers that can practice about the usage of the APIs and also collaboration with open source communities, hackathons, plug test for interoperability and also some sort of podcast tutorials to explain the difficult parts of the standards.
(42:18):
So this is just an example of a few APIs that we've published, but I would like also to give you some pointer if you are interested about the pragmatic approach that we have. Really we wanted to clarify this huge space of APIs. Everybody today is talking about APIs, especially now that we are talking about AI. Consumers of APIs will be also AI agents and they will be AI to AI communication via APIs. APIs are important today. They will become even more important in the future through MCP and A2A, any kind of protocol and gateway that serves to communicate in terms of that exchange between entities, consumer and producer of services. And then there is a lot of confusion in terms of APIs. Should I use a standardized API? Should I use an open source a p? I can produce my own proprietary API. If I'm a developer, I am really confused about that.
(43:13):
So maybe if you want to learn more, there is this kind of white paper written by delegates from ETSI MEC and CAMARA and the TM Forum delegates together to shed some light, possibly give some guidance humbly to developers to let's say deep dive into this kind of universe of APIs with different flavors, how they can fit in the bigger picture, how they can complement to each other and how they can be consumed with some commonalities and some possibility to expose the services in an interoperable way. So moving forward, we are working also seriously in the space of AI, which is actually many domains. In some domains, we are seeing a lot of potential for the future and for example, abstraction of APIs sensor sharing, but also for the security part, for the security monitoring management, the cm, the API gateway. And also there is a working item, especially focused on artificial intelligence at the MEC and also other, there was someone maybe Nico talking about let's say constrained devices. Also possibility to have in the future more and more workloads that are also in unpredictable connectivity or connected remotely at the far edge. So distributed networks and IoT deployments. This is a very important flagship project that we have. And yeah, with that I believe I conclude my presentation, we are looking forward to see more advancement in the standards and with more work on AI and edge native design support for application developers. Thank you so much. Thank you very much.
Guy Daniels, TelecomTV (45:02):
Your day sit down. Great, thanks so much Dar. Lot of information there. Thank you very much. Do we have a quick question for Dario? I neglected you earlier, Francis, so let's quickly get Francis and I think there's another question over here. Let's get to him. Yep. Thank you Francis. The gent over in the second row there Tony. Yeah, we'll find someone for you eventually.
Audience Member (45:23):
Thank you. Thank you very much for the presentation. Sorry for my voice. So basically my question is regarding ETSI effort in term of future A2A and MCP standardizations. If there are any effort on that, let me give a little bit of context for that. So basically if we see yesterday for the presentation we have this agent AI layer, which is very important and if we look to what we have seen that we have multiple AA, A2A implementation from Google, IBM and others and so on. So how to ensure interopability and open ecosystem within agentic AI layer and what are its effort in this regard beyond API.
Guy Daniels, TelecomTV (46:12):
Great. Great question.
Dario Sabella, ETSI MEC & xFlow Research (46:12):
Thank you. Thank you for the question. Of course, it's difficult for me to give you an answer because there are some decisions which are ongoing. But I can tell you that the group in Mecca and driving, I'm pushing them to evolve to what I call Friendly Mac 2.0 strategy to evolve the group. And I've also sensed the temperature from companies. I see also many inputs for new ideas for the topics, for the future standardization. And also I see a lot of interest on edge AI and these topics are really on the radar. I truly agree with you. We need for interoperability in this kind of new set of technologies and the standards will play a key role. I'm driving this group to migrate into a technical committee to announce the footprint and have more impact also for European regulation when there is a lot of issue of course for companies to be compliant with and a lot of unclear situations. So the standard may help on that and I believe we are considering this really, it's on the radar, I cannot tell you now, but maybe next conference I can give you some updates on the standards where we are going.
Guy Daniels, TelecomTV (47:18):
Fantastic. Thanks Dario. That's great. We're out of time, so any more questions, maybe we can lead them for our coffee break, which is coming up in about 30 minutes or so. So I'm going to start our next session now, but please a round of applause for our three guests now.
Please note that video transcripts are provided for reference only – content may vary from the published video or contain inaccuracies.
AI at the edge: New services
This session delves into the advancements and strategic approaches to AI at the edge, featuring presentations from Colt Technologies, Wind River and the chairman of ETSI MEC. The discussions cover Colt’s edge AI propositions, Wind River’s view on the intelligent edge, and ETSI’s standardisation efforts for edge computing. Together, they present the importance of edge computing in achieving low latency, high performance secure networks.
First Broadcast Live October 2025
Participants
Chetan Narang
Principal Innovation Strategy Manager, Colt Technologies
Dario Sabella
Chairman, ETSI MEC, VP, xFlow Research
Nico Marziliano
Regional VP, Telco Sales, Wind River
