Global Electronics Association CTO on AI Infrastructure and Supply Chain Resilience

[00:00:00] - [Speaker 0]
I'm only able to record over 60 interviews a month and cover events around the world. Thanks to partners like NordLayer, who not only support the show, but the entire Tech Talks network. But today, I wanna talk about a challenge I hear all the time from founders. You know security matters, but you don't have the budget or the luxury of a big team to run enterprise grade tools, and yet you're still a target. Now the NordLayer new browser feels like it was built with this exact scenario in mind because it gives you enterprise level control over how you access your SaaS apps, move data, and browse the web.

[00:00:42] - [Speaker 0]
But most importantly, without the cost or complexity that often comes with it. Best of all, it's easy to deploy, simply to manage, and will feel very familiar for your team from day one, which means you'll actually get adoption instead of resistance. And that is the part that often gets overlooked. But the best security tool is of course the one that people actually use. So if you're scaling a business and want to get ahead of risk without slowing growth, take a look at nordlayer.com/browser, and you can find out more information there.

[00:01:17] - [Speaker 0]
But now, on with today's show. What happens when the race to scale AI runs straight into the realities of energy, materials, and a supply chain that still has to deliver every single part on time? Well, this is something I'm gonna discuss with Matt Kelly, CTO and and vice president of standards and technology at the Global Electronics Association. And this is one of those conversations that gets past the hype and straight into hard practical questions that business and technology leaders are asking right now. Because while there is no shortage of excitement around AI infrastructure, there is a much tougher story unfolding underneath it around sourcing, resilience, power, calling, and the sheer physical demands of supporting what comes next.

[00:02:12] - [Speaker 0]
And Matt is gonna bring a perspective you don't get to hear every day because he works across thousands of member organizations spanning industries like automotive, manufacturing, and electronics. So he gets to see the bigger picture. And in this conversation, we'll also unpack why procurement is shifting from cost based thinking to confidence based sourcing and understand why dual sourcing is no longer something you just keep in reserve and why a fully local supply chain sounds great in theory but gets a lot more complicated in practice. And, yes, as this is a tech podcast, we'll also get into the reality that AI growth is now tied just as much to energy efficiency as it is to compute power. And from silicon and packaging to thermal management and end of life reuse, this conversation is a reminder that the future of AI is not just software talking to software.

[00:03:06] - [Speaker 0]
It's actually about infrastructure, materials, and whether the physical world can keep pace with some of the big promises being made right now. So if you're looking for a more grounded conversation about AI infrastructure and supply chain resilience and what leaders should actually be preparing for over the next twelve to eighteen months, then there's a lot to offer you in this episode. But enough from me. Let me introduce you to my guest right now. So a massive warm welcome to the show.

[00:03:36] - [Speaker 0]
Can you tell everyone listening a little about who you are and what you do?

[00:03:40] - [Speaker 1]
Hi, Neil. Thanks for having me. My name is Matt Kelly. I'm chief technology officer and vice president of Standards and Technology here at the Global Electronics Association. We are thrilled to be here with you today and looking forward to the discussion.

[00:03:56] - [Speaker 0]
Me too, and thank you for sitting down with me today. And when we talk about AI infrastructure this year, there's so many big talking points. Everyone's going talking about everything from agentic AI to to quantum and everything in between. But I'm curious from what you're seeing, what are the biggest pressures that you're seeing on supply chains, especially that most IT leaders maybe underestimate today? I'm sure you'd you'd have a lot of conversations out there, but what are they underestimating most?

[00:04:26] - [Speaker 1]
Well, I think that, you know and we're gonna go through this in in in our discussion here, Neil. But I think the the kind of the overall message that that we're definitely seeing here and that I would like to convey is that the electronics industry as a whole from, you know, OEMs to the final products that are used in the IT world, these electronics products and the industry is really shifting from a cost optimization to a system level performance and resilience model. And this is really all underpinned by AI, the capabilities that it that it gives us, the hardware and the packaging that drives these systems. So the computing systems that are enabling AI. And then the last part is really the intelligent manufacturing elements that are within our factories here today.

[00:05:19] - [Speaker 1]
The the idea of factory of the future and industry four dot o are are very much in motion and and are part of the factories today. So these altogether are really the forcing function that's translating, you know, our changes faster than ever. So I think that would be my opening remark.

[00:05:36] - [Speaker 0]
Love it. And I think there's also been somewhat of a shift from cost base to confidence based sourcing. But how are you seeing this mindset changing procurement decisions? And what does that look like in in practice for tech leaders and business leaders today?

[00:05:51] - [Speaker 1]
Well, this shift has really been building definitely, not just in the last couple of years, but beyond. I I think that some of our information that that we've been tracking takes us back to more around the 2017 time frame. So, you know, a good few years before COVID onset, we were very well aware of those news headlines, you know, constraining sourcing. But, you know, starting kind of around that 2017 time frame that the shift has really been building, And it's it's really been pushed over the edge from a combination of, as I mentioned, I think COVID, the geopolitical disruptions that we have, and now AI driven demand. These things altogether, it's usually not just one.

[00:06:34] - [Speaker 1]
These three things are are really compounding that need for confidence based sourcing. And companies really are now investing a lot of time and energy into that, that realizing that that lowest cost sourcing doesn't really matter if you can't get the either the incoming parts or the subsystems that you need or the product that you're buying. So I think that's really been kind of the the the driving force behind this. So, you know, a couple of real examples, think that, you know, the the the latest sourcing issues of of of memory shortages. We're also seeing significant shortages in printed circuit boards, if you can believe it.

[00:07:17] - [Speaker 1]
You know, these AI servers are are very large and and, you know, you can think about 19 inch or even 18 by 24 inch type of panels, and you multiply these out, guess what? You're you're talking a lot about a lot of material and and leading to shortages. So those are two examples of why I think confidence based sourcing is is really becoming, front and center.

[00:07:41] - [Speaker 0]
And another big change is dual sourcing used to be seen as optional, but it feels like it's now becoming almost expected. So how should organizations approach building redundancy with without creating unnecessary complexity or cost, for example?

[00:07:58] - [Speaker 1]
Dual sourcing is, I think, for a long time been viewed at as a kind of a as a necessity. Right? It was more of a of a risk mitigation technique, and as well just keeping costs and pricing under control so that, you know, you you had, kinda leverage during, negotiation phases of whatever you were procuring. The difference is is that the the need for dual sourcing has now become front and center. Leading companies are are really using this as a a need to drive supply chain resilience.

[00:08:37] - [Speaker 1]
It's really front and center of that messaging and that that strategy. So leading companies are now really measuring resilience with with with KPIs and, you know, making it much more real within not just in their strategy, but their execution of their operations. So dual sourcing is not really a I don't wanna say a nice to have. It's an actual fundamental to drive supply chain resilience in their companies.

[00:09:05] - [Speaker 0]
And there will be many people listening in organizations that are exploring regional supply chains now, but you've suggested that I was reading before you came on the podcast today that a fully local model isn't that realistic. So how should businesses strike that right balance between regional resilience and global efficiency? I would imagine it's it's quite a tricky, quite tricky to achieve.

[00:09:29] - [Speaker 1]
Yeah. Neil, I'm gonna answer the question in the context of the electronics manufacturing and electronics sector because, you know, various sectors are different. I'd I'd like to be a little specific on my answer. So, you know, when you look at and this goes goes back to my industrial days, you know, working out in industry for many decades. The reality is that it it requires both, you know, regional supply chain resilience and global supply chain enablement.

[00:09:58] - [Speaker 1]
They work together. I don't look at this as being two separate things. They are both needed to produce the product you're trying to deliver. On the regional side, the reality is that there are some things that you may wanna build domestically. The reality is they're never really going to come into the region.

[00:10:18] - [Speaker 1]
There's various reasons for that. Infrastructure is is is too costly. There's a a barrier to entry for that. A lot of it is is is workforce and labor related, particularly in lower cost regions where you just really can't replicate those types of things. So even though you may want to have regional sourcing for particular elements in your system, unfortunately, it's not gonna happen.

[00:10:44] - [Speaker 1]
What you're going to really need to be focused on on are what are the things that you can bring into the regional domestic supply, but then also be aware that you're likely gonna have to be paying a premium. Not all the cases, but in in many cases, particularly in North America or Europe, there would be fundamental things in your system that could be for security reasons. It could be for just, you know, local continuity that you that's a critical part supply that you need to have and are willing to pay for. Those are some good reasons to have domestic supply chain. On the global supply chain side of things, I very often when I speak to to to people, I say, make no mistake, the global supply chain is alive and well.

[00:11:31] - [Speaker 1]
And what we need to do is we need to harness that for the things that are commodities or have established footprints that make sense to keep where they are, work with trusted partners, work with trusted companies, work with, you know, allied partners, from a geopolitical sense. So to summarize, I don't look at this as, you know, you have to select a regional supply versus a global supply. In the electronics industry anyway, the number of elements and components in the system is extensive. And what you should be focusing in on is what am I selecting for my domestic supply and what needs to continue on as a global supply.

[00:12:10] - [Speaker 0]
And AI infrastructure is another big talking point right now, driving huge demand for both compute and energy. And I'm curious from what you're seeing out there. What are the real world constraints around power, cooling, and sustainability? Because they do conflict with each other sometimes. And and what what should organizations need to be planning for now?

[00:12:31] - [Speaker 1]
AI is obviously just in its early stages. I I think we are all thinking about what AI is gonna bring to us in in five years from now. The I've often said the AI timeline seems to not be, you know, linear. Three years from now, just seems like an eternity and imagine what will happen. So in in terms of what we should be preparing for for AI, I mean, I think I think there's the the user side of how you're using AI inside your companies, but also what the the reality of AI is on our AI infrastructure and the energy related side.

[00:13:07] - [Speaker 1]
So you you touched on on on the energy, so I'll talk a little bit about that. Energy efficiency of of HPC AI based data centers is really now the primary limiter of AI and what's called high performance compute scaling. These these advancements we're about to see really are kind of premised on how much power is available. So for example, over the next decade, we must think of energy efficiency as the most important challenge. Energy efficiency of compute has not been keeping pace with what we need and improvements in our hardware as well as, you know, some of the the silicon advances in the hardware really need to keep pace.

[00:13:56] - [Speaker 1]
So without these these elements, the the biggest single bottleneck will be power delivery to these systems. So the the promise of AI that we're all waiting to to kind of see and go begin this journey on really is premised on the power that that that fuels that in the first place. We we need to get this right, and there's a lot of challenges. We can talk a little bit more if you like on what those are.

[00:14:22] - [Speaker 0]
Yeah. Please do. Expand on that. What what are some of the other challenges that you're seeing there?

[00:14:26] - [Speaker 1]
So a few things. You know, in the in the sense of AI energy efficiency, we really need to be thinking again, this is kind of from a system level thinking perspective, that we need to be taking a what we call a silicon to systems design view. And that is the choices that are being made, and we see this every day in the news with advances at the silicon nodes with TSMC and and and the the IDMs, whether it be NVIDIA or AMD, Intel, others. All of this is really, you know, kinda starting at the silicon and and that cascades itself all the way down through these massive AI based data center systems. So power, not compute, is now the limiting factor for AI scaling.

[00:15:14] - [Speaker 1]
That's number one. And what we need to do is we need to think about co design at the very beginning so that we can manage electrical, thermal, mechanical, optical domains. All of these things are all working together. And if we don't get it right, then we will not be designing systems that have this power efficiency in front of us in the first place. So we're going to not be able to keep up with these power demands.

[00:15:40] - [Speaker 1]
And then I guess the last thing that I would say in this area is the amount of materials that it that this is taking. I mentioned earlier, I mean, these systems are very large. So we need to be thinking about ways to be able to reuse systems, to design them in a sustainable fashion, all of these things. Because energy efficiency, energy usage, yes, it's coming in the form of when that server is up and running, but we also know that these are large systems that need to be reused or recycled or kind of the back end of life of these things because we're at the very beginning of this journey and we just simply don't have enough materials to keep on the trajectory that we're on. So those are a few of my comments from the kind of the hardware side of AI energy efficiency.

[00:16:30] - [Speaker 0]
Big thank you to Denodo for supporting the Tech Talks Network and making these conversations possible. Because when your lakehouse stores the data, the real challenge is getting that data where it needs to go and faster. And your lakehouse stores the data, but Denodo helps deliver it faster. So with real time access, built in governance, and a business ready data marketplace, Denodo can help your teams unlock insights without costly duplication. And you can learn more by simply visiting denodo dot com.

[00:17:07] - [Speaker 0]
And looking at what you do here, you're in somewhat of a unique vantage point. And I'm curious, from your perspective, working across thousands of member organizations, where are you seeing companies getting supply chain resilience right, and where are they still falling short?

[00:17:23] - [Speaker 1]
Oh, it's a great question. Where they're getting it right is the a a lot of the the back end of the process. So, you know, things like mechanicals, the let's face it, all of all of the electronics that that are built, you're you're holding on to casings or housings or those types of things. You're not holding on to, you know, raw electronics. Right?

[00:17:47] - [Speaker 1]
So the sourcing of of those types of commodities can be sourced pretty much anywhere, as long as you're qualifying and making sure it's of high quality. So I think that supply chains are doing a a good job, or companies with of those supply chains are doing a good job of of sourcing and keeping those elements in the system running. I think the some of the biggest bottlenecks that we're seeing are on the kind of the front end of things where it would be, you know, particularly things like semiconductor or a chip fabrication where, there's a lot of very, unique materials, rare earth type of materials for examples, where there simply is a very limited supply and, that supply only exists in certain regions. So those would be two of my examples. I I think that this is still gonna be, you know, years in the making of of really trying to redefine where some of this is brought in, but, those are a couple of my examples.

[00:18:49] - [Speaker 0]
And I think when we're talking around expensive tech projects and resilience, there's a big ROI question mark, and how do we measure the success here? So as resilience becomes the almost a measurable KPI, how should leaders justify investment to the board, especially when the ROI is about risk reduction rather than immediate revenue? Because it's slightly different.

[00:19:12] - [Speaker 1]
Yeah. Certainly. Leading companies are now measuring resilience with real KPIs, as you mentioned, things like time to recover from a disruption or percentage of a dual source bill of materials, and, you know, things like, supplier geographic diversity. Those are some some real examples of KPIs that you just mentioned. So, you know, mature strategies really don't just mean you have a second supplier that you can go and act as needed, but in fact, you've actually gone through the full process.

[00:19:42] - [Speaker 1]
Right? You've qualified them. You verified and validated their performance, and you can switch that production on with minimal disruption as needed. And so it's it's really about that readiness, being able to switch as you need, not just, you know, having kind of a backup or a redundancy plan. The second part of the question you asked me was, you know, making this kind of like a case to the board and and why that matters.

[00:20:07] - [Speaker 1]
I think things have really changed. I think the the way to frame it is relatively simple, and it's this, you know, the cost of dual sourcing is predictable and the cost of disruption isn't. So you can imagine a line down situation that a particular component isn't available. I'll give you a a great example. We we talk about server enterprise server systems.

[00:20:32] - [Speaker 1]
You can think about a system being worth we'll call it like a million dollars just as a as a nominal figure, just making it up. There are hundreds of thousands, if not millions of capacitors and resistors in these systems.

[00:20:48] - [Speaker 0]
Yeah.

[00:20:49] - [Speaker 1]
If one of those capacitors or resistors is not available, you can't build that system. So you are holding up the ability to design or try to to build and deploy a a million dollar system and you can't get a half of a cent part in, it shuts your production line down. So it's a it's a really stark, but real example of of the importance for having dual sourcing, not just for your most expensive elements in a hard piece of hardware, but, you know, the idea of last part in is is very real, and you have to have all your parts in to build the product.

[00:21:24] - [Speaker 0]
Wow. Incredibly powerful point there. And if we look further ahead, are there any practical steps that the leaders listening should, do to ensure their organizations could take in the next, what, twelve to eighteen months and and maybe better ensure that their AI infrastructure and supply chains are are ready for what's coming next. And I appreciate the speed of technological change at the moment is off the charts. You look at how much has changed in the last two, three years alone.

[00:21:51] - [Speaker 0]
But any practical steps that you'd advise leaders listening there?

[00:21:55] - [Speaker 1]
So first, you know, plan for this massive energy surge. It it's already upon us. Plan for energy and cooling at every step, that you are thinking of through through your deployment of of of IT data centers. Secondly, diversify your supply chains. This includes, you know, you know, materials, components, all of the kind of subcomponents that go into those systems.

[00:22:24] - [Speaker 1]
And third, you know, if if you're in the position to design these systems, we should be thinking about designing for efficiency, better thermal management, better optimized architectures, more efficient packaging. Again, I'm talking very specific about the hardware. And so I think that what gives me optimism is that the industry, the electronics industry is very innovative, and it's very good at solving these challenges when aligned. So I think that we we really just need to be taking a system level approach again as opposed to just kind of I do my part and someone else does there. So those would be some of my practical example and and ideas.

[00:23:07] - [Speaker 0]
Well, it's been an absolute pleasure talking with you today. So many big takeaways. I'd love for people listening to feedback on their experiences and insights and anything that they're gonna be taking away from our conversation. But for anyone that wants to carry on what we started talking about with yourself, with your team, learn more about all the work that you're doing at Global Electronics. Where would you like me to point everyone?

[00:23:31] - [Speaker 1]
Yes. If there's questions, they can reach out to to to yourself or or to myself here, as contacted on the the podcast, and certainly reach out. I encourage everyone to take a look at our website. It's electronics.org here at the Global Electronics Association. So thanks for having me today, Neil.

[00:23:50] - [Speaker 1]
Appreciate it.

[00:23:51] - [Speaker 0]
Thank you. And we did cover a lot there from the shift from cost base to confidence based sourcing, regional plus global supply chains, AI infrastructure realities, and also some supply chain resilience as a KPI. Lots of great points around that. I'll have links to absolutely everything. I encourage anyone listening that would like to carry on the conversation to reach out to myself or you and your team.

[00:24:15] - [Speaker 0]
And, more than anything, just thank you for starting that conversation, Steve. Really appreciate your time.

[00:24:20] - [Speaker 1]
Thank you, Neil.

[00:24:21] - [Speaker 0]
So many big standout moments in that conversation. One in particular is that you can have a million dollar system ready to go, but if a tiny low cost component is missing, the whole thing stops. And that says everything about the moment we are in right now. For all the talk about transformation, intelligent systems, the next chapter of AI, all the basics still matter. Availability matters, readiness matters, and resilience matters.

[00:24:50] - [Speaker 0]
And I think Matt made an important point today about the way leaders need to frame these conversations internally because the cost of building in resilience may feel uncomfortable when budgets are tight, but the cost of disruption is usually far worse and far less predictable. And that's why this is no longer just a procurement discussion. It's a boardroom discussion. It's a operations discussion. And the other big takeaway for me is that there is no neat either or answer here.

[00:25:21] - [Speaker 0]
It's not local versus global. It's not cost versus resilience. It's not compute versus sustainability. We do need to move away from binary thinking because the organizations that do well, they'll be the ones that can balance those tensions without pretending that they do not exist. And I think it for many reasons.

[00:25:41] - [Speaker 0]
I think this conversation felt so timely because AI might be moving at remarkable speed, but the infrastructure behind it still depends on smart planning, real world trade offs, and disciplined execution. But what are you seeing out there? Are organizations treating supply chain resilience and AI infrastructure as long term strategic priorities, or are many still just reacting to problems only after the pressure shows up? Please, tech talks network dot com, drop by, send me a message, check out the links associated to this episode, and please share everything that you've taken away and things you'd like to add. But that's it.

[00:26:21] - [Speaker 0]
We're out of time for today. I'll speak to you all again very soon. Until next time. Bye for now.