Transcript

Nick Earle: [00:00:00] Hi, and welcome to this episode of IoT and AI Leaders. Nick Earle, and, I’m the exec chairman at Eseye. Uh, this is a great episode. Of all the episodes that we’ve done over the five to six years, I think this is probably the one where we’ve really got down to what it actually needs to happen to bring IoT to mass deployment, and in particular, how AI is going to drive that as opposed to IoT feeding AI.

And specifically, how the model for networking needs to change. My guest is Nir Shalom, , the CEO of, , FLOLIVE. And he- you can hear in what’s about to follow a very passionate overview of how the model needs to change to be customer-centric out to the operators as opposed to operators into their own customer base.

Lots of great examples, lots of, uh, discussions on policy and [00:01:00] AI agents and the evolution of the network. And I think that if you are in, uh, IoT or you’re just in IT or you are looking at AI in the context of IoT, this is, this is, uh, essential listening. So with that, uh, let me hand you over to my discussion with Nir Shalom, the CEO of FLOLIVE.

Enjoy.

Nir, welcome, uh, to the IoT and AI Leaders podcast. Great to have you.

Nir Shalom: Yeah. Thank you very much, Nick. It’s great to see you, and great to be here.

Nick Earle: Yes. And we’ve known each other for many years, um, so, uh, we’ve had multiple conversations about the future of the industry. But now we’re talking about something a little bit different, which is how IoT is going to, um, expand and embrace AI. In fact, you could argue it’s the other way around, how AI, uh, needs to expand and embrace IoT.

But we’ll get back to that. But in the meantime, for our listeners, um, [00:02:00] not everybody knows exactly what FLOLIVE does, so how about a quick advertorial for everybody, uh, you are the CEO of, um, what FLOLIVE is all about.

Nir Shalom: So FLOLIVE is a global solution for IoT connectivity and networking, uh, which basically means, uh, a global cellular network, uh, fully operated by, by FLOLIVE, which includes a distributed point of presence with local core networks in different parts of the world, integrated with different mobile operators, and they’re all orchestrated under the same, uh, OSS/BSS solutions, aiming to provide our partners, a single platform, uh, to manage, uh, IoT connectivity and networking for their customers anywhere in the world, where we make sure that we are compliant, secure, uh, optimizing, uh, latency issues and, uh, and other performance, uh, [00:03:00] requirements of the customers on a global scale.

Nick Earle: Yeah, absolutely. That’s a very concise, uh, explanation of something that’s fairly complicated, so well, well done. You must … And I’m sure you’re very use- used to, uh, doing that. So let’s dive in. Um, I always say at the beginning of these podcasts that, um, uh, this used to be called IoT, um, Leaders. Uh, and, and now, uh, um, uh, we’re calling it, uh, um, IoT and AI Leaders, and that was based on this premise that I referred to in the intro that the, these two areas are going to come together pretty quick.

In fact, they have to come together. Um, uh, and it’s all about the evolution of AI, um, and, um, as opposed to an, an evolution of IoT, because there’s a ton of data at the physical layer. What, what’s your, what’s your view on that?

Nir Shalom: Yeah. It’s a very good question, and I, and I would like to start with the AI aspect of it.

Nick Earle: Mm-hmm.

Nir Shalom: Then we’ll link it to the IoT, uh, piece. Today, [00:04:00] as, as we see it, when we talk about AI, and we think about companies that are, are implementing AI to get better business results, the vast majority of it is basically running in what we call the logical world. Meaning take an insurance company, they take a lot of data, uh, they analyze it, they run the models and everything, and out of it, they will maybe come back with a better s- you know, product for their customers and stuff like that.

So this is really a lot of data being processed, uh, to deliver some good business value. But when we talk, when we talk about physical AI, we talk about something slightly different. We talk about the ability of the models to control the physical component in the field. Think about a humanoid robot, uh, or, uh, traffic video, uh, cameras in the street that basically, uh, helps traffic to, to go more smoothly, uh, or auto- autonomous cars.

In all these use cases, AI is [00:05:00] being used not just as a, as a way to analyze data, but to help physical devices, I’m sorry, to make decisions about the device behavior. And when that happens, and by the way, this, this big evolution is happening as we speak. Yeah. We don’t yet see in the field so many autonomous cars or humanoid robots or, uh, you know, self-managed, uh, traffic videos and stuff like that.

It’s still, it’s still to come. And I believe that when it comes at scale It will, it will basically change the requirements from the network, from the underlying network, and it will introduce, uh, new sets of, uh, capab- or requirements for the network that we, we don’t fully realize today. And, you know, we always have this tendency to think, “You know, it’s work- the network is working.

What’s the issue? [00:06:00] It’s been, it has been working for many years, it’s going to keep working.” And we believe that this is not necessarily the case. We are… We may be just at the right point to say the networks were okay up until now, we know that in IoT it was very complex anyhow, but the physical AI, meaning devices that are going to use AI to manage the way they behave in the, in the field, w- may require a different set of capabilities from the network, and we need to analyze it, to understand it, and to start get preparing for it.

Nick Earle: Yeah. Let, let’s go a bit deeper on that. Um, and, um, you know, you and I both know as both being CEOs of IoT companies, I’ve become chairman now, stepped down as CEO, but, but, um, you know, for the last few years, we’ve been… When we looked at the network, it was about connectivity. I mean, it was… And the conversations with customers are connectivity.

Can I connect to the device? Can I see the device is active? It’s almost like a, a, a network [00:07:00]map of what is connected. But there really wasn’t, um, uh, too much, uh, back and forth between the network and the device once we got beyond that, that level. What you’re talking about now with, you know, uh, as you say, these things are coming.

I mean, you know, the Android robots are coming, the intelligent drones are coming, the a- uh, uh, autonomous cars are already in some cities and, um, uh, coming to it. I mean, just whichever area of life, it’s coming. Um, so there’s… It’s not like this is a debate as to whether this is needed. It, it’s definitely needed.

So let’s go, let’s go, you know, uh, uh, deeper in that. Um, uh, why is it important to get more, much richer information from devices, um, uh, up into the network layer? We’ll talk about how, but why is it important to get richer information than [00:08:00] connectivity for those who perhaps, um, aren’t, aren’t as familiar with this as, as you are?

Nir Shalom: You know, j- speaking very broadly, when you think about physical AI, uh, customers that are going to, or enterprises that are going to implement physical AI, we need to make a, to make a very simple decision to start with, and this is where the inferencing, where the AI is going to run, right? Is it going to be at the edge, meaning on the device itself?

Near edge, for example, like a near cloud location which is not far from a group of devices, or maybe- Uh, a little bit more, uh, up in the, in the hyperscalers or, or place like that. A-and there are different solutions and different combinations and everything. But if you analyze every– any of these use cases, you will get to the conclusion that each of these use cases will now require something a little bit different from the network.

Let’s take an example. If, if the AI is running at the edge, meaning [00:09:00] the edge device has all the capabilities to run the models It will need to have the interaction with the network to, for example, to do the model updates, uh, to get the trained data, et cetera. So this one now requires a different type maybe of, uh, of data download than, than we used to had, uh, to have in the past.

Just, just as one example. If the, if the model is sitting in the cloud, then the device and, and the model’s insights are r- relevant for the device operation, then now we need to think on, on how we are having a communication which is very resilient because the device cannot basically, uh, behave properly without being connected to the model sitting far away.

So the resiliency of the network, the availability of the network, the latency between the model and the device, uh, are becoming more critical. If now we’re talking about, uh, for example, the, uh, [00:10:00] the data that is being used for the models, we go back again to all the data compliance issues in a, in a, in a bigger way than we used to see it up until now.

So if you analyze the different deployment options for AI to run physical devices, by definition you start to realize that now these different topologies may require different network behaviors. And by the way, it’s not the same network behavior for each of these different topologies. We need to have a flexible enough network to be able to address, uh, different topologies and, and customers may change topology over time because we are just at the beginning of this revolution.

Nick Earle: Yeah, I, I, I… In fact, I was gonna pick up on that a- as the word top- topology because what you’re describing, if you think of it as a picture, a visual picture, is like a, a network diagram. But, but actually as the applications and the intelligence move to the edge and the need for, as you [00:11:00] say, ultra-low latency, so as you think 5G, moves to the edge, then actually a, a lot of the intelligence, the processing, the…

And we’ll get back to policy, um, uh, later on, um, is, is moving to the edge. A- and the reason I state it like that in terms of a topology or as a picture is that if we drew a picture today, if you said to someone, “What does the network look like?” You know, you gave them a whiteboard and a marker pen, they’d probably start off with the operator.

They, they, they… At the center of the circle, they’d put the operator, and then they put, you know, circles around of capability, and right round the edge they’d put their customers, and round the edge of the their customers they’d put their devices. And that’s based on a model where everything is from the center out.

Because the center of gravity is one operator Uh, one operator coordinating, uh, orchestrating the connectivity. But now the edge is tra- is changing because these devices are bec- as you rightly say, they’re [00:12:00] actually computers, they’re aggregation devices. They have their own firmware, they have their own memory.

They, they can run software to allow them to make more decisions. Um, but at the same time, as a result of standards in, in, um, i- in, uh, the IoT industry, we have standards like SGP.32, which most of our users will be very familiar, listeners will be very familiar with, which actually, you know, breaks the proprietary lock with the operator and, and allows choice and flexibility.

So suddenly what you… The, the idea of, you know, the, the layers of the onion with the operator at the core and, and the, and their own customers locked in breaks completely. Now we have a constellation and the eSIM orchestration layer, which is actually making a, a, a decision. So s- it’s not just that the data and, and intelligence is moving to the edge and that creates opportunity, it’s that it, it, it’s a new [00:13:00] topology that, that is, that has a different center of gravity to the old top- topology.

And speaking as someone who was at, you know, Cisco for 13 years, uh, most of our customers were operators. You know, we, we, we, um, we sold most of the equipment to operators because they were the guys who operated the networks. Operators and very large enterprises, uh, to be fair. But this is a, this is a very different, um, uh, topology.

So, so, how do you envisage this, is going to happen, um, uh, uh, in terms of, you know, the flexibility, the allocation of resources? Um, uh, how do you think that, that, that these networks, uh, need to be changed? Um, uh, because SDN, software-defined networking, allowed the network resources to be dynamically allocated.

You could have different QOS depending if you were doing a financial share trade, um, uh, [00:14:00]transaction versus a report creation. You could, you could pass the QOS as a simple example, or even you could choose latency gates. But now the operator is a one provider in a constellation. So, so how, how as an industry can, can that capability be implemented in your view?

Nir Shalom: Yeah, it’s a, it’s a very good question, and, uh, we need to do something which is hard for us, and to basically acknowledge the fact that the, the network itself is not in the center Meaning the center is the customers-

Nick Earle: Yeah …

Nir Shalom: need to drive real business value, uh, in, in how he implements his AI vision And as we said before, there are different ways for customers to deploy AI, but the network needs to do something that was never done, and in a way [00:15:00] to be adaptive enough and to fit itself towards the customer’s AI topology and not vice versa, meaning up until today, as you, as you mentioned before, the operators were building networks and they told customers in a way, “That’s the deal.

This is the network. This is where we have the point of presence.”

Nick Earle: Yep. “

Nir Shalom: The capabilities we have, this is the different quality of service levels we have, and, you know, this is what it costs.” It’s, it’s, it’s pre-provision. It’s quite rigid in a way. Um, and then customers needed to, in a way, use it in the way it was built and provisioned with fairly little flexibility, uh, on the network.

And, and they also needed to compromise, uh, upon the, uh, the network limitations. For example, coverage limitations, uh, performance limitations, cost limitations, availability limitations, compliance limitations. You know, customers were basically [00:16:00] getting the MNO’s network as if it is the only thing exists.

But we cannot assume that this is going to be the case in the future. Why? Because when customers are moving the, the decision-making to, to an agent, now the agent, in a way, expects different things. The agent will, will basically need to come and say, “This is what I need to operate in the best way. I’m sitting in…

I’m a, I’m an agent that runs the humanoid robot. I need to get something from the cloud. I need to get it now. It needs to come in with a certain, uh, low latency. It needs to come in, uh, uh, holistically, uh, or I need to keep the data in the country because of A, B, C, D.” So the networks, in a way, will need to be adaptive, and instead of pre-provisioned rigid architecture, to allow dynamic routing, dynamic decision-making, [00:17:00] dynamic policy changes, dynamic optimization of performance based on each device need in the network.

So an agent can wake up in the morning and say, for example, “Good morning. I now need to download a big amount of data from the cloud because I need to update myself, and it’s, I need to do it within the next 25 minutes because I’m going to go off network in 25 minutes,” and the network will need to find a way to do it for this specific device on the spot in real time, and then cater for another device needs, which may be a little bit different.

For example, security needs, uh, addressing security needs. That’s the way to think of it. The way to think of it is if the customers are thinking about the agents delivering value to their business- And now the network is basically the enablement layer underneath. And if the agents runs the show, then the network should be [00:18:00] another area where the, the agent can come and ask for the network capabilities, and the network should be adaptive and responsive to address these needs and, uh, and let the agent, uh, perform the business needs of the customer.

Nick Earle: Yeah. And I, and I, I… This is such an important, uh, issue. It, it really is a transition of, of the architecture model. So I, I’m gonna give my own interpretation, uh, of it. we’re using a different example. I’m gonna… I, I’m a great believer that, that you can… If you look back at IT You know, you can see that there are patterns that repeat and, you know, uh, on some of these podcasts I’ve been doing over the last five, six years, I’ve told the story of, um, uh, is, you know, when we had, um, uh, ma- let’s take SAP.

It, it sounds like a, a, a crazy example, but I believe it’s relevant. SAP used to be, have R2, which ran on a mainframe, and R3 became client-server. And, and okay, that’s a [00:19:00] different architecture. It could run on mini computers and PCs, a three-layer client-server model. But what was actually happening is that the, the, the, the, the logic of the decisions were being driven from the client.

In this case, it was a person in front of a PC, but it was delivered from the client back th-through to the layers and, um, not from the, the, the top. So in my example of the onion ring, if you have the operators, and there’s over 800 of them around the world, typically three or four per country, um, protected by regulation, um, uh, not sharing towers and all, all of the proprietary stack element that’s built into that.

But what you’re now painting is a picture, if we had a diagram, a whiteboard in this, in this, in this podcast, you picture where the, as you say, the user is at the center, and the operators are network nodes, but around the outside. But there could be 20 network nodes. And so if you take something like 5G, which is absolutely needed, [00:20:00] I mean, in the future, there’ll be 6G, but absolutely needed for cars, uh, robots, um, uh, whatever, then you can’t get universal 5G, which behaves the same way from every operator because they’re implementing their 5G stacks separately.

So now that we have a federation, a, a, a federation layer, in other words, the ability to, to choose the operator, so we’ve, we’ve broken the proprietary lock, that’s the eSIM. And then we have eSIM orchestration, which is basically… And, and both companies do this and a few others, but not many. A few others do it.

E-Eseye, a rules-based engine that says, “This is the level of service I want. I don’t care where you get it from, but I get it, and I want one invoice, one support contract, one portal.” Suddenly, you’ve broken that model. So, so something big needs to happen to break the model. That happened with SAP, uh, e-essentially the, the internet and the [00:21:00] browser and everything, um, various things.

But I think a similar thing is happening here in the, the, with the introduction of the eSIM, the breaking of the proprietary lock, and then the SGP32 standard allowing the user to be in charge of, uh, the choice, uh, and, uh, and then, uh, companies which were, um, MVNOs, but MVNOs with an eSIM orchestration capability that says, well, it, it, this rules engine will then get you the level of service that you want is very similar to what happened in client server.

And we know that what happened in client server actually became the fundamental model until then after that was mobile, and it’s just the latest in, in, in one. So I think it’s, it’s definitely happening, and people definitely need to plan for it. I use the SAP example in order to ask you the next question.

My learning, uh, I used to be based in Germany, and I was, uh, the account executive for [00:22:00]Hewlett-Packard, uh, for, um, SAP, uh, also when I was over in Silicon Valley, and I had long conversations with people like Hasso Plattner and, and the execs at SAP. And, and my learning was that, um, uh, they provided consistency, uh, workflow.

Um, you– It didn’t matter where the data came from, and whoever looked at it in the organization would get the same view, so that’s really important. So it has to go through a layer of the onion. The devices have to go through a consistent layer, not s- not like a bicycle wheel with spokes to different operators who don’t talk to each other.

5G needs, you know, 5G orchestration. But one of the biggest challenges was policy, um, because at the end of the day, you can technically federate between, um, operators and say, as you mentioned, the AI agent, “Find me someone who can provide this service now at this latency, uh, with this, uh, level of connectivity, [00:23:00] maybe with this price in this country that keeps the data i-in country,” because that’s the data sovereignty requirement that my device needs right now.

But how do you get corporate policy, um, how do you, um, uh, uh, define policy so that from an enterprise perspective, who, you know, large enterprises will own eighty percent of the devices. It’s the c- typical Pareto. Do you believe that the, the intelligent network or the, uh, device-centric or whatever de– phrase, interested in your view of what, what we call it, does that have to have consistency, consistent policy capabilities in order to implement this?

Nir Shalom: I think the answer is absolutely yes. And I think that we are going to see, um- Uh, two trends coming in, in this regard. One, one of, one of them is what we discussed before, meaning to give the, uh, the agents more autonomy [00:24:00] in, uh, defining their specific needs from the network and make the net- network built out of many atomic capabilities, and the agents are able to come and ask for each of these capabilities and release or, uh, stop using it when it’s not needed anymore.

So it keeps the resources, uh, efficient on one hand, but at the other hand, on, on the other hand, provides the capabilities the agents needs in real time. So this is, this is something that will happen. Why? Because it happens already for agents with other resources which are not cellular connectivity, so why not for cellular connectivity as well?

But then we are going to see a second thing, where companies are going to come and say We would like to define a behavior or a policy, if you may, and we would like the network to be smart enough to make sure that the devices behave according to the policy we define. Let’s take a couple of [00:25:00]examples to explain it.

Let’s say that the customer defines, because of the customer’s specific use case, the customer defines that the most important thing is minimum latency.

Nick Earle: Yeah.

Nir Shalom: What do we need to do now? So the customer doesn’t, doesn’t tell which devices connect to which network edge, meaning packet gateway or something like that.

There are many, many packet gateways in the network around the world. The net- the, the customer doesn’t say, “Send it here and there.” It just says minimum latency. The network will then need to go and check the devices every ti- every now and then, let’s say every few hours maybe, maybe every few minutes, depends on the customer needs, and basically change the route from the device to the packet gateway based on the lower, lowest latency.

One example. The second example, customer may come and say, “I need to adhere with this, this compliance, this regulation.” The network will need to understand what the [00:26:00] regulation means for the devices, and again, go and check the devices on a periodical basis, and in a way make sure that, uh, uh, that they reconnect the devices to be always compliant.

The customer will not deal with it. That’s the second example. Or the customer may say, “The policies that we define the routing in real time,” why would that happen? Let’s say that a customer has 20 inferencing locations to run their AI in the cloud or the near cloud. And now we have, we have a network edge near each of these inferencing location.

Meaning what we do, we bring the network to fit the customer’s topology. We don’t tell the customer, “That’s the network, uh, architecture. Now go and deal with it.” We basically ask the customer, “What is your AI topology?” And we put the network layer as an adjacent component to it. So now what we have is 20 inference [00:27:00] locations with 20 corresponding network edges, UPS packet gateways.

That’s great. But now the customer needs to decide which devices go to which inference point and network edge, right? And we have 20 of these around the world. So again, it can be based on latency, it can be based on policy, but it can also be driven by the customer. The customer may say, “You send all the devices to this location-” Until I tell you otherwise.

Why? Because in this location, my inferencing cost is the lowest one. So I want to use this one first. And when I exhaust it, I want to move to the next one. So the policy that I tell you what to do, Mr. Network. And now the network start changing or moving devices between inferencing locations of the customer upon the customer’s real time decisions.

But it all goes to the same, to the same place. We are going to have enterprise policies deploying, deployed on the network. We’re going to have agents speaking with the network. Mm. And the network [00:28:00] is, is eventually going to be having three mechanisms to fit this new world. Mechanism number one, that a network is flexible enough to fit the customer’s AI topology.

Meaning, in my view, every location where the customer has an inferencing location, there is a corresponding network edge, one. Second, the network is built out of many atomic, many subcomponents and many subcapabilities, uh, which are available for use and for, uh, triggering in real time. And third, the ability of either the policies or the agents to trigger any of these capabilities upon it.

This is, this is the new world. Now, connecting to what you said about SGP32 and network, network orchestration. SGP32 is great because, as you mentioned before, it unlocks this, uh, um, limitation of having a single operator serving a fleet of devices. [00:29:00] But it’s still … If you look five, seven years forward, it still has a limitation, as you mentioned before.

The limitation is that once you switch to an operator profile using SGP32, basically you are, you are using the MNOs, these specific MNOs capabilities, advantages and limitations. And these may be a little bit different from another MNO-

Nick Earle: Yeah …

Nir Shalom: trying to do the same. Meaning if you’re trying to get a network that behaves exactly in the same way anywhere in the world- And, uh, expose the, the network capabilities in a very simple way for the policies or the agents to use, SGP 32 may not take you there.

What you need rather than that maybe is basically one global network that really controls all the network capabilities independently of the MNOs. [00:30:00] Uh, and so that basically use the antenna coverage of the MNOs, but the logic, the resource allocation, uh, the decision-making is sitting on a, on an independent core that can work with any MNO’s coverage points around the world, but provides a global consistent, uh, behavior for every device using SGP 32 or not.

It’s becoming less important because the SGP 32 is only, is only the wrapper. It’s not the-

Nick Earle: Yeah. It, it, it, it, it, um, you know, um, I think we’ve, um, as we get towards the end of the podcast, I think we’ve, we’ve, we’ve identified the elephant in the room, and it’s a very big elephant. Because people don’t talk very much about it.

I, I, I’m gonna go back to my, um, 5G example. The, the industry, the telecom industry globally, I believe, is spending up to a trillion dollars rolling out 5G globally. I mean, the numbers, they’re, they’re just enormous [00:31:00] But that’s from the operator out. And as we’ve talked about, if you redraw the onion slide with the customer out and multiple operators around the outside, you cannot get a con- you cannot, uh, get a consistent 5G service.

You cannot. I don’t care if they have $5 trillion set aside, no one operator can ensure that all the other operators are consistent with their implementation of 5G. There, there, there’s no way you can coordinate 820 operators, or even if only the 20 biggest ones did it, there’s no way they can coordinate between each other on a consistent way of implementing 5G other than a, a consistent technical way, but not in the way that you’re talking about policy and, and choice and, and et cetera.

S- A-and they’re all competitors. It, it will just never, ever happen. The moment you see that, you then suddenly say, “That’s the elephant.” The elephant is that they can’t do it. So [00:32:00] but if you look at it from the customer’s point of view, you’re not gonna, back to the beginning of the podcast, you’re not gonna get the benefits of, of, you know, there’s 50 times more data in things than there so far has been uploaded into the large language models.

So getting granular, immediate data with, uh, on things where the device can make the choice, the device can react, and it can be by use case, it can, it can be by latency, it can be by policy. If, if that has to happen, but if that is going to get the benefits of AI, if that is going to happen, the elephant in the room is that the operators cannot implement it.

And I’ve, I’ve been, and you, I know you have, and I’ve been saying this to operators, and it doesn’t make you popular, but it’s like, “Well, how are you gonna implement it?” And typically it comes back to, “Well, I, I’m a, I’m a very big brand, and I have a lot of customers that are loyal to me.” But those customers are global And, and they want single product SKU, and they want this capability globally.

So, so [00:33:00] what then is going to happen is, as you said, you used a phrase which is like, I think you said something like the new network, or it’s almost like the, the, the network intelligence, the network control point is going to shift, and it’s going to be from the device out. It’s gonna be called by an agent that’s device resident and back to the onion.

One of the, uh, first layers of the onion is going to be this in- this intelligent network layer where you, where, which does the orchestration and the policy and the brokering and the, and the, um, uh, federation, and that is a big change. I think it’s not just a change for the operators and, and you know, we both got examples of operators who understand that.

We, we’re working very closely with, um, AT&T and others, and you’re working very closely with other operators. But it’s still only a subset of the total global operators. But it, it’s not just a big change, um, uh, [00:34:00] for the operators. It, it’s also a big change for the IoT landscape. And I was gonna, you know, ask you the question, uh, uh, I don’t know how many MVNOs are out there claiming that they can do SGP32, and there’s lots of them have, have, have been popping up lately.

But not every MVNO, but very few MVNOs have invested in the software stack like we have to create an orchestration capability. Can’t do exactly what you’re describing, but it is that layer. We’ve created, both created that layer of the onion. All our customers go through our layer of the onion. It’s too thin, but it’s a layer of the onion before they connect it.

We both federate to the operators. Um, uh, what’s your view on, on, on the IoT landscape of all the different companies that are doing connectivity? Because the way this conversation is going is that unless you have deep software stack capabilities to do orchestration, federation, [00:35:00] uh, device-centric, uh, um, capabilities, then it’s not like the MVNOs can, can…

The traditional MVNOs can do this because it’s not just saying we support SGP32. You can’t do the federation.

Nir Shalom: Yeah, and, and I agree. And I, and I, and I would say the following. I, I think it all starts with the customer’s needs, and we, we spoke about these needs along this conversation. But if you think of it, it goes eventually down To the fundamental requirements of customers that with a physical AI are becoming even more critical.

We’re talking about, first of all, very wide coverage. You know, customers expect that anywhere in the world where there is even a slight cellular reception, we should fetch it and use it to connect the devices. Customers are expecting a certain level of network availability and redundancy. They’re expecting certain, uh, level of, of, of latency, uh, downlink, [00:36:00] uplink, uh, capabilities, compliance capabilities, security capabilities.

So when you take all these and you say, “Okay, we know that they’re going– all of them are going to become 50 times more important because we are going to have 50 times more data,” and, and agents are going to be dependent on this connection to, to, uh, to operate their devices, otherwise the devices will not simply not operate

Nick Earle: Yeah

Nir Shalom: you now come and say, “So how do I make all these capabilities available all around the world?” Coverage, compatibility, security, and low latency and compliance, uh, and, and ease of use for customers and, and control on costs. How do I make all these available all around the world in a way behaving the same way so customers do not need to train agents to treat with different network behaviors in different parts of the world because it’s not scalable?

Nick Earle: Yeah.

Nir Shalom: And if you realize that this is the problem, you need to build something which is on one hand very global, [00:37:00] very broad in terms of use, uh, uh, coverage, but on the other hand is very deep in terms of the capabilities it can provide. And not just the orchestration. It’s really the network behavior itself.

It’s the quality of service and latencies and routing of the, of the device to network connection. If you understand that, then you realize that the operators will not be able to do it. They will not be able to do it because they may be able to do it with their own coverage in their own country, but they will not be able to do it because they cannot control the other MNOs-

Nick Earle: Yeah

Nir Shalom: behaviors in other parts of the world, and the customers don’t care about it. They want a single global solution with the right coverage that they can control the network capabilities upon need. And therefore, you need three things. We, we spoke about it, but I just summarize. One, you need to be able to really control the network capabilities, not just to orchestrate, to control.

The difference between orchestration and, and control is [00:38:00] the following. Orchestration means that I can move you using SGP 32 from AT&T to Verizon. This is orchestration. But then when you are AT&T, use AT&T full-blown. When you are at Verizon, use Verizon full-blown. This is orchestration. Control means that I can, I can dictate what is the route of the data between the device and the packet gateway.

I can def- I can decide in real time if I’m instantiating a VPN or APN or release it. I can control the network, not just orchestrate the network. So I believe that first of all, customers will, will want to be able to have the control, not only the orchestration. This is why SGP 32 is good, but not good enough, to be completely honest.

This is number one. And number two, you need to be able to expose these capabilities so that policies on one hand and agents on the other hand will be able to use these, th- this, this level of control To [00:39:00] optimize each connection in the network to do what it needs to do, to run the robots, to run the cars, to run the drones in the right way for the specific device.

And operators, while they may be able to do it to some extent on their own network, will never be able to do it globally, and I, I’m not even sure that they want to release the control of

Nick Earle: the network No, they don’t. Yeah,

Nir Shalom: correct, they don’t … but because it’s against their, their, their- It’s

Nick Earle: against their financial model, yeah, yeah

Nir Shalom: against the DNA. And I think we, it has to change because AI at the edge will require control

Nick Earle: Mm.

Nir Shalom: And networks will need to give this control in a consistent way all around the world, and this is the change we are going to go from the network being in the middle, pre-provisioned, managed by the MNOs, works okay locally.

We need to move to democratic network, running globally, running in a consistent way globally with capabilities that agents or policies can use in real time upon need, not controlled by the MNOs, [00:40:00] but controlled by the customers running it. That’s the evolution re- revolution we are going to go through.

Nick Earle: Well, I can’t, um, beat that. I know that, uh, both companies are working very hard with our engineering investment on building that. Um, we have been doing for many years. Um, but I… Why don’t we leave it there? I think we’ve certainly given our, our listeners and viewers plenty to think about and absorb, um, a- and we haven’t held back.

Um, so Nir, I, I just wanted to, uh, thank you for, um, your, uh, your wisdom, your emotion, and your insight, um, into what needs to happen. We definitely are on the same page as companies. And, uh, thank you for, uh, being my, my guest, uh, on the podcast and for explaining your view of the future and the, the new network model, I’m gonna call it, the, the, the, with the customer at the center, um, uh, so clearly in this podcast.

So thanks very much.

Nir Shalom: Thank you very much, Nick, for [00:41:00] having me today. It was a very interesting conversation, at least for me, so thank you very

Nick Earle: much. Yeah, I th- I’m sure it was for everyone, uh, me, but in particular the users. Thanks.

Nir Shalom: Thank you very much. Bye-bye.