The Rundown - How Nebius is Becoming the AWS of AI (ft. Marc Boroditsky)

Episode Date: October 5, 2025

Nebius is one of the fastest-growing players in the AI infrastructure race. In this episode, Chief Revenue Officer Marc Boroditsky explains how Nebius is supplying compute power to giants like Microsoft, building next-gen data centers, and navigating the gold rush for Nvidia chips. We break down the economics behind the $17 billion Microsoft deal, the rising demand for inference, and growing fears of overbuild in the AI economy.

Welcome back to the Rundown, one of the top business podcasts in the world. On today's show, we are talking to Mark Forditsky, the chief revenue officer at Nebius. Nebius is one of the many neocloud companies that has emerged from the AI boom. They've been one of the most hype names in the space. Their stock price is up over 300% this year. So I had a chance to ask Mark about Nebius's role in the AI boom, how they differ from other neocloud companies like Corweave, their relationship with Nvidia. if he's worried about an AI bubble,

what it's like hanging out with Jensen Huang. I mean, this was such an awesome conversation. It was just great to hear from someone on the inside of the AI hype. So I hope you guys enjoyed today's conversation. All right, let's get into it. All right, guys, today we are talking to Mark Boro Didski, the chief revenue officer of Nebius. Mark, thank you so much for hopping on the show today.

Hey, Zade, it's a pleasure to be here. Thank you for having me. Of course. I want to first start talking about what Nevis actually does, because there's a lot of people listening to this show that've heard the name. They've seen the stock price go up. They think it's somehow related to the AI boom. So I want to turn it over to you. Can you explain to us in simple terms what Nevis does?

Certainly. We typically position ourselves as a full-stack AI infrastructure player. What that means is we provide the hardware, the software, and the software. the services that an AI engineer would need to be able to build the special AI applications or models or services that they're standing up. You can think of us, if you go back to the early days of cloud, compute, and storage like AWS. We are the AWS of AI. Okay, it's the AWS of AI. I'm curious, though, like, how do you, you know, we've seen the name neocloud being thrown around a lot, right? Like Nevis is a neocloud, and there's other neocloud companies out there as well.

How do you stand out from the other neocloud players like a core weave? Or is it very similar? You know, there's overlap between us and the other players. The category of neocloud is sort of hodgepodge of companies that are bare metal like core weave. all the way over to companies that are just software that are a layer on top of the infrastructure, the hardware infrastructure that you're going to need. The way we position differently is we are the full stack player. So we think of ourselves more as a neoscaler, so the next version of the hyperscalers, as opposed to just a neocloud, which could include portions of the requirements.

You know, the vision for the company is to meet the AI engineer where they are. AI. Engineers are building really important special things. They shouldn't have to think about cobbling together or integrating or managing their environment. It should just work. And that's what we're trying to accomplish with Nebius. So if it's so if I'm if I'm a company and I'm like, okay, I need a ton of I need access to invidia chips. I need access to to to compute. I don't know where I'm going to get it. I can't just go build a 15 billion dollar data center. I call I call you mark and then you can get a set up with Nebius, whether it's bare metal, which I'm assuming means access to the chips itself for the computing power, or you have other services as well that you can bring in that can help the AI engineers.

That is correct. That is correct. On top of those chips, you're going to want an operating system, you're going to want virtualization, you're going to want a whole bunch of tools to be able to manage models, to manage your data flows, to manage setting up inference and operating inference. We supply all of that. So you as an AI engineer go on and build what it is that you want to be. built. Gotcha. And I read an article recently. You guys announced a major partnership with Microsoft. So it seems like you guys are, you know, NetNabias is making some partnering with like other hyperscalers to supply them with AI computing power. According to Bloomberg, it was at least $17.4 billion through 2031. I'm curious to know, though, like these hyperscalers, they have a lot of money. Why do you think that they're choosing to rent? through Nebius versus just building it out themselves?

Well, I think their preferences to build things out themselves. The pace of the market that we're in the midst of is unprecedented. There's no other category that's grown at the rate that we're experiencing. And this is real growth. These are customers buying infrastructure to be able to build whatever they're building. And I think that right now in this moment in time, the demand that they need to service is more easily serviced by companies like Nebias. So for us, it's a great privilege to not only be able to service them, but to be qualified and supported by them.

It's going to help us to scale, and it's helping us to be able to execute on our core mission, which is to be that full stack vendor for the rest of the market. And, okay, that makes sense. And I think as part of this deal, I was reading that Microsoft gets access to 100,000 Nvidia GB300 GPUs, which is their state of their art chips. which brings me to the conversation about, like, Nevis's relationship with Envidio. I'm so curious to know how all that works, because there's so much demand for Nvidia's chips. Everyone says it's like the most in-demand asset in the world to get access to these state-of-the-art chips. How do you make sure that you're in the front of the line to get access to these chips? Are you texting Jensen every day?

Are you wishing them happy birthday the first thing in the morning? Are you getting dinner with it? Like, what does that look like to make sure that you guys are the ones getting their chips? We have a very special partnership with NVIDIA. It's a privilege, to be perfectly frank. We are actually supported in a lot of different ways by NVIDIA. I mean, obviously, we're a big seller of their chips in a form that actually provides a high degree of flexibility in the market. So for them, I think they see us as an important channel to meet demand where it is.

For us, I think Jensen's taken very seriously the importance of relationships like Nebius, and his entire leadership organization works with us to make sure that we are actually appropriately delivering the right kind of capabilities, meeting customers where they are, and helping us to be able to develop the market for our services, which benefit them. And I think NVIDIA is also an investor in Nebius as well. I think it's a $700 million investment. last year. But they also have a stake in core weave as well, which is why I'm always interested to know, like, how does that work? Like, how do you make sure that, like, you're the one that

gets the access versus their other relationships that they have? It's got to be so competitive. Oh, it is very competitive. But it's a massive market, and there's a ton of opportunity out there. And the kind of customer that we are servicing, as an example, we're very heavily involved in the Nvidia Inception program. Inception is them helping startups. As I mentioned earlier, we're meeting the AI engineer where they are. We spend a lot of time with the VC community and the startup community. We're very privileged to have this fantastic relationship helping startups with

Nvidia through Inception to actually get the compute they need and build the special things they're building. Corwin's go-to-market is different than ours. We are overlapping and competing periodically, but candidly, we see the hyperscalerscalers more frequently. that's the real competition. And that's where we're focused to make sure that we're actually making inroads into the rest of the market. I think the theory right now is that Nvidia is more willing to sell their chips to Nevis versus the hyper-scalers because companies like Amazon and Google are developing their own chips, right? And, you know,

Jensen's a smart guy. He wants to prioritize people that are going to be buying more invidia chips in the future and not trying to replace them. Do you think that plays a factor in all this? It does. It does. There's no question about it. I mean, if I was Jensen and I'm far from it, so take this as a very minor opinion, but if I were Jensen, I'd be focusing on people that are going deep and investing specifically in helping him to build his business. The hyperscalor business is different than the nebius business. Hyperscalers make money by ensuring that things are highly systematic. systematic, cookie cutter in nature, so effectively commoditized. Their desire is to actually drive down the overall price of compute so they can actually sell,

price to them, that is, so they can sell services on top at greater margins. Our business is specialization. We are looking at the specific use cases and requirements that the AI engineer and the AI companies have that we're able to help them to accomplish. In this period of, let's say, dramatic market growth, a lot of innovation is necessary. We're investing in making sure that we're delivering the capabilities that are going to make a difference for those AI engineers. I think the pace of our investment is going to outpace the hyperscalers for a number of years. And we're going to be able to service important workloads that go beyond the capabilities that are embedded in the hyperscalers' platforms.

But at some point, you want to be, like you mentioned early in the interview, like you want to be, be the AWS of GPU. So you want to eventually get to that level and kind of offer the same level of services and stuff that the hypers are offering with the high margins and everything. 100%. And that is, that's the expectation that the kind of applications we're going to have in the future are going to be very different from the applications that were built in the past.

I don't want to dive into the technical details, but the reality is the architecture of future computing experiences is going to be dramatically different. And we are building towards that future vision. And we want to be that supplier that's going to be delivering the full range of capabilities that allow our customers to meet their requirements in a consistent, reliable fashion. I'd like to get your take on the concerns these days around the circular financing nature of, you know, especially Nvidia.

They're investing in a lot of companies, NeoClouds like Nebius. what's your take on that? Is it something to be concerned about or do you think that it's just kind of blown out of proportion? Well, this is a massive market. And I think you've got to take everything into consideration in the context of the broader market. We are on the brink of a multi-trillion dollar transformation that's taking place. And yes, there are some questionable transactions that are happening out there. Capitalization is a critical.

component of what's going to be required in order to be able to meet this multi-trillion dollar opportunity. And for us, at Nebius, we're very focused on making sure that we have a solid capitalization strategy. So as you can see from the transactions we've been completing, we have been relying on utilizing our value to raise additional capital resources, the convertible offerings that we've done, the equity sales that we've completed, allowing us to leverage our increasing value in order to be able to turn around and build more capacity. And even the transactions that we're doing, like the Microsoft deal that we mentioned earlier, that's helping us, again, to be able to finance further capacity expansion.

So our approach is to make sure that we have the capital necessary to continue to grow at the rate that we are. Yeah, because there's so much demand right now, you want to get as much capital as you can to keep building and building because you can't meet the demand right now. 100%. 100%. That's the reality. I mean, if we could, we would. And that's what we're trying to do. And by the way, it's not just down to capital. We have to build these data centers. We have to actually cite these data centers. You know, we have to actually put them in the markets where there's

expectation of having a location. So there's a whole physical delivery aspect here and, you know, a local readiness aspect to this that constrains the pace at which that we're putting these data centers in place. That's a good transition to talk about the actual physical data centers, because I think you're right. Most people don't realize it. Like, it's physical infrastructure being built. It's buildings being put up, pipes in the ground, cooling systems, power, which I think is a big thing right now, right? I'd like to get your take. Are you, are you more concerned about getting access to chips, or are you more concerned about getting access to enough power to power these massive data centers? The pendulum swings. I mean, you know, the, you know, the, you know,

the most recent, I don't know, half a year or so, it's been data centers. We're now getting into the main delivery period for the next generation of GPUs, the GBs. And if you go back and watch the delivery experiences that took place with the previous generations, there's going to be a supply bottleneck at some point. But the reality is that you're going to see us jump from one, let's call it, scaling challenge to another will ultimately meet it. You know, the reality is customers want compute tomorrow. We're having conversations with them about how do we actually partner around your compute requirements,

not just tomorrow, but for the next year or two? Because tomorrow's problem turns into your scaling problem next quarter and the following quarter and the following quarter. So being able to move their demand to the period when we have the supply is what we're working. And actually creating the means within the way that we're selling, the way that we're building, so we can actually drive expansion

in a very methodical fashion. We want to be able to not just support you for that spike in training that you're doing tomorrow. We want to be able to support you as you scale over the next several years. I think originally there was a huge demand for the GPUs for model training, right? Have you seen that switch over to more of the compute,

which is like for people at home? That's like when you run a query on chat GPT, that takes computing power to run those. Are you seeing that switch over from the training to compute now? It is. It is. And by the way, to be specific, the technical term is inference. Inference, thank you.

That's where the request when you're, you know, clicking on the application on your machine sends the request to the service. It's actually looking for an inference that generates a token and sends that back to the application that then presents in your browser or the mobile app you're using. Yeah, we are seeing an expansion and inference. That's exciting to me because the reality, all due respect to the model builders that are out there, keep doing what you're doing, critically important. We are nowhere near the maturity necessary to meet all the specialized things that people are trying to do.

But we, in the consumer world or the business world, should be watching, how does the actual utilization grow? because that's ultimately the application being utilized, commercial or consumer value being derived, and that's where the revenue opportunity gets generated for those applications and services. We are seeing apps dramatically expand. Actually, we've seen customers and partners that only focus on inferencing. And that's because they're actually now scaling out an important application. Yeah, and that seems to be like the big opportunity right now.

And I think that was like the concern early on. everyone's like, well, these models are going to get more efficient, these open source model that's going to require less and less compute, which is like what a deep seek, deep seek scare was earlier this year. But now with all these, with all the inference demand from these thinking models from all these AI applications like video generation, which we're going to talk about in a second, it's just increased the inference requirement, which I guess is good for you guys because you guys can provide the compute power for those applications. That is correct. That is correct. The interactions that are happening are multiplying.

Yes, the infrastructure, no question about it. Models and hardware are going to get more and more efficient. If anybody believes otherwise, you've not been in the tech industry. You're new to this. The reality is that we're going to see quantum improvements like the deep seek moment over and over again. That shouldn't be a surprise. What people should be looking at is what's the future requirements look like? The future requirements are dramatically more compute intensive and dramatically more complex than what we're experiencing today. Imagine a world where it is a live, full-frame, video interaction that you're having with, you know, an avatar that's having a, you know, a human real communication experience with you on everything. You know, autonomous driving, robotics, the app that you used to have on your phone.

own, all of that's going to require a dramatically greater number of inference interactions to be able to make that, you know, real experience believable. Has the thought of overbuild ever creeped into your mind, even when you're getting lunch or you're waking up in the morning, you're like, we're spending a lot of money, building all these data centers, billions and billions of dollars. Is there, is there any concerns about overbuilding? How would you know? How would you prevent that from happening? I'm just curious to know what you guys are thinking about that. It's a very good question. It's a very good question. I mean, it's happened in every single previous tech wave. And like every single, every single previous tech wave, you have a reset and then you have a catch up. Okay. My suspicion is we're going to have the same occur in this category. There's no question about it. It's very, very hard to see past all of the excitement. and exuberance around what we're able to accomplish.

And yeah, you do see overinvestment. The difference, though, with this tech wave, and I think this has been spoken about quite a bit, is how fast things are happening. You know, I've watched startups in given functional or technical areas, you know, raise $100 million, try to build something, and then disappear. And when I go find the founder, it's like, what happened? They said, well, it didn't work.

Or, oh, you know, we miscalculptial. whether there was actually a real use case there. Or Zodd hired them for $300 million. Hopefully that's where they're all going, but there's a lot more of them than there are people being hired. The point is that I think the fail-fast reality is going to make that reset period a lot faster. The other thing that's happening is it's not just startups that are fueling this.

We're starting to see the real adoption in the enterprise. It's starting with major software vendors, and we're seeing very specific use cases in like pharma and media, you know, all the way across enterprises that are showing us that there's going to be significant scale adoption there, and we should be watching that very closely. That goes back to your statement earlier, looking for the commercialization. We need to be watching that, and my suspicion is we probably have a little bit of a plateau, but then we'll see the growth curve occur again against enterprise expansion.

My concern here is that we're spending, you know, a lot of money on buying these AI chips, these GPUs, which what I've read is that their useful life is between three to five years. So even if, so all this money that's being spent on the overbuild is not going towards infrastructure that's going to last 50 plus years like the railroads were or all the fiber was from the previous bubbles. It's going towards chips that are not going to be useful in three to five years. Is that a concern for you guys at all? No, because today we're constrained by the capacity that's out there, the amount of chips that are available, and the demand that we have far outstrips the number of chips. Now, the question is more two to three years in the future. Yes. And again, this question of or the expectation that the demand for more compute is going to go up.

No question about it. We're going to need more and more capabilities on the hard. Local news is in decline across Canada. And this is bad news for all of us. With less local news, noise, rumors, and misinformation fill the void. And it gets harder to separate truth from fiction. That's why CBC News is putting more journalists in more places across Canada, reporting on the ground from where you live,

telling the stories that matter to all of us. Because local news is big news. Choose news, not noise. CBC News. Okay. When I sell my business, I want the best tax and investment advice. I want to help my kids, and I want to give back to the community. Ooh, then it's the vacation of a lifetime.

I wonder if my head of office has a forever setting. An IG Private Wealth Advisor creates the clarity you need with plans that harmonize your business, your family, and your dreams. Get financial advice that puts you at the center. Find your advisor at IG Private Wealth.com. I got you again. Oh, I got you. Yeah, you're back. Okay. Something strange is happening with my iPhone. I don't know why. But the compute demands are definitely going to be going up. Another interesting observation, people are still using prior generation GPUs. Even though the bees are out today, the Blackwells, and the GBs are the Grace Blackwells are just getting deployed, we're selling the hell out of hoppers. Okay. And those have been out for several years. And I've actually got customers that have told me, can you get me 10,000 hoppers? I'll take any hoppers you have.

And what they're doing is they're utilizing them for other scale requirements. We're going to see a range of requirements. And the most advanced chips are going to be needed for the most advanced workloads like training. And the production workloads like text-based inference will be using hoppers. Video-based inference? We're going to probably need to use more high-performance chips. be able to support those requirements. That's a great point. I didn't even think about that, how, like, the less compute stuff

can go towards the older chip. So, like, maybe their lifespan is longer than three to five years. They can be used for lower compute stuff. That's a good point. I have to ask, though, right now there's a lot of concern with the power demand, causing the price of electricity to go up in areas where there are data centers being built. how do you think that there might be some regulatory backlash for for nebius and other and other data center builders out there that is causing you know these prices these electricity prices to go way way up and and do you think that there's something that should be done about that well there are already difficult regulations and a lot of the let's call it densely populated or high density markets so

the reality is we are dealing with a greater degree of regulation in, you know, like New Jersey or New York or L.A. In less populated areas, it's not as big of an issue. And by the way, there's, we have a lot of flexibility. We don't need to put a data center in a specific location. We've been building in places like Iceland and Finland. And we as a supplier are looking for the right combination of location, capital, and overall market opportunity. And we're going to figure out how to actually get the combination in the context of regulation, right. So, yeah, I would expect that you're going to see in some markets more difficulty in building data centers.

It's going to make the economics more difficult to have locally hosted GPU capacity and will be building adjacent or in the next country or in the next region, and people will just get the service from there. It'll make the experience more difficult, maybe a higher degree of latency as we get more advanced applications out. there and then maybe then we'll see a deregulate, you know, a reduction in regulation that opens up the capabilities in those tiger markets. I think the best case scenario here is that like we're going to see a huge boom and power generation in the U.S. hopefully with more power coming online, renewable, nuclear, all kinds of power to meet the demand for these AI data centers. And then hopefully

that's going to result in too much power generation and that's going to result on lower electricity prices for everybody. I think that's the best case. scenario of this AI build out is we just get a ton of power and we get cheap electricity for the next 20, 30 years. You've definitely seen a renaissance and power generation. There is no question about it. And you can see it in the projects that are taking place and the reconsideration of all forms of power, even nuclear.

Yeah. In the United States, that's a bit of an unpopular topic. But you can definitely see that there's a motivation to expand the grid's capabilities. We're not completely dependent on that. You know, we can go into a location and. tap natural gas, put a turbine in place, and then an independently powered data center that allows us to continue to build on our plan. That's awesome. And yeah, I think that's a space that gives people it gives people some nerves because they're watching their electricity bill and that's something

that they can tangibly feel. But that's going to be something interesting to watch how all that is handled from a regulatory side and from like how, you know, hypers and companies like Nebius react to that. I have some lightning round questions as we wrap up. wrap up today's interview. I want to get your take on, have you tried this viral SORA app from Open AI that came out this week? Have you tried it yet?

Not yet, not yet. Very excited to try it. I'm actually curious, when you see an app like that, are you just like, yes, more compute or more inference, which means they're going to be giving you guys a call for a potential big deal?

No, I think about it differently. Raising the bar by pushing the frontier of expectations, expectations raises a bar for everybody. So the way I think about it is I'm excited as a consumer, as a business user, and I'm excited to see what the other suppliers end up doing, you know, the Anthropics and what have you. And more interestingly, I'm excited by what the disruptors are doing because they're going to have to leapfrog. All of that collective activity that translates to GPU capacity that we can supply. So like all rising tide lives all boats and that's going to be.

Exactly. Exactly. And that kind of, you kind of mentioned earlier as well that like you're seeing more adoption of AI across enterprises and everything. I mean, how are you using AI in your day-to-day work? I mean, are you using like a chat GPT or another AI tool to like review your agreements with Microsoft? I'm curious to know how you're using it. Yeah. So in our back office operations, we utilize all types of AI tools in order to make sure that, We're getting very efficient capabilities and processes in place.

We're actually in the middle of a, boy, I'm going to get flooded with inquiries. We're in the middle of a GTM infrastructure buildout, and we're building an AI native GTM infrastructure. So we are going to be looking at all the tools that are out there. So yes, everybody's. You're about to get 300 emails on Monday morning. I'm sorry about that, Mark. My reason is now just going to pile up. I already feel it.

I welcome it because I really want to see what the innovators are doing. In my personal life, yeah, I do utilize, I love perplexity for doing research. I don't know if that's just because I landed on it and it worked for me. I also enjoy granola. Granola lets me be present and not have to be sitting there keeping notes and then still be able to have a strong and representative set of notes that I can use for whatever purpose that comes afterwards. Yeah, granola is great. It's an AI, it's an AI note-taking tool that you can run in the background for people that are unfamiliar.

Great tool. I use it myself as well. Last question, and then we've got to get out of here. Have you met Jensen Huang yet? I did. Jensen cares deeply about his partners. And GTC Paris, he came and spent time with us. Was he wearing the black leather jacket? He wore his black leather jacket. The last time I saw him, he had to explain why he was wearing a suit. And that was at an event that was in the London just a couple of weeks ago where he was on stage with the prime minister of the UK and he was wearing a proper suit. So he meets the customer where they are. So that black leather jacket isn't always on. Is he as cool in private as he is on stage? Because man, that guy is just the charisma coming out of him is just insane. He is and knowledgeable and genuine, really genuine.

He came, spoke to the Nebius team, was very aware of what was going on with the company, He was very aware of what was going on with our founders, you know, made shoutouts for Arcadi, our CEO. Yeah, he's he is a he's a consummate statesman type CEO and very impressive. Awesome. Well, that was awesome, Mark. Thank you so much for hopping on today. I learned a lot and hopefully, you know, when you guys announce your next big deal, you can come on again and we can have another chat. Zaid, any time.

I'd welcome the opportunity to catch up another time. Thanks again, Mark. Take care. Thanks. Well, all right, guys, hope you enjoyed that conversation with Mark. I really enjoyed talking to him. It was just really cool talking to someone that is on the inside of this massive AI infrastructure buildout. You know, it'll be interesting to see what kind of role neoclouds end up playing moving forward.

Are they going to end up disrupting established players like AWS? Or are they going to be the first to blow up if this AI bubble pops? We'll see what happens. I'm definitely keeping my eye on Nebius. And hopefully we'll have Mark come on the podcast again soon. Let me know what you guys thought about this conversation in the comments on Spotify. or YouTube, are you bullish on Neo Clouds, or are you more worried about the price of electricity going up

with all these AI data centers being built? Let me know in the comments. And while you're at it, don't forget to hit us with a five-star rating on Spotify. Hit us with a thumbs up on YouTube. That engagement really helps us out and it helps other people find the show. Thank you guys so much for listening, watching, and commenting.

Shout out to Mike and Connor for all the work behind the scenes. And we'll see you guys back here on Monday. In lasagna, medium power, 15 minutes. Sounds like Ojo time. Let's play. Feel the fun with Play Ojo. The online casino with all the latest slot and live casino games. What you win is yours to keep with no wagering requirements.

Instant payouts and no minimum withdraws. Hey, I just won. Woohoo. Feel the fun. Play Ojo. Honey, forget about the lasagna. Let's celebrate. 19 plus Ontario only.

Please play responsibly. Concerned about your gambling or that if someone close to you, call 1866-531.2600 or visitconnxontera.