T-Mobile for Business on The Intelligent Edge Enters Its Operational Era

Enterprise AI at the edge now lives or dies on who owns it when it breaks.

At The Six Five Summit: AI Unleashed 2026, Moe Beydoun, Senior Director of Product Specialist Sales, and Matthew Feider, Senior Manager of Physical and Edge AI at T-Mobile for Business, join David Nicholson to open the Intelligent Edge Track.

Their conversation reframes workload placement as more than an architectural choice. It is an economic and operational calculation balancing proximity to the action with provisioning, lifecycle management, network resiliency, and total cost of ownership.

Feider raises the stakes further. While generative AI is pursuing a trillion-dollar coding market, physical AI is tackling the hundred-trillion-dollar challenge of moving atoms in the real world—a problem he says is an order of magnitude harder and will require greater compute paired with carrier-grade network resiliency.

Key Insights:
🔹 Enterprise edge AI is moving from proof of concept to fleet-scale operations, making provisioning, authentication, patching, and rollback more critical than the model itself.

🔹 Workload placement is an optimization problem, not simply a latency decision, because fragmented device management can cost more than the underlying compute.

🔹 Physical AI requires more compute closer to the point of action, positioning network providers like T-Mobile to connect cloud-scale intelligence with real-time decisions at the edge.

🔹 Resiliency and security now matter as much as latency, with enterprises targeting five-nines availability that the cloud alone cannot consistently provide.

🔹 Data sovereignty is becoming a critical source of enterprise value, giving organizations that adopt federated, privacy-first architectures now an advantage as physical AI scales.

Both guests frame the next year as a design window: enterprises that work backward from the business outcome and build the network investment right the first time will be positioned to scale physical AI when everyone else is still rebuilding.

Watch the full video at sixfivemedia.com, and subscribe to our YouTube channel so you never miss an episode.

Explore more sessions from Six Five Summit: AI Unleashed 2026 at sixfivemedia.com/summit.

Disclaimer: Six Five Media is for information and entertainment purposes only. Over the course of this video, we may discuss companies that are publicly traded, and we may reference their equity share prices. Nothing discussed during this webcast should be considered investment advice or a recommendation to buy or sell any security. We are not investment advisors, and you should not rely on this content as financial advice. Six Five Media collaborates with technology companies and industry leaders to produce research-driven interviews and multimedia programming for enterprise technology audiences.

‍

Moe Beydoun:
The goal shouldn't be to maximize the amount of AI running on a device. The goal should be to minimize the cost and complexity of delivering this for a reliable business outcome at scale.

David Nicholson: 

Hi everyone and welcome to the 6.5 Summit AI Unleashed 2026. For our Intelligent Edge track opener, we're exploring how AI is moving beyond experimentation and into real-world operations, and what that means for enterprises building the next generation of connected intelligent systems. Joining me are Mo Beydoun, Senior Director of Product Specialist Sales, and Matthew Feider, Senior Manager of Physical and Edge AI at T-Mobile for Business. Mo, Matthew, welcome to The Six Five.

Moe Beydoun:

Happy to be here.

David Nicholson: 

It's great to have both of you here. Mo, let's start with you. Last year, we spent a lot of time talking about where the edge is. A year later, where are you seeing the biggest signs that enterprises have moved from talking about the intelligent edge to actually operating it?

Moe Beydoun:

I think the clearest sign is that the conversation has actually shifted from technical feasibility to operational accountability. Enterprises are no longer asking, hey, can this workload run at the edge? They're asking, who owns it? How is it monitored? How does it recover? And what business process changes when it actually produces that insight? I think we're seeing this shift in several areas. First, those deployments are moving into repeatable operational workflows. For example, in manufacturing, healthcare, retail, logistics, public infrastructure, AI is beginning to actually present itself and trigger actions that are just more than dashboards, informational dashboards. Um, a detected event is actually initiating something. It's initiating a maintenance ticket. It's initiating, uh, alerting a safety system, uh, potentially redirecting inventories, uh, creating upstream and downstream, uh, supply chain changes real time. Um, and potentially changing how a physical system behaves. Secondly, enterprises are planning for fleets rather than individual sites. A successful pilot at a single facility is very different than managing hundreds or thousands of locations with different devices, different connectivity conditions, potentially different software versions, and obviously varying security requirements. Third, the edge is increasingly being treated as part of that enterprise operating model. We're seeing network teams, cloud teams, security teams, ops teams, and actually business owners are actually coming together. They're being brought together much earlier in the design process. And that really right there is a transition into this operational era. That intelligent edge is no longer a destination or a piece of infrastructure in the traditional sense. It's becoming this distributed operating environment that has to deliver measurable outcomes every single day for that organization.

David Nicholson: 

Matthew, in that operating environment, we're looking at AI moving into the physical world. So we talk about edge, and a lot of people think about inference on some device at the edge. But what about real-time decision-making when you've got robots and camera systems and connected things at the edge? How does that change the way we should think about edge architecture?

Matthew Feider: 

So simple, right? It's increasingly more complex, right? So when you're doing inference and you're chatting with a bot on your phone or anything, it's going out to clouds, completely controlled environment. right, from down to how we cool and the gigawatt and what is actually flowing through that and what context. When you go out into the real world to drive inference and make it happen in real time, not only does the environment and the data that you're inputting into those systems change, right? I got to understand the light. I have to understand tactileness. I have to understand what else is going on around me. I'm also changing where my compute lies, and that can also change that real-time aspect. Most of the people traditionally, when we were looking at this about a year and a half ago, were saying, put it all on board, whether that's a robot or at my facility. Now, as they try to interact multiple things and the complexity of that intelligence grows, and we're seeing tokens in this space grow three times faster than the actual devices. they're needing something that starts to understand the bigger picture, right? What does that camera see? What is the robot actually experiencing? How are the routes that the cars need to go? And so if you want that to be a decision, you need a bigger brain, right? And that bigger brain though, needs to be closer to where the actual action is happening. And so for us, that's where T-Mobile fits in. As we look at physical AI, as we look at the actual edge, we want you not just to do inference to get an answer. This is an answer machine now. What we really care about is like, how do we actually drive action? And how do we drive decision through that? So the customers that we talk about are me. First, they care about latency. It's the first thing they come to us for. But then we start to expand on TCO. Right? How is it cheaper than you continuously updating your hardware locally? How do we make it better and faster in the cloud? They want bigger compute. Right? They're used to more of the AGX type of series. That gets them to a certain place. How can they expand past that? And then finally, and Mo talked about it at the end of his section, was resiliency. Right? Resiliency and security. We were at the very early innings of how do I get five nights? I don't think you're going to get that through the cloud today. And we're not seeing that. There's a lot more people getting involved in this. And as it enters the physical world, it actually can impact traffic, can impact throughput in a facility. There's nobody else who has this carrier grade like security, both from the network side, all the way up into the compute that we're starting to deploy into our edge.

David Nicholson: 

In that quest for five nines of availability and beyond. Mo, we had a chance to chat earlier about this idea of, you know, sometimes just because you can do something doesn't mean you should. And often operators will look at, you know, okay, when will we be able to run this level of inference at the edge on some device? Set aside inference for a second, just think of it in terms of compute and everything that's required to do something at a certain point. What's the reality when you think about decision-making and inference? Should we automatically think that as soon as you can run it as far out on the edge as you can, you should or no?

Moe Beydoun: 

The first reality is that a technical capability and operational suitability, they're not the same thing. A model can technically run on a camera, it can run on a sensor, it can run on a robot or a gateway. doesn't automatically make that endpoint the best place for it to operate, especially at scale. Enterprises are starting to encounter and quickly see that there's constraints around power and thermal performance, memory. We all know what's happening in the memory space, the cost of memory, observability from a security standpoint, and then most importantly, lifecycle management. You know, this challenge becomes much more visible, like I talked about earlier, when you're moving from 10 devices to 10,000 devices. You know, in those cases, every endpoint has to be provisioned. It's got to be authenticated, monitored, patched, updated. You know, these models have to be versioned and distributed. You know, performance can be measured, must be measured under those changing environmental conditions. You know, a failed update is going to need a rollback. You know, a security policy has to stay consistent across this highly distributed footprint. And as I mentioned a minute ago, you know, there's this economic question, right? An endpoint-based architecture, it can appear inexpensive during a pilot, and those initial compute requirements might be small. But as you scale this thing, the cost of managing that fragmented device environment, it can really exceed the cost of the compute itself. I think that's why workload placement should be treated as an optimization problem, not an ideological choice, if that makes sense. Some inference does belong on the endpoint. Some belongs potentially on the local edge where resources can be pooled. Some belong in the cloud. You know, the right answer really depends on the decision that's being made and the operational burden created by that placement. The goal shouldn't be to maximize the amount of AI running on a device. Really, the goal should be to minimize the cost and complexity of delivering this for a reliable business outcome at scale.

David Nicholson: 

Yeah, that's a great point. Coincidentally, I had my own encounter with this just looking at the variety of home security cameras that are available and trying to make the decision. Okay, how portable does this need to be? How much intelligence does it need to have? Okay, well, how much battery is it going to need? And I ended up wiring a bunch of things up. a bunch of dumb devices. So to your point, just because you can, it doesn't mean that you should. And on that reference to complexity, Matthew, my home camera system isn't the most complex environment, but largely we think complexity is being one of the biggest barriers to scaling the intelligent edge for enterprises. So what are the kinds of architectural decisions that people should pay really close attention to? as they move forward over the next couple of years. Nobody wants to paint themselves into a corner.

Matthew Feider: 

Den is in, man. Den is in. As far as, I wish, you know, Zumpik as well, but more importantly on the client side, right? Like we used to, before all this started to blow up and like, you look at memory and where that's going and start to understand what that means for the future devices. more and more on a separate part of our business when we're talking to, you know, OEM providers on the device side, they're looking at hybrid computing, right? And trying to understand what should I keep on device? What should I offload off device? And where should that workload actually land? Not every single endpoint and, you know, some of our largest partners are promoting GPUs on device and continue to stack that. Not each one of those can be a roaming data center. It's just not possible. And when you go into your home security example, I actually love that one because that single camera cannot make a decision. It sees only a certain view of what is out there, right? So start to understand What is work back from the action? What do you want to get done? What are you worried about? What is it? What's the KPI? And then start to design from there. Today, or not today, actually, a year and a half ago, when we started to engage with clients on this, everything was a POC, isolated inside a lab, and it was really easy to stand up, right? And the moment they started to grow it and understand the cost implications, understand how they needed to get the data to mesh together and get federated, it started to fall apart. And so physical AI is having this kind of scaling issue and it comes in a couple areas. And we do think the architecture compute and connectivity wise is what is broken and is what we are here to solve. So first, focus on the actions you want to drive, then work backwards. Understand what this looks like in a fleet or environment wide or not isolated to a plan. Next, you know, The way robots were building models a year ago versus how they are today with more of the MOEs and things on that area, open and modular, right? You've got to be able to swap these out because every single six months, maybe every six weeks, something's going to change. And the last thing is understand your data. I think what Palantir is saying about data sovereignty and owning it and being able to create value out of it, I think we're in the very early innings of what that's going to look like. But I think most of the value today is going to accrue to that layer. So whatever you can do, both to enable it for yourself, but start to think about how you enable it for others in a federated way, in a privacy-centric way. Those are decisions or strategic decisions we should start to think about together.

David Nicholson: 

Well, gentlemen, you have been kicking off the Intelligent Edge track. So the folks engaging with our content here are going to be industry leaders who are looking into the future and trying to make sense of this question, how do I get positive ROI out of AI? How do I keep my job while doing this? So Mo, I'm going to start with you, but Matthew, I'm going to ask you the same question. If you're talking to an enterprise leader, what are some of the things or what is the one thing that you believe they might be at risk of underestimating about operating AI in the physical world, and how should they be approaching this? How should they be thinking differently than they are now?

Moe Beydoun: 

That's a great question, and I just can't give you one answer. I think the response from my standpoint is really twofold. First of all, I think the operating model is underestimated. Many organizations are still approaching physical AI in a traditional sense. They're selecting a model, they're selecting, purchasing hardware, connecting those devices, and then launching a use case. Um, you know, in practice, what we see is, you know, the long, long-term success of that deployment really depends on how the people and how the processes and the technology actually operate together. Um, you know, when an AI system, uh, identifies an event, an issue, you know, who, who receives it, who sees it, what determines whether it's accurate or not, you know, when does the workflow begin? Who's accountable if that system's unavailable or the model's not behaving as intended? Over the next year, enterprise leaders should really be spending a lot of time designing how the human and operational workflows come together as they select where they're going to spend their capital on AI technology. The value is not created when the model produces inference. The value is created when that organization can consistently turn that inference into a trusted action for their business. And then the second piece, my second response is, I can't tell you how many transformation initiatives I've seen fail because the network. And the network is an afterthought or the wrong investment is made in network and an organization has to reinvest to get it right the next time around. The network really is that common enabler that's gonna bring these new technologies together into the workplace. It's the underpinning for everything that's happening from an AI, from an IoT, from a hyper automation standpoint, I mean, trillions, trillions of dollars are being spent and invested as we speak into this space. And the network and the connectivity, it really is that digital currency. It's that critical currency that's gonna drive success really for that intersection. If you envision that Venn diagram, that intersection between the business outcome that they're trying to achieve you know, enterprise IT and everything they need to do to manage and sustain and support and observe, as well as the actual enterprise mobility that can be an application, a use case, an AGV in motion that has to operate seamlessly within the four walls, outside the four walls, on the go. And that's why I think network, obviously selfishly, I think T-Mobile plays a critical role in delivering that outcome for our customers.

Matthew Feider: 

I think Mo covered it all. I think, look, I'm stealing this from NVIDIA, but their leader of physical AI commented how what ChatsVT and Anthropic and most people are racing towards on the coding side. It's like a trillion dollar industry, IT. Moving atoms, which is what physical AI is focused on, is a hundred trillion, maybe more, industry. It's the core of what we do as humans. And training generative AI to solve that $1 trillion industry of intelligence is actually much easier than addressing our physical world. We're extremely good at it. you will see human in the loop, human on the loop, trying to deal with these edge cases and unpredictable nature of it for quite a long time, right? But if you get it right, and you do all the steps Mo talked about, which I think was really spot on on how you back your way into it, start slow and grow and scale and understand that, you will deliver lots of value into your business, right? But it starts today. It can't be one of those things where you wait for somebody else to define it because they will define it around somebody else, right? And so, but we're here, like I wouldn't say as T-Mobile, we have the answer, we have it figured out. If anybody tells you they do, I'm gonna laugh them out of the room. Our main focus is how does converging the network with the right compute, the right architecture as well. If you want to do it on cloud because it needs to be there, do it there. You need to have it on device, you do it there, right? What we are in a journey of is figuring out what that right architecture, what that right business model and what that right commercialization is for you to unlock that. If we get that correct, that's the next connection for us, right? Like everybody has these, but I'm more interested in the robots than the cars. So we're open here to talk and learn more, but it's unpredictable. Take it one step at a time, get it down, but start today. Starting tomorrow, you're already bound.

David Nicholson:

Matthew Fider, Mo Beidun, thank you gentlemen from T-Mobile.

Moe Beydoun: 

Thank you for having us.

David Nicholson: 

That wraps our Intelligent Edge track opener at the Six Five Summit AI Unleashed 2026. Thanks for joining us. Don't forget to subscribe, follow us on social and visit sixfivemedia.com for more conversations and insights from across the summit. We'll see you next time.