RAISE Summit 2025 | Roman Chernin, Nebius

Clips
More from RAISE Summit 2025

Roman Chernin

Co-founder, CBO

Nebius

play_circle_outline Exploring Global AI Infrastructure Advancements: Insights from the Raze Summit and the Future of Productivity Investments

play_circle_outline Introduction of Robin Chermin, co-founder of Nebius, specializing in AI cloud services.

play_circle_outline Insights into changing market dynamics regarding GPUs and enterprise strategy.

Info
Transcript

Roman Chernin, Nebius

Roman Chernin

Co-founder, CBO Nebius

In this interview from the RAISE Summit 2025, Roman Chernin, co-founder and chief business officer at Nebius, joins theCUBE’s John Furrier to unpack the new economics of AI infrastructure. Chernin explains why Nebius built a full-stack, AI-focused cloud that spans everything from data-center design to a developer-ready software layer, giving startups and enterprises a supercomputer experience delivered as cloud.

Chernin shares fresh milestones: Nebius now runs seven facilities across Europe, the US and the Middle East, with tens of thousands of GPUs on... Read more

explore Keep Exploring

What event is being covered and what topics are being discussed? add

What does Nebius do in the field of AI infrastructure? add

What recent trends and statistics indicate about the evolution of technology and its impact on computing? add

bolt Powered by CUBE AI

Roman Chernin, Nebius

search

>> Welcome back everyone to theCube's live coverage here in Paris, France. I'm John Furrier, your host of theCube. We're here for the Raze Summit, two days where all the leaders around the world come together to talk about the advances in AI infrastructure, data platform layers, and of course the applications where the agents will be taking over the world. That's what they think, but more it's a productivity game. But the big story is the AI infrastructure. It's truly enabling lots of value creation, lots of value extraction. Of course, very investment oriented right now. Robin Chermin's here, the co-founder and chief business officer of Nebius. Thanks for coming on theCube. Appreciate it.

Roman Chernin

>> Yeah. Hey. - So AI infrastructure, a lot

>> of investment going in. We're starting to see some visibility into the enablement, and I want to get into that because now as the GPU's get better, the three hundreds are coming, the two hundreds are out. On the Nvidia side, you're starting to see open standards, so you're seeing both camps of AI infrastructure rising up. Thank you very much for what you do. Thanks for coming on.

>> I like the job we do.

Roman Chernin

>> So tell us what you guys do. Explain to the audience

>> what your company is, and then we'll get into some of the conversations.

Roman Chernin

>> So in simple words, what we built is cloud specialized in AI. So we built all the layers starting from the data centers, design, and construction if needed. Then the hardware, then the pretty much sophisticated software platform. And I think this is the main kind of angle, how we differentiate ourselves that we are full stack and we are very much in the software.

>> So you're specialized, what they call it, specialized cloud, NeoCloud, whatever you want to call it. I call it a computer. It's a super computer that's available to the masses.

Roman Chernin

>> I think I like cloud, but because people have a certain expectation how cloud operates. So I think in the evolution of the offering of AI infrastructure, we are moving from, we build clusters for dedicated users to more cloud- like experience. And I think it's important to continue this direction, moving this direction because the more we move in a sophisticated use cases, diversified users from large labs and research institutes, like in hyperscalers to enterprises, the more demand will come to the more cloud experience.

>> Yeah. Consumption paid by the drink and-

>> Developer excellence and so on.

Roman Chernin

>> Yeah. This is really cool. How many data

>> centers do you guys have? Can you share some stats?

Roman Chernin

>> Yeah. I think important thing about us, we are operating on a certain scale in the current shape of the company one year. We'll actually celebrate our first official anniversary next week.

>> Congratulations. - Yeah, thank you.

Roman Chernin

>> And we now, we started from Europe.

>> Our infrastructure footprint started from one data center in Finland, and now we have seven across Europe, US, and a little bit of Middle East. We are operating on by the end of the year, we'll operate, let's say high tens of thousands GPU's. I cannot disclose too much because we are publicly traded company surprisingly, so young but ...

>> Congratulations on that too. Yeah, you don't have to, don't get in trouble. I really want to get into more of the technical conversation because I think I love the cloud analogy because if you look at this past year, just some stats that we've been tracking. Intel's X86 market shares dropped from 73 % to 13% in one year. If you asked me that 10 years ago, I would've said, that's never going to happen. But it just goes to show you where the world's going. Okay. So obviously GPU's, XPU's, the software is now integrating into the hardware. You see robotics, you see AI factories, supercomputer clusters for the masses, not just domain- specific, HPC use cases now opening up. It's causing an explosion of innovation. Entrepreneurs are energized, enterprises are refactoring how they think about distributed computing. And so now you have this kind of ...

Roman Chernin

>> Inflection moment. - It's an inflection moment.

>> And if the tides are in between, and we're going to see once there's a crossover on the enterprise side, I think we'll see AI in everything. That's what we believe will happen, but it's just limited by where do I get the power? Like legit power, the horsepower, the speeds and scale. And that's really the only thing. So I don't want to build it all because the time it takes for me to build the data center or retrofit, it's just-

>> Yeah, economically wise, the cycle of

Roman Chernin

>> the new chips coming online, the new requirements, the physical infrastructure coming is so fast that if you're not specialized in that, if you're not building your economic model around this high pace, it's just difficult to justify why you procure that.

>> Yeah. If you go back 20 years. I've been doing this, I founded this company 16 years ago, this SiliconANGLE. The data center business was buy some real estate from a real estate investment trust in the US or get the buildings, got internet, got some power, get some racks. Now it's completely designed. Can you share your thoughts? You mentioned that earlier, because you're taking an intentional approach to the footprint. I'm sure every square feet meter of their facility is planned, meticulously. Share the design criteria you guys take to go, I'd say full building and full stack.

Roman Chernin

>> So first, I'm not the best engineer in our company. Far from that, but we intentionally go down the stack to, we obviously ran some of the capacity in data center, but the most of deployments we do, we at least design and we have a lot of IP and cooling systems and engineering system to make sure that the idea there that if you design the data center with the idea in mind, what is the rack design then and design the rack with the idea in mind, what is the software on top. You can do just more optimized. And I'm not saying about only cost saving, but it's efficiency. I think that what helps us, the core team went out from the large company that has had the experience of building on all these layers. A lot of people ask at the beginning, "Why you do it? Isn't it slower when you are not just integrating the blocks? " But we think that if you go in the kind of every single detail of the design of entire stack, it long-term gives you the efficiency.

>> Talk about some of your customers, how you guys deploying

>> and price performance comes up a lot. I want diversity of choice. I want my workloads to go super fast. I'll pay whatever it takes within reason, and then I want to have some good inference, but I don't want to pay a lot. I mean, I'll pay, so there's different levels of service.

Roman Chernin

>> Someone wants go on spot, someone wants very dedicated guaranteed capacity and so on. Yeah. So I think we are quite different from some of the players in this market because we ... I'm kidding that our business is quite simple. Like you raise, you deploy, and you sell and then you repeat. You raise, you deploy, you sell. Some other people actually do it in different order because they need sell, then they raise, and then they deploy. So we work quite unique from the capital perspective because we started with the capital. So we were in the position to deploy capacity in advance. So we don't deploy cluster for the customers, but we plan for the future demand. It give us the flexibility to be in a position to provide all types of on-demand short-term reservations, spot instances from capital and capacity perspective planning. But on other side, you need to have a software platform to do it. And we have real cloud experience for our customers. So they can run on demand, they can use spot instances, they can do the reservations obviously. And of course large customers go to the long- term contracts also. But yeah, I think it's very kind of accurate point that we have a lot of diversified workloads. Like you have real-time inference that only cares about latency. You have batch inference that you can, if you do right things, do super cost-efficient because you can orchestrate the workload across leftovers of your compute. And you have large training jobs that require dedicated interconnected clusters. And so we try to combine it all together. And our customer base now is quite diversified. It's mostly still from the i-native world, the guys who come to do their technical savvy. But we see more and more customers coming from the enterprise world. First of all, digital companies, but also life science, also fintech, like hedge funds. And yeah, I think the future is moving from this startup world to more to established businesses.

>> Roman, how about the power requirements on the energy side? Because some build-outs require massive amounts of gigawatts, megawatts, in some cases gigawatts. I've seen folks handle the power quite elegantly where they don't have to be, they can design it properly. If you have the capital budget in mind, you got to look at the power envelopes you got, but then the building configurations could change based on power. I've even talked to some folks that are using air cooling because they design things properly where they're using air cooling. It's not as dense as open standards. Some are more Nvidia oriented. So it's like there's different approaches. What approach do you guys lean to? Or does it differ by deployment?

Roman Chernin

>> Yeah, I think now we are on the stage of building large campuses, let's say up to gigawatt kind of locations. And of course the access to the power is the first question you need to solve. But I think that what we find out that actually the problem a little bit accelerated. So there are certain places in the world when you still can go and build efficiently. And we started with air-cooled systems. Now most of the things we're building are liquid cooled. Yeah. But again, having-

>> It depends on the environment too, right?

>> Yeah. Yeah. - All right. Final question for you.

>> Growth strategy, your plans for the second half

Roman Chernin

>> of the year going into 2026. There's a lot of demand. So it's good. Business climate's good. What's your strategy? What's your growth plans?

Roman Chernin

>> Yeah, so we promised to our shareholders officially that by the end of the year we need to be in a range of one billion ARR. So it's, I don't know, growing 10 times during the year, which is good, but maybe not enough even for this business. And the second layer is how we change the value proposition and offering. Again, I think the most important transition we started from AI startups and now the two dimension we are working on is A, secure enough of data center and power for largest customers of this market, like super labs and hyperscalers and be able to serve their needs. And second, continue to build, upstack our software offering to be relevant for the enterprises. And I think next year, next couple years is the transition to enterprises for us.

>> Enterprises are slowly getting ready.

Roman Chernin

>> Yeah. I think the last six months are super promising because some of the use cases have proved to work and now we see a huge acceleration of the adoption from enterprises. They cannot say anymore, "It's not working."

>> Yeah, they can get ragged, they can get some use cases, they can automate some things. Great stuff. Roman, thank you for coming on theCUBE. Appreciate you.

>> Thank you for having me. - It's great to have you.

>> We'll be following you. We'll do another follow up session.

Roman Chernin

>> We'll do a deep dive. Again, the AI infrastructure is diverse, but the one common thread is the demand for the enablement of super computing capability. Multiple workloads, different kinds of workloads. The AI wave is hitting. It's theCUBE coverage here in Paris. Thanks for watching.