In The Arena by TechArena - Nebius at SC25: Building the Neocloud for Enterprise AI

Episode Date: January 23, 2026

From SC25 in St. Louis, Nebius shares how its neocloud, Token Factory PaaS, and supercomputer-class infrastructure are reshaping AI workloads, enterprise adoption, and efficiency at hyperscale.

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 Allison Klein, and this is another Data Insights episode, which leads on here with Janice the Rescue, Solidine. Hey, Janice, how are you doing? Hi, Allison. Doing well. How are you? Good. We're here at Supercomputing St. Louis, which is always one of my favorite events of the year. It's so cool to hear about how technology is being used for scientific insight and the next discovery that's going to shape the planet. How's the week been for you?

It's been a great week. There's been a lot of discussion about the planet, about high performance computing and believe it or not, cooling, right? But also just infrastructure in general. So we have a really cool guest with us. I am so excited for this interview. Tell me who you brought with you and what we're talking about today. I'm very excited because today we have Dan. Daniel Bounds from Nebius. And as most folks know, the Neo Cloud is a fascination, right?

It's a new way of doing cloud, and it's not seen as the traditional enterprise. And I think what you guys have achieved so far, it's just simply amazing. So welcome. Thank you. It's a pleasure to be here. You know, Nemius has been on an absolute tear to scale. Tell us a little bit about what you've been up to. I've been with Nebius about a year. In that time frame, this industry has really just taken off. I think we've all been waiting for decades for all of the necessary componentry, software, hardware infrastructure, capital, everything to be in place. And this is the right place at the right time, especially for supercomput. When you think about all of the amazing things that we create together through AI, this is a special week for us. But we just had earnings last week. We're

selling everything we've got, which is amazing. We're building at massive scale. And even as we look forward to next year, we're looking at run rates in the $7 and $9 billion range. And so it's truly amazing. And also the customer stories that are pulling through at this time are inspiring and propelling us forward. Now, I think that one of the things that I do is follow the neoclod space so carefully and the new providers of compute hardware to fuel AI. I've been so intrigued by Nebias. Can you tell me a little bit about what you think sets Nebriots apart? So when I answer this question typically where I go immediately is the people. If you look across the neocloud landscape, everyone is in a race to build more faster. We've all looked at accelerated computing as the answer.

We've all looked at purpose-built infrastructure as the answer. And so the question becomes, how do we create that build it fast enough with the high enough efficiency so that we can actually power these things and achieve output? So the people that we have that are making that happen, four or five hundred engineers from guys who know how to develop silicon up through the rack in the row to the folks who understand Kubernetes and sluram to operate the environment at hyperscale, but also build really the infrastructure of a traditional supercomputer that everyone has access to. That is really what sets us apart versus the rest of the Neocloud competition. Yeah. And, you know, supercomputer is always a really great place to gauge where the industry

is headed. What are you guys seeing in terms of how AI is kind of evolving the technical compute space? So technically, we're on the cusp of yet another inflection point. These inflection points are happening really fast. If you look at the performance of the GPUs, of the high bandwidth, low latency fabrics that we're deploying, and of the storage, you know, data is such a critical component to all aspects of AI, from ingestion to preparation, to the way that we compute on GPUs, to really infusing that into applications, data is the common denominator. And so very excited about the innovations that are happening all along that data pipeline for AI workloads and how that's unlocking a lot of value. For us, we look at it through the lens of the industries. We're incredibly excited

to serve as some of the biggest foundational model builders and now a ton of AI native startups that are going to be those foundational model builders, that wavecress has happened. Now we're also seeing the at-scale enterprise class inference that everybody can actually monetize. And that's the big inflection point that we're really excited about right now. And as supercomputer, we're talking about that,

not just in terms of commerce, but in terms of development and research and the promise that we've all talked about from an AI perspective for a very long time. You know, it's interesting that you say AI for a very long time because, yes, we've been working on this for a very long time, but we're still kind of in the efficacy of this era of really understanding the workflows that organizations are going to be operating, you know, LLMs are just getting to that maturity phase where enterprises can start

deploying workflows based on them. When you talk to customers, what kind of workflows are they talking about today? And as you look forward in the landscape and you think about things like congenetic computing. How do you see those workflows changing and morphing in terms of hardening workflows that enterprises are adopting at scale? Yeah, I think there's two aspects. One is maturation. More people are experimenting more broadly than ever before, and that's because they have access. It's not just the biggest of the big who can access supercomputer-like performance. Now, any startup, any enterprise can go in. We have a product Token Factory. It's essentially a PAS platform for AI where you can go choose from a range of models

off the shelf and you can start to play with them and then you can start to scale them. And to your other point, you can do that and match the ambition of your company from a quality, from a resiliency, from an SLA perspective, and more importantly, from a compliance and from a security perspective, what we've seen is a maturation of all of these components now so that enterprise class, customers can deploy in with cost seats. Okay, cool. Love that. So, you know, when you're working with our customers,

how do you see you guys kind of standing apart in your work and how you're partnering with these guys? How is it helping Nebius? Yeah. So there's a lot to know in AI and it is part hardware, it's part software, and software can take many forms, you know, from the orchestration layer to the way that you schedule jobs and distribute workloads. All of that takes a lot of engineering.

engineering horsepower. What sets Nebius truly apart is the fact that we bring all of that from an open source perspective. So we're one of the biggest contributors to Kubernetes, massive leader in the area of slurm. We have an MLOPS layer that we deploy that helps developers come and build on top of Nebius. We say this a lot internally. We want to be a community of builders. And a lot of that for us is offering this wide range of capabilities, whether you want it just from a past perspective or an IAS perspective, if you want to reserve capacity. And now we have big companies like META and Microsoft

that are building bare metal, but taking advantage of what Nebius brings to bear. I think we're uniquely positioned in the market against other purpose-built AI clouds because of that full-stack approach. Others will say it. Ours translates into results for customers. That's incredible.

I think that we talk about the community of I think about one thing that I know all customers are going to talk to you about, are performance and then cost and that careful balance between those two vectors. When you talk to your customers, how do you navigate that careful balance between performance scalability, efficiency, both from a cost and an energy perspective, and then the actual workload delivery and innovation and leading room for innovation within the way that you're working together? Yeah. So innovation is at the top of our priority list. Everything that we do can always be perfected. And so some of that, you mentioned performance. I saw something on your site around ML Perf 5.1. The industry advances. And when we talk about generative AI performance, we want our cloud operations, our clusters to perform at the absolute top end of that. We're at supercomputs. We have number 13 on the top 500 list. I want to say like number 39, 49, somewhere in there. This is coming from an AI cloud provider that just submitted results. It's a testament to the fact that we have a heritage of building supercomputer class performance with the flexibility of a hyperscaler. That's impressive.

And also, if you filter that list through commercial availability, I think we're like number two or number three in the world in terms of performance. The other thing that's really important for us is as customers look to deploy on Nebius, especially as we talk about enterprise customers, workers kind of work. And so meantime between failure at a cluster level has to be sky high. They have expectations that they have built over decades of operation on premise, and even in the hyperscale general purpose clouds, that now a purpose-built AI cloud has to merge. And for us, that's a key point of differentiation of value that would apply customer. You know, it's funny, and we need to get to our next question, but when you consider that for a second,

And you just said, you're like, when you consider commercial availability, it's number two. We are moving so quickly. The fact that we were not even talking about NeoCloud a couple of years ago at super computing. And now Nevis is sitting at the positions that you are incredible level of innovation and deployment in such a short period of time. They only impress me. I know we all share a lot of history in this industry. And I keep coming back to the same phrase of there's no place else I'd rather be right now. I think the promise of AI across our lives is commits, almost incalculable,

but also the economic value in what we're doing to change the way that computers have affected us. This is an important moment of time. And so we love the opportunity to work and partner with our customers to build that future. Incredible. Yeah. So how do you see different parts of your infrastructure meeting to deliver on customer requirements specific to, say, storage requirements. Yeah. Like I mentioned earlier, data is paramount. It's critical. What we find is most neoclows at this point have really majored in compute and networking. We offer a complete range

of storage solutions across that data pipeline that I talked about. Of course, it's file an object, and as we continue to not only mature with LLMs, but in multimodal models, all of that comes back to the underlying data performance that we have. And so we've invested heavily not just in our own capabilities, but also partnering with the likes of Weka and VAS. A lot of that underpinned with the absolute latest bleeding edge technologies like QLC, partnering with folks like Solidim, so that we can deliver to customers a high performance environment that scales with them, creates tremendous flexibility. We don't know what they're going to do next.

We're doing it with them. And so from a Nevis perspective, it's incredibly important that we look at our customers as partners. we develop, we build, we iterate, and help them secure the future. Love that. I'm an infrastructure geek. So whenever anybody talks about AI infrastructure, I think it's so interesting how storage architectures have become so critical and so central to this moment in AI. And I love what you talked about there.

I think that the other side of that is just all the data that it's bringing into an environment and that security and privacy needs to be really carefully managed. How are you managing that with your customers? And how do you do that across an environment where some data is residing in a nidious environment? Some might be residing in a customer environment. So I think the paradigm is it's not a one-size-fits-all. So when you look at the way that we approach the market, I mentioned our PAS platform, that's for folks who don't want to go and stand up a cluster and have to manage it.

They just want to hit the easy button, and they want that easy button to turn into an everyday button. We have reserve capacity. We also have bare metal. We also have what we call a white label service where we have either municipalities or governments or nation states that are looking to deploy some level of sovereignty around the cloud. We can do that. But as we look at servicing a broader spectrum of customers, security and compliance are expectations. And that's the way we look at serving the enterprises. What are the expectations that they have built and honed over time? Do you have SOX compliance, you have COIPA compliance, you have all of these different standards that are table stakes. Do you do that with a full range of security capabilities that we would expect in a traditional, either hyperscale cloud or an on-premise cloud? We have very quickly addressed a ton of those, and we've got a quarterly release schedule that has continued to strengthen and strengthen over time. We'll exit this year, I believe, as the preeminent AI cloud for enterprise adoption. by feature set. That's how it's correct. And it wouldn't be at any conference, supercomputer otherwise,

and we didn't talk about efficiency, right? And efficiency has come up quite a bit for NABIA. So what are some of the things that you're doing with your technology and making efficiency a part of your DNA? Yeah. What are kind of the guideposts that you follow? I remember a few questions ago. I talked about a full stack approach, and that's not just software, but it includes software. So again, something that sets us apart, we design our own systems. So from the fence that we choose on the motherboard to the efficient drives that we put into our compute and storage chassis to the way that we build our data centers, we have a data center in Senate line with insane PUE, so much so that between the free air cooling and the fact that we pass the heat back into the local community, we heat their houses, thousands and thousands of houses benefiting from the operation of this data center. So efficiency is absolutely in our DNA, but it does it in there. We're constantly looking at this balance between performance per watt, performance per watt per dollar,

because it has to be commercially viable. It has to be highly efficient. It has to be something that yields results for the customers. So each one of those things factor in. And on top of that, every customer is slightly different. They all want to build the infrastructure that matches their AI ambition. And so we take that into account as well. So I'm very confident in the level of overall efficiency that we're building into our entire stack.

But I really love when we have stories with individual customers where they can maximize for their use case. I love that example you gave too. That's just really cool. Finland data centers, yeah. I love the heat repatriation examples. The thing that I'm thinking about was start of the interview about Nebius has been on a tear. You've also been on a tear. Your career trajectory is so interesting.

and joined Nibius a year ago. You did that because you saw a future that was bright. What are you most excited about in terms of the future at Nubias? So we've already acknowledged that things are moving fast, and they're moving in ways that not everybody can foresee. And I would venture to say no one can perfectly foresee it. Where we're moving next is this focus on individual segments of the market. So protein folding is a great example inside of health care and life science.

inside of bioinformatics, inside of, you know, there are use cases where AI has really started to have an impact. And that gets our juices flowing because then we can start to think about financial services. We can talk about retail. A lot of people don't know this. Nevis also owns a autonomous driving vehicle company called A.V. There's tremendous innovation that's happening in robotics. And autonomous driving, AI is the backbone. So it's the application in.

daily lives in enterprises beyond just this idea of having agentic support, that's important. It's a first step, but there's so much more than we can do as an industry. And I think supercomputer is a great place where we can have that conversation. Sure, that sounds so. Yeah, I mean, folks need to know where can they go to learn more. So if they want to reach out to Nevis, you or figure out more information, what do they do? Reach out to us. There's an easy contact us button on nevius.com. We will respond immediately or as close to immediately as humanly possible.

But yeah, reach out to us. We're here at Supercompute. We'll be at all the GTCs where almost all of the vertically oriented trade shows around the world. And we are craving engagement with anyone who's looking to stand up a highly flexible, highly performant, highly efficient AI purpose-built infrastructure. We love what we do. Well, I can't wait to follow your story. You're on a great trajectory.

And I can't wait to learn more. we'd love to have you back on the show anytime you want to cologne. But thank you so much for being here today. It was my pleasure. And thank you so much for the questions. And we will talk soon. Thank you, Daniel. Thank you.

Thanks for joining Tech Arena. Subscribe and engage at our website, Techorina.ai. All content is copyright by Tech Arena.