Building the backbone of AI with Mahdi Yahya, Co-Founder @ Radiant on Compute, Power and the Future of Infrastructure

0:00This episode is sponsored by Deck Dolphin. Fundraising can be tough. You build a deck, you send it out and you hear nothing back. You don't get enough feedback. Deck Dolphin changes that. Get honest, actionable feedback from real VCs within 48 hours. Submit your deck today, go to deckdolphin.com and start booking more investor meetings with a better deck. Hi Mahdi, welcome to Riding Unicorns. Hey Hector, good to be here. Today, everyone, we're joined by Mardi Yaya, the founder and the CEO of Ori, which is now part of Radian. Today, he's operating at the center of one of the biggest structural changes in technology, the explosion of demand for AI computing infrastructure.

0:43Mardi, you started your career building a telecoms business that exited. And I wonder if that's a good place to start, if you could walk us through that early journey and what you learned from it.

0:53Mahdi Yahya:yeah i mean telecoms is a very broad umbrella it was an infrastructure business and we worked a lot in data centers so that was my entry into the whole data center world and what i call the backstage of the internet of how that operates and that was 18 years ago i was 20 years old and you know i've learned everything about infrastructure global infrastructure how data centers operate how networks operate through running that business back then. I didn't know much. I've learned a lot through that process. And then since then, I've been an operator within the infrastructure space. And obviously, being an infrastructure, this is what's happening today is the biggest, most exciting infrastructure built ever globally.

1:41Mahdi Yahya:So I'm quite lucky to be part of the infrastructure world now as well. The path for Ori has been far from a straight line and it's been through various iterations and obviously you're now at a scale that was pretty much unfathomable when you started. The Radiant Brookfield Ori merger values the entity according to Reuters at 1.3 billion dollars so a far cry from the initial pre-seed valuation I'm sure we at FSO weren't invested at I can't remember exactly what that valuation was but can you talk to us a little bit about those the sort of twists and turns through that journey? Look, there's two ways to build a company.

2:19Mahdi Yahya:One is you identify a very clear problem in the market. You build a solution for that. It's pretty straightforward in that sense where you go after the customers that have that problem and then you grow the business on that basis. And then the other way is you try and imagine what might happen in the future and you start building a technology that may or may not solve a problem that might happen in the future. Obviously, Ori was in this bucket. It's a very hard way of building a business, but it's also exciting and rewarding as well. So when we started Ori almost eight years ago, AI wasn't as it is today.

2:57Mahdi Yahya:It wasn't as present or as in high demand. But we had an idea that the global infrastructure stage is going to change and we're going to need new ways of running applications, including machine learning applications back then across edge computing, distributed infrastructure, and the likes. And the idea behind Ori at the beginning was to build a platform that can basically manage that relationship of applications and this new wave of infrastructure that was grown. Over the years, you have to listen to the market, you have to listen to the technologies, you have to understand where the use cases are going.

3:35Mahdi Yahya:And eventually that led us into AI. And with the rise of LLMs like a few years ago, it became very apparent to us that the use case we've been building for all these years is the AI applications that we use today. And we focused the company and we built our technology platform around basically AI infrastructure and delivering applications across AI infrastructure. Can you explain in terms that our audience may understand, given a huge breadth of technical understanding, what is it that you guys actually do? We build physical GPU clusters in data centers around the world. We design those, we build them, we deploy them, we operate them.

4:20Mahdi Yahya:But also, most importantly, we have a software platform that sits on top of this infrastructure that basically allows our customers to run applications or run training, runs on top of training models or running models or inferencing models. Whatever use case they are basically building within the AI space, we only focus on one use case, which is AI, and allows our customers to run these applications that they are building and developing over this infrastructure. People call us a neocloud. I'm not a big fan of that terminology, but that's kind of the bucket that people put us under. And, you know, like this infrastructure is very expensive.

5:01Mahdi Yahya:It's a CapEx heavy business. It requires huge amounts of capital to scale it. And as more demand comes your way, you need to build more and more infrastructure. And very quickly, we realized that in order to scale the business from where we were, we needed two things. We needed a capital partner that can work with us and allow us to scale this business alongside the demand that we were basically seeing. But also, we needed access to power and energy and data centers. Because this is a very big component. Like these GPU clusters, this AI infrastructure is extremely power hungry. It consumes a lot of power.

5:41Mahdi Yahya:You know, we went through a fundraising process. We were talking to a number of investors, and then we met Brookfield, who, if you're not familiar with Brookfield, they're one of the largest asset managers in the world with a trillion dollars under management. They are a very large investor in the infrastructure space, in the data center space, so they own a number of data center platforms, and in the energy and the power space as well. And they set up a fund, a new fund, called the Brookfield AI Infrastructure Fund. and effectively we decided to partner with Brookfield to basically launch a new company which is called Radiant and Radiant today combines the four components needed to actually deliver AI infrastructure at scale which is power, capital, compute and software.

6:29Mahdi Yahya:The compute and the software came from Ori, the power and capital came from Brookfield and that basically allowed us to launch this new entity, Radiant, to tackle the growing demand of AI infrastructure globally. Yeah, Madi, at what point did it click that AI compute wasn't just needed to keep up with demand, but actually much more strategic than that for companies, governments, etc.? Like, where did that tipping point kind of come in? AI itself is becoming a critical asset for governments in general. So when you look at AI on its own, it's very obvious why using AI as nations and entities drive us towards this sovereign conversations.

7:12Mahdi Yahya:Now everyone's talking about sovereign AI and how sovereign AI needs to be developed and deployed and managed. Now, obviously, if you take that a step down, then you start looking at the infrastructure that is running this AI. So when governments basically started to realize that AI is all about data, the data that they have, but also the data that this AI is going to generate through the interactions that we're having with it, through the different tasks that it's doing, through the different applications that are basically running on top of these different models, then you quickly realize that this data is becoming more and more valuable.

7:49Mahdi Yahya:and that's what drove a lot of governments, nations or even very large organizations today to start thinking strategically about AI from a sovereign angle and a control angle. That obviously captures a very large portion of infrastructure. So this AI, where would this AI run, how we can run it, how we can control it, both on the physical hardware side but also on the software that basically runs on top of this compute to manage the different applications that are being used within a sovereign framework. And this is a growing trend, and it's increasingly becoming a topic that is discussed at high levels in government.

8:27Mahdi Yahya:And that also creates business opportunities for companies like ours, where we have a sovereign AI product ourselves that basically touches the infrastructure side of things when it comes to building sovereign AI solutions. And do you see a world where, I mean, there's been talk of sort of governments having their own LLMs and infrastructure? Do you see that as the route forwards? Or is the government's role in this to provide a fertile sort of playing ground for the people who are building? I don't think governments should build their own LLMs. I don't think everyone should build their own LLM from scratch.

9:02Mahdi Yahya:It's impossible to do. I would say it takes a lot of investment and time and energy and thinking and brainpower and so on to build LLMs from scratch. The nearest parallel I always think about in this is like, you know, like the electricity grid you have as a nation. It's a sovereign critical infrastructure component within the country, and you control it. You control all the different ins and outs of it and so on. Not every government is building every single component in the electricity grid to make it happen, and I say it's sovereign, but you build the controls around that. So you can import a variety of different components from all over the world and put them all together to build your sovereign electricity grid and then control that in the right manner and right way.

9:47Mahdi Yahya:So AI is no different than that. AI is critical infrastructure. Intelligence is becoming infrastructure, which is true. It's a utility. And more and more is becoming a utility. So once governments start treating AI as a utility, then you start thinking about it as a sovereign, critical infrastructure piece that you can import models from different places and so on, but you build the controls within the country to actually achieve sovereignty in AI. Yeah, I mean, I absolutely agree. I mean, it is a utility. It's something that at the moment, actually, relatively in the grand scheme of things, relatively few people are using it.

10:23Our power users know the capabilities of it, but that way it filtered down to the wider population. And every worker will be using AI to augment themselves in a similar way to We're using electricity at home and at work. That makes us faster for all sorts of reasons. But for that reason, the space is completely exploding. The amount of money going into this is the foundation. Labs are competing incredibly hard with one another to build more powerful models. And that money trickles down to providers like you guys. How can Radiant find itself an edge in this incredibly noisy market? What is the vision?

10:59What can this become?

11:00Mahdi Yahya:As Ori, we had an edge with technological edge. So we were ahead of a lot of the peers we have in the industry in terms of our software capabilities, our orchestration capabilities, and how we run applications across this new type of infrastructure. With the partnership with Brookfield, we now have also more of a strategic edge because a lot of our peers in the space either look at financing on a project-by-project basis, whereas we have a$100 billion fund now behind us. And Radiant is one of the deployment vehicles of that fund. So we can tackle the largest infrastructure projects in the world today.

11:42Mahdi Yahya:And, you know, like Brookfield are one of the best capital allocators in the world. So beyond just deploying capital, this new asset class requires an innovative ways of actually deploying capital, structuring capital, structuring cost of capital. And our partners at Brookfield are the best in the world at that as well. And this gives us a competitive edge in terms of how we're deploying the capital and eventually trickling down to how we price these things to our customers as well. Of course, the fourth piece is you need somewhere to put all of this infrastructure. Even if you finance it, even if you buy at a certain scale, you need data centers, you need power that is ready, that is not years, years away as well from here.

12:25Mahdi Yahya:And the people who have access to these data centers power, today we have access to about three gigawatts of power through the Brookfield ecosystem across the world. And a lot of our peers in the industry struggle with that. The combination of these four components gives us a very strong competitive edge and creates a very compelling product that has never been done before. You have to think about this, that this type of business is bringing the old world of investing, which is real estate and power and so on. And the cutting edge of investing, which is VC and technology and so on. And it's merging them together into a new type of offering and a new type of asset class in a way.

13:06yeah it's incredible it's incredibly interesting because it's right back to basics the next phase of the global economy in the same way as electricity or rail transport any of these were back in their day and that means you know you have at every layer of the stack if you can call it that people are basically only going to be able to charge a small margin so it's like starting with contractors like building energy sources and it's building data centers and then it's the infrastructure and then it's the chips and then it's the models and then it's the application there on top of that and every layer can only charge a small margin so at the end of the day the winners are going to be the people who can optimize every layer of their own stack as much as possible and having this access to even marginally cheaper capital through Brookfield and marginally contractors who are slightly more efficient who've done this stuff before can squeeze out a percentage here half point here uh it'd be a competitive advantage in the long run exactly and like it compounds so every

14:07Mahdi Yahya:every basis point that you shave off basically creates compounding effect across the whole value chain and you know vertically integrating the entire thing like our marketing team calls it from shovel to model basically so which is really what we're doing today like we we're starting projects where we're building our own data centers we're building our own basically facilities and we build our own compute clusters, obviously with NVIDIA chips, but we design those, we manage those with them. And then our software on top basically gives us a lot of optimization capabilities that we can also optimize the entire stack for the product that we're eventually given to a customer.

14:47Mahdi Yahya:And you have to think about this infrastructure build. This is like a civilizational build at the moment. Our entire civilization is being transformed with AI. So the infrastructure that we're building today is a civilization, but it's what's going to fuel that transformation that is happening today. And the more you integrate that, the more you think about new and innovative ways to build it, not just technically build it, but on the capital stack, on the software stack, on the construction stack, and so on, then the more you're basically accelerating this transformation that is happening globally.

15:24Mahdi Yahya:And that's how we see our vision and what we're doing today as radians. where do you think the spend ultimately gets captured because it's in many ways the opposite of the sort of last generation of technology where value really accrued in the application layer this time around it's more like accruing with semiconductor companies then the foundation labs themselves and probably least of all the application layer but you could take the argument further you could say actually it's going to be construction firms power providers there's an element of skill and of scarcity of product where nvidia for example has defensibility because it's very technical product whereas constructors just don't but where do you think the spend is going to be captured spend goes in cycles and technological cycles that like the ones we're going through it goes in cycles the hardware will always come first let's say the physical not say just the hardware the physical so today the spend is being captured in the physical build out.

16:21Mahdi Yahya:And that includes, you know, the data centers, the builders, the power, the compute, and so on. That's where the spend is happening, I would say, and that's where it's being captured. But we will always go back to the application layer, because at the end of the day, we're building all of this for the application layer, eventually, that will come in later. So, and that happened before as well, we started with the hardware, we built, like, personal computers became a thing, right? And then we started building applications for those, then smartphones became with that you always started with the hardware and then we started we gave a platform for people to build applications that were not possible before but you need to have the physical layer built first in order to go back to the application layer and from a cycle perspective we are in the physical layer build out today it will definitely go back to the application layer over time Mario I want to just tack a little bit and you were a fund two company for episode one and you kind of an E1 treasure at this point.

17:17You know, there's been so many iterations of now ultimately Radiant, but you're kind of a cockroach, you know, like you've survived nuclear war after nuclear war. You come back stronger every time. Where do you get this resilience?

17:29Mahdi Yahya:There's something I tell to the team all the time is the rule of the game we're in is to stay alive. As long as you're alive, there's always a way to move forward somehow. And it's as simple as that. That's how I apply this. Like when things are tough, I always basically narrow it down to the simplest thing. What is the most important thing that will allow us to stay alive and then work outwards from there? And going back to my first point, like building a technology in search for a market is hard because you don't know if there's going to be a market. You don't know if it's going to be customers for that or investors that will believe in this as well.

18:04Mahdi Yahya:I mean, obviously, episode one believed in this from day one. And I think we wouldn't be here without that first check that episode one wrote us many, many years ago. But if you keep that basically simple thinking is the game is to stay alive. That's the goal of this game. Then eventually you'll end up with opportunities and building more and more opportunities to continue scaling and growing. Yeah, super interesting. Can you pinpoint a moment, a sort of non-obvious strategic decision that you've made that has helped with that trajectory of the business? I would say two moments. One was very early on, we decided to acquire another company.

18:47Mahdi Yahya:We were very small. And it goes back to the ability of actually securing talent. And it's very hard when you're a new company in a very unknown industry. So we decided to acquire a company which had three co-founders who had a variety of different skills. And that, I think, was a pivotal moment for us where it allowed us to bring in a specific type of talent. Being a solo founder is hard sometimes. That was one of the ways to actually bring in a form of co-founders into the company that allowed me to take it forward as well. because if you already started your own company, then you understand what it means to have a skin in the game.

19:28Mahdi Yahya:Basically, a conversation I had with them and convinced them to come join forces with us. And it was a form of a merger acquisition, basically with that team who's still there. And now they hold senior positions even within Radiant today as well. And all of the early team, to be honest, I consider them as co-founders, including those guys that came in. And I think that didn't sound logical back then. You know, we were a company. We had no revenue. We were trying to figure out what the market is and so on. And even I spoke to episode one back then and they said, that sounds crazy. Are we going to do it?

20:00Mahdi Yahya:Are you even going to be able to convince them and so on? But I kind of felt that that was what we needed back then. I didn't know the implications of that, how it works. But if we felt that's what we needed, it was counterintuitive on many levels. But it worked out really well, I would say. And that was a key one. the other point was when we decided to start building our own physical infrastructure because we were a software company for a very long time we built our software and we decided that we're going to go and start building our own physical infrastructure in data centers because we wanted to vertically integrate that with infrastructure that we control as well and again that was very counterintuitive to a lot of people within the team within our investors as well, because we're starting to buy CapEx.

20:46And you're talking about buying tens of millions worth

20:49Mahdi Yahya:of equipment. And the company had very small revenues back then as well. So to invest tens of millions in infrastructure as well didn't make sense. That points to a lot of people. But if we hadn't done that, we wouldn't be where we are today. Because we took on a very big challenge of deploying large-scale infrastructure that is worth tens of millions that kind of catapulted the company into the next level which allowed us to be where we are today as radiant as well and those big decisions they seem risky at the time but in hindsight they seem like completely the right move and making yourselves a more critical part of the stack which is ultimately what you're trying to achieve the more people rely on you the more you're going to generate in revenue really in the long term so i want to ask about risk taking and founders those what seemed like big risks but it felt like the right thing to do were they actually risky what were the risks how did you calculate them and do you look back on them and go well actually it wasn't that risky it was the right thing to do they were definitely risky now sometimes you take decisions very risky decisions and they end up being the right decisions.

22:07Mahdi Yahya:My job is to allow the company to make as many right decisions as possible over a period of time. That's how I always see it. And we don't take the right decisions all the time, but minimize the wrong decisions as much as you can is kind of what you try and do all the time. I mean, the first decision was risky in the sense that we were a very small company. Like we weren't a large organization without running business and running revenues and running products and everything else. And we decided let's acquire another company and scale a specific side of the business, scale a specific product or whatever the reason usually companies have for inorganic growth.

22:41Mahdi Yahya:The risk there was that it might not work. The acquisition might not work. That team would not have worked within the business and so on and would have a devastating effect on the company moving forward as well because we made that decision counterintuitively in terms of where we are as a stage of business back then. And many things could have gone wrong and would have had an effect on the company, on the team, on our growth, on the confidence of the investors in us as well. And the other one was we were taking CapEx risk. We were spending tens of millions of dollars in CapEx and infrastructure without proving the thesis back then as well.

23:18Mahdi Yahya:So if it turns out that AI infrastructure was not in demand and nobody wanted it, or if it turns out that we couldn't deliver on it, our software didn't work or our capability to actually manage this infrastructure were not there then that would have been disastrous say and the company would have shut down yeah well i'm really glad i asked that because there's a couple of absolute golden bits in there so maddie we're gonna move on to our final two questions so the first one is a future unicorn prediction so is there another company you've seen early stage that you think has a lot of potential there's a company in the inference space they're called what i like about them like there's a lot of inference companies out there, but they kind of position themselves as like the Costco of inference.

24:00Mahdi Yahya:So like the cheapest inference you can get, because I feel like that's where the world's going as well. And it has a very strong team behind it as well. I've had multiple conversations with them and I feel like they're onto something really interesting there. And I feel if they manage to crack that thesis of actually Costco of inference, there's a big business to be built on the back of that, I would say. And there's also a reason in the robotics, like I'm really interest in robotics space as well i'm telling you saturn dynamics is another one i would say that is doing some really interesting stuff yeah i feel like robotics is sort of bubbling it hasn't quite had its moment yet but i think there needs to just be a bit more talent focusing on that area all it takes is a couple of big funding rounds dexery is doing really well we've had awana on the podcast a couple of times that it's not really like a consumer brand but all we need is a couple of consumer brands to break through and then i think a wave of talent is going to rush towards it and that will be the moment so investors need to be there for infrastructure and then our final question is our dinner party guest game so if you could have dinner with any three people who would they be and this was a tough question because if i had to choose right now i would choose three people connected by lineage and i would actually choose the first human that ever descended from.

25:19Mahdi Yahya:I would choose one of my ancestors from 500 years ago when we started building our modern civilization that we have today. And I would choose someone from my descendants like a thousand years from now as well. And I'd love to have dinner with all these three people and see how that lineage basically affected what things are and what they were thinking back then and how things are in the future as well, which is a world I will never see, which I'd love to see. And I'd love to see it from the lens of a legacy and the lineage that I came from and the one I create from here as well. Yeah, that's awesome.

25:57I love how creative people can get with this. But we've had people say, me 10 years ago, me now and me in 10 years. But we've never had someone look through the lens of civilization like you are. And you really are walking the walk when you talk about those things. because that answer, we would never have predicted that. But actually, in hindsight, it feels like it completely makes sense with how you talk about the problems that you're solving. So I love that. Thank you so much for sharing that. And it's been a great episode. Hopefully it's really interesting for everyone listening. Everyone's thinking about AI and the infrastructure behind it is critical.