In The Arena by TechArena - Inside Arm’s Drive to Advance AI’s Data Center Transformation

Episode Date: October 24, 2025

As AI fuels a $7 trillion-dollar infrastructure boom, Arm’s Mohamed Awad reveals how efficiency, custom silicon, and ecosystem-first design are reshaping hyperscalers and powering the gigawatt era.

Welcome to Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Alison Klein. Now, let's step into the arena. Welcome in the arena. My name's Alison Klein, and we're coming up on the AI Infra Summit in San Jose. I can't wait for this conference. It's going to be fantastic. And to set the stage today, I've got a really special guest. Mohamed Awad of Arm with us. He is a keynoteer at the event and a fantastic leader in the silicon industry that is driving incredible innovation for AI. Muhammad, welcome to the program. Thank you. That's quite the introduction. Thank you. Well, I think you're deserving of it. And just to set the stage, this is the first time on

Tech Arena for you. So you're the senior vice president and general manager of infrastructure business at Arm. And you're responsible for such a broad portfolio across cloud, data center, networking, AI, and HPC. Why don't we just start with a little bit about your background and how you entered this role and how you view your purview at Arm? Yeah, it is really interesting to be in the semiconductor industry real quick. My mom spent 30 years in the semiconductor industry working at TLA. And I used to always say, I never want to be there. And here I am and enjoying it quite a bit. I went to school for a software engineering when I First came out, I worked at companies like Nortel and Lucent, kind of in the infrastructure space.

Prior to Arm, I spent a decade at Broadcom in various different roles, obviously well known for their semiconductor capacity and been at Arm now for seven years. So it's been quite a ride and having a lot of fun within Infra. Yeah, during that seven-year tenure that you've had an arm, there's also been a massive disruption in the data center market. I've spent the last couple decades in data centers. And one of the things that I think about is that we've moved from the rackum stackum commodity server market to ones where racks of technology are being designed uniquely for AI workloads. When you think about what major cloud providers and neocloud vendors are delivering in terms of core capabilities, what are the patterns that you're seeing and how is

the silicon industry central to that effort? Yeah, I think you hit the nail on the head. When the cloud industry was born, it was born of, hey, we're going to take a bunch of commodity hardware and cobble it together and provide massive amounts of compute based on. Everyone's heard the story about how Google started in a Stanford dorm room and off-the-shelf silicon and hardware. I think what's changed over the last, certainly several years, but probably 10 years or more, is that the world has adopted the cloud much more broadly. The amount of data flowing into the infrastructure, the amount of data being pushed out of the infrastructure has really meant that cloud service providers have had to find a way to scale to keep up. And scaling to keep up

is both physical scaling, like how fast can I build another data center, all the way through to how do I find enough power to be able to support this or enough compute? And I think that's really been the big transition. And of course, all that's been incredibly exacerbating in the last couple of years through AI. That transition has made them really look at how to get as efficient as possible, how to get as much performance as possible out of every watt that they have available to them. And I think that's where building your silicon around your, you data center, i.e. doing your own grounds-up silicon design in support of a system architecture or a data center architecture that you're trying to achieve has really become the de facto standard

as opposed to the other way around, which was, hey, I'm just going to build my data center around whatever silicon is available. I think that's really been a big change in the industry, and we're seeing that happen even more aggressively as we move forward. There was commodity servers and then there was specialized service, things like DGXboxes. Today we talk about AI factories, these full-on racks that have been designed, that the networking and the compute and the accelerations been designed very closely together to kind of enable optimal performance and efficiency. Now, even if we go back a few years, think about the moment chat GPT came out,

and I think everybody realized, hey, a new moment of technology is beginning. This is something completely different than what we've seen before. but I don't think anybody really understood at that point how big this was going to get. And I guess one question that I have for you, as you look at the broad proliferation of AI factory buildout, is how big of an opportunity is there in front of us over the next few years? And how is Arm positioning itself to guide the industry towards making smart investments during that time? Yeah, the sort of scale of AI, it's really kind of wild when you think about it. I've read statistics about how there's going to be six and a half or seven trillion dollars worth of investment in infrastructure by 2030 just in support of AI.

You know, you look at how much compute was required or how much data was required in order to train GPT4. It was petabyte worth of data, right? So this sort of amount of data is just staggering when you sit back and think about it. For our part, what we focus on is the sort of efficiency associated with enabling folks to build systems which are optimized for their problem set. And so that really means a couple of things. First, it means driving an incredible level of performance. So that's kind of what we focus on is that per walk, credible performance and an incredibly efficient envelope. And of course, Arms Heritage has been efficiency.

That's what we're very well known for. you know, we've got a long history of 35 years of being the platform around efficiency, but it's not just about the efficiencies of the CPU, it's the efficiency of the compute subsystem, it's the efficiency of all the other IP that's required to pull that together, things like our mesh networking, our mesh interconnect between the CPUs, etc. But beyond all that, it's the efficiency associated with allowing our partners, in many cases, all of the large hyperscalers to go off and build silicon, which is specifically optimized to their use case, which is designed around the acceleration that they're trying to achieve, the workload that they're

trying to create. That's kind of what's very uniquely armed, is that we give you that flexibility to go off and innovate. We enable you with that kind of technology leadership, and then we provide that ecosystem. Of course, we've got that incredible ecosystem standing behind us that you can take advantage of. And I think it's the combination of those things, which is allowing our partners to go innovate in these incredible ways to kind of meet that challenge. Now, I think that one of the stories, we talk a lot about different trends in AI and what's happening in terms of the change of composition and infrastructure. But one of the stories that I don't think has gotten as much attention as it deserves to is the remarkable growth

we've seen in segment share for ARM in the hyperscaler market. You've forecasted. that half of the compute shipped to top hyperscalers in 2025 will be arm-based. What do you think has prepared the company to take advantage of this moment of the AI era and drive new leadership? Yeah, it's an incredible exciting time. I mean, you know, AWS talks about how 50% of the compute they've shipped in the last two years has already been armed. So now we're talking about other major cloud service providers joining those ranks.

it's an incredibly exciting time for us from that perspective. This is really driven by a couple of things. One, as they achieve better and better TCO from these solutions, which are optimized for their infrastructure, better per per watt across the board, that kind of high performance, low power consumption solutions, it becomes clear to them that's the right path to go down and that's the right avenue to deploy.

And so we're kind of riding that tailwind on one hand for just broadly general purpose compute, but now with the inflection of AI and these full-on racks and systems, which are being built specifically for AI, whether from invidia or from many of the large hyperskills, they're building CPU, GPU acceleration altogether specifically in service of this massive influx of AI systems. And so that's what's driving a lot of this. It's really about them seeing the sort of benefits that Arm brings a table and taking advantage of it. For our part, and for the ecosystems part, ultimately what this means, and I think this is the really interesting part of this, is that you've got this massive amount of AI software and infrastructure,

which is being optimized for Arm and being optimized for Arm first, whether you're on a Grace platform or you're one of those homegrown platforms at one of the hyperscalers. It's all being optimized for Arm first. And I think that's an incredible place to be. We went from in general purpose compute, where at one point we were trying to get the software ecosystem to the point where it was as good as. And now, whether it's general purpose compute or AI, in many cases, software's being optimized for ARM first and not second. That's the big transition that we're seeing happen right now. No, another trend is the fact that we're entering the gigawatt era in terms of power consumption as hyperscale or data centers scale.

And one thing that I know about ARM is, and you mentioned it earlier, that you guys are known for compute efficiency. I think that energy efficiency in delivery of platforms is becoming non-negotiable in AI, given the pure demand for compute. How is ARM helping address this challenge within the way that you're evolving your technology and working with others to deliver solutions to the marketplace? Yeah, I mean, we obviously have a long heritage, as we talked about earlier, around designing, both CPUs and peripheral IP, which is designed to be incredibly low power. So that's the first thing. And just to put that in perspective, when you're talking about a 500 watt CPU, if you can pull 20, 30 percent of the power out of that CPU, it may not seem like a lot for a single CPU. But when you start multiplying that across an entire data center, that means a lot more AI you can

fit into those platforms. The reality of it is that the world is struggling. keep up with power today for AI. And so getting more efficient on something like your compute, your general purpose compute, is not going to mean that you need less data centers. What it's going to mean is that you can fit more compute into that existing shell, which is going to be there for 15 years or into that existing power budget that you've already got defined for that system. And so that's where we come in, which is just helping them pull that overall power consumption down on some of those components, which ultimately drives a much more efficient system design. Now, we've been talking a lot about hyperscalers, but of course, enterprises today are

running workloads from across multi-cloud into on-prem and to the edge. And moving workloads across the seamlessly has historically been a challenge. How is arm addressing that need for workload portability across such diverse environments? Yeah, it's a great question. And I'll tell you, it's really interesting because one of the unique things about ARM is that because our ecosystem, you know, if you look at AWS, you look at Google, you look at Microsoft, you look at those kind of cloud service providers, they're all building spoke custom silicon for their infrastructure, but they're doing so based on our CPU implementations. And because they're based on our CPU implementations, the software that's optimized for one can be leveraged across all of them.

And I think that's an important point that maybe sometimes people miss. And that allows for those enterprises to go take advantage of all the benefits, TCO and power benefits that oftentimes is hyperscale as pass along. And they all sort of boast about 40, 50, 60% per or Perf per TCO benefits associated with their arm-based solutions. And that's obviously a great place for the broader ecosystem to go take advantage of Arm. No, one thing that I know from being in the Silicon Arena a long time is that when you can actually customize Silicon to address specific workload requirements, you're going to end up with either a more performant or more efficient solution.

I think that this moment is really a renaissance of custom silicon delivery. How do you see Arm playing in that space? And how do you in your ecosystem help enable that shift to more custom designs? So there's lots of things. I mean, obviously, our focus at Arm has always been about how do we enable our ecosystem to go off and build the most optimized, most efficient, most performance systems possible. That's what the company was born on back 35 years ago with starting with the Apple Newton, right? So that's literally always been our genesis.

I think for our part, today what we're doing is moving even further down that path by enabling things like our own total design ecosystem. So this is about, hey, we'll take a compute subsystem where we'll stitch together a lot of the CPUs and interconnects, we'll provide that to the ecosystem so that they've got a starting point that don't have to start with just a bag of IP that helps accelerate their time to market, that helps lower the overall investment. It's going to require to achieve that. solution, allowing more and more companies to participate in the design of products based on

them. The other thing that we've done with our own total design ecosystem is we now have an ecosystem of partners who are now engaged with our compute subsystems. And so these are IP vendors who are taking their IP and integrating it into our compute subsystem. So if you need a memory controller or a PCIE interface, you can go to one of our partners, whether that's a cadence or synopsis or others. And they've already got it preintegrated. and ready to go on our compute subsystem, so further lowering your cost. We've worked with our EDA tool vendors so that they've integrated compute subsystem directly into the EDA tools to help streamline the development process.

We're working with all of the major foundries, whether that's Samsung or Intel or TSM, where they have used our compute subsystem to help go do performance benchmarking and optimize their process nodes around this so that you're confident in what sort of performance you can achieve. These are all the things that we do through our arm total design. ecosystem and our ecosystem more broadly to go accelerate and streamline and shorten that cycle of what it takes to go build silicon, shorten the sort of overall investment. Now, I know that we started the conversation here, but I do need to ask you, your keynoteing

at the AIM for Summit this week. What can attendees expect to hear about from you and your keynote? And are there specific messages that you want to send the attendees on the imperatives for what the industry and customers need to do together moving forward. Obviously, I'm very much looking forward to the keynote. And I guess what I would suggest that the industry needs to sort of remember is that we are at the beginning of this renaissance, as you might say. And then a lot of it is being driven by just the sheer demand and the transformational nature of AI.

And if we think about aspirationally what we would like AI to achieve and the potential for AI. I think it's clear that no one organization, one company, one technology is going to be able to solve that all by itself. And so for our part, and certainly within OCP, I think it's incredibly important that we collaborate together and look for ways to kind of advance for the common good so that we can all benefit from the sort of potential that AI has brought to the table. I think you're going to hear a lot about that at OCP, and that's certainly going to be part of what I want to talk about. That's awesome. Well, Mohamed, I think your mother would be proud of you for actually choosing the silicon path because it sounds like your work is having a wide impact on where the

industry is headed at this very important technology inflection point. I'm sure that folks who are listening online want to engage with you. Where can they connect with you to continue the story and find out more about what technology arm is delivering for these important workloads? First of all, let's say that my mom's going to be proud of me because she's my mom. She has to be proud. It's part of what she signed up for. But, of course, I will be at AI Infra Summit. So I'd love for folks to come say hi there.

But if I miss you there, I'm obviously on LinkedIn and all the typical platform. So please feel free to reach out. I'd love to. I cannot. Fantastic. I can't wait to hear your talk. I will be there. And I'm sure that a lot of folks will be jumping at the bit to hear what you've got to say.

Thanks so much for the time today, Mom. and it was a real pleasure. Thank you. Thanks for joining Tech Arena. Subscribe and engage at our website, techorina.aI. All content is copyright by Techarena.