Nebius Group N.V. Q1 FY2025 Earnings Call

CEO and members of the management team. Our remarks today will include forward-looking statements, which are based on assumptions as of today. Actual results may differ materially as a result of various factors, including those set forth in today's earnings press release and in our annual report on Form 20-F filed with the SEC. We undertake no obligation to update any forward-looking statement. During the call, we will present both GAAP and certain non-GAAP financial measures. A reconciliation of GAAP to non-GAAP measures is included in today's earnings press release. The earnings press release and accompanying shareholder letter and an investor presentation on our website at group.nebius.com/investor-hub. We ask that you enter your questions into the webcast portal, and we will be reviewing and consolidating the questions for Q&A. And now I'd like to turn the call over to Arkady and the team, who will go over a few slides that we've presented for this investor presentation.

Yes. Thanks. Thanks, Neil. Thank you, everyone, for joining our Q1 2025 results call. I will start by saying that demand for AI compute was very strong in the first quarter. Our results show it. Our revenue grew nearly 400% year-over-year, and our annualized run rate revenue grew nearly 700%. We saw great momentum in our core infrastructure business. We ended the quarter with a solid cash balance of $1.4 billion and we actually continue to invest in our infrastructure. To that point, we are rapidly building our capacity to serve customers around the world. This is a global race as you understand, and we are well-placed with our footprint in the U.S., Europe, and now in the Middle East. As you can see here on the slide, we added three new locations recently, and there is more to come. We're exploring new locations for capacity build-out, and we hope to share more news with you very soon. We also announced some new partnerships this quarter to strengthen our relationship with NVIDIA as well as Meta and Llama. And finally, we had a very productive quarter with respect to building out our technology stack, and we are getting industry recognition for our AI cloud offering. For example, SemiAnalysis ranked us in the GPU Cloud ClusterMAX rating system. Now I'll hand over to Andrey Korolenko to discuss some of the key products we launched in Q1.

Thank you, Arkady, and hello, everyone. We believe the depth and the quality of our care significantly differentiate us against the other Neoclouds. We made great progress in Q1 in further developing our AI cloud offering and had a number of notable product launches. We launched the Slurm-based cluster upgrades such as automatic recovery for failed nodes, proactive system health checks, and detecting issues before jobs actually fail. These changes reduce downtime for customers and improve capacity availability on our infrastructure, which led to around 5% improvement in available nodes for commercial use, which is quite significant. Several platform services were released and moved from the beta phase to general availability, MLflow, and JupyterLab notebook as examples, but there was much more. We invested a lot of time and effort in the reliability and performance of the platform. Notably, we launched an enhanced object storage, which ensures that large data sets can be accessed and saved quickly during model training runs, reducing client waiting times. Building on that foundation of our homegrown storage capabilities, we have also partnered with three leading storage providers such as DDN, VAST, and WEKA. This enables us to deliver the best possible experience for all customer scenarios going forward with the Blackwell generation clusters. Last, but not least, we expanded integrations with external AI platforms such as Metaflow, dstack, and SkyPilot, which allows customers to bring existing workloads into our ecosystem with minimal friction. Now I'll hand it over to Daniel.

Thanks, Andrey. In addition to strengthening our product in Q1, we also made significant progress toward expanding our partner ecosystem. From further building out our data storage solution portfolio, as Andrey mentioned, with industry leaders, we extended our core AI cloud capabilities to the ISV landscape with tight technical integration, and we made announcements enabling customers to consume Nebius infrastructure across a wide segment of the industry. Equally important are the relationships we have with the full range of AI marketplaces and established channel partners that help us meet customer demand for our AI infrastructure across the globe. I'd also like to talk about NVIDIA. As you know, NVIDIA is an investor in our company. We have a long history of working with the NVIDIA team, and we want to continue to build on that relationship. In Q1, we made several announcements with them. First, in the Q1 timeframe, Nebius and NVIDIA announced that Nebius would be one of the first AI clouds to offer the NVIDIA Blackwell Ultra AI Factory platform. We also became a launch partner for NVIDIA Dynamo, one of the most efficient solutions for scaling and computing during inference. Nebius was also named one of five reference platform NVIDIA Cloud Partners, helping us as we specialize in delivering AI-accelerated services built on NCP reference architectures. Finally, some breaking news: Nebius will support the NVIDIA DGX Cloud Lepton marketplace at launch. We couldn't be more excited about our partnerships, not just with NVIDIA, but across the landscape. So with that, I would like to hand it over to Roman.

Yes. Thank you, Daniel. Let's speak a little bit about customers. Our strategy is to serve a wide variety of customers with our robust platform. We have hundreds of managed and self-service customers using Nebius Cloud platform for training and inference workloads across various industries such as tech, media and entertainment, life sciences, and more. With our expanding capacity footprint and global sales support, we are now able to serve customers 24/7 with a truly tailored approach from our high-level experts on both sides of the Atlantic. This, together with the advanced software platform, goes beyond commoditized GPU as a service offerings. This highlights our flexibility and ability to rapidly adapt to the evolving needs of our diverse customer base while delivering high-quality solutions powered by our tech stack. This is what our customers value the most. They recognize that we are building an AI-specialized cloud with upscaling capabilities. All those factors contributed to our strong Q1 results. The demand environment for AI compute remains robust, and our sales momentum has continued into Q2. April's annualized run rate revenue was $310 million, and we continue to experience strong demand into May. Now I'll pass it to Tom Blackwell to walk through our guidance.

Yes. Thanks very much, Roman. As Roman said, we've had a great start to the year, a very strong first quarter, and we're carrying strong momentum into the second quarter. We feel very confident in our ability to achieve the ARR guidance for the whole year that we gave, which was $750 million to $1 billion. We're well on track to achieve this. We're also reiterating our overall revenue guidance for the group, which is in the range of $500 million to $700 million. As we think about profitability here, we're maintaining our adjusted EBITDA guidance for the full year. While we expect adjusted EBITDA to be negative for the full year, we plan to turn positive at some point in the second half of 2025. On CapEx, we're currently planning CapEx of approximately $2 billion for 2025. This is a bit up from the previous guidance of $1.5 billion due to a couple of factors. First, we had some CapEx spend that had been planned for late Q4, which actually fell in early Q1, which leads to the increase toward $2 billion. We want to be opportunistic when it comes to ramping up our infrastructure capacity as we see demand. We've considered additional investments beyond the initial data center expansion plan. You may have seen some coverage recently around the data center in Israel, which we think is a great opportunity. It's a great market, and I will give some more color around that later on in the call. Looking to the midterm, we believe that this business will achieve mid-single-digit billions of dollars in revenue, and we're actively building out our capacity pipeline to support that scale of revenue growth. The reality is that there are also scenarios where we could grow more aggressively. And Andrey and his team are very focused on building out the infrastructure potential pipeline to enable us to deliver potentially more than 1 gigawatt of capacity in Midtown. That would allow us to achieve significantly more revenue than the midterm guidance we’re talking about. We’ll be opportunistic and go after opportunities as we see them. Some factors could drive that additional incremental growth, such as more adoption from enterprise-level customers and larger, longer-term contracts. Again, Arkady will give a bit more color on that later in the stage. In terms of profitability, this is a business we can grow profitably, and we anticipate medium-term EBIT margins to range between 20% to 30%. This is supported by our AI cloud business reaching scale. We also have an important differentiator, which is the full stack, particularly the software at the top end of the stack. The software is a crucial part of our business model, making us attractive and sticky to clients. We believe it’s what’s going to allow us to create higher-margin business models and service a wider range of customers, allowing us to increase the revenue per GPU, not just through the GPU-as-a-service model, but with a broader range of revenue sources. It's also important to note that we take a very conservative view on depreciation. So with all of these numbers, we apply a full-year depreciation schedule while others in the industry typically use a 5- or 6-year depreciation schedule. Longer term, while we see 20% to 30% as our midterm EBITDA margins, we could exceed that with continued scaling. To summarize, we're successfully building AI infrastructure at scale. As you've heard Arkady talk about in previous calls, our differentiation comes down to two main factors: the quality of our technology and our access to capital that enables us to take advantage of that technology to ramp up quickly. We have a fantastic team of engineers building exceptional hardware, software, and services. Building out our native AI cloud and expanding the range of AI-native customers we're able to service well beyond a classic bare metal offering. We’re establishing strong partnerships, as Daniel discussed within the ecosystem, and all of this is enabling us to reach and serve a broader range of customers. We believe we have a unique position among Neoclouds to finance our future growth efficiently. We have significant capital funding potential for the core business from our various ownership and equity stakes in noncore businesses. Monetizing these equity stakes can effectively translate into the core business's bottom line. For instance, we have a 28% stake in ClickHouse, which according to recent reports has a valuation of around $6 billion. We believe it will continue to perform exceptionally well. We also have Toloka, which recently attracted strategic investors like Jeff Bezos and Mikhail Parakhin, which should help them scale up globally. This is great for Toloka, but also for us as we maintain a significant majority economic interest in Toloka. Similarly, Avride is one of the top autonomous vehicle teams globally, and we’ve announced partnerships with companies like Uber, Hyundai, Grubhub, and Rakuten, indicating the strength of our tech team. We’re in active discussions with potential third-party investors to help accelerate their growth while retaining a significant economic interest. Our ability to use these assets gives us a very attractive source of financing. We anticipate billions of dollars in investment towards the core business over time, with minimal dilution to existing shareholders and a disciplined approach to debt. Once we achieve adjusted EBITDA profitability, our strong balance sheet will allow revenue growth to translate efficiently into bottom-line results. I’ll stop there and hand it back to Neil for Q&A.

Great. Let's start with our first question. You just mentioned midterm revenue and margins. Can you clarify what you mean by midterm? What are the key components to achieve that? Roman?

Yes, thank you, Neil. Our base case plan calls for several billion dollars of revenue in the midterm over the next few years. While this assumes we grow our capacity to support this type of revenue from 2025 levels of 100 megawatts, our ambition is to provide much larger and faster growth. We are building a data center pipeline to provide scalability to more than 1 gigawatt of power. How quickly we get there will be a function of how fast we can capture demand through enterprise-level customers and longer-term contracts. Regarding margins, our target of 20% to 30% EBIT margins is driven by a greater mix of workloads where we can run our GPU fleet with high utilization for extended periods. In addition, we’ve put significant efforts into developing our software, which will contribute to high-margin software and services revenue over the long term. Furthermore, we take a conservative view on depreciation by utilizing a full-year depreciation schedule, while others usually apply a 5- or 6-year schedule. As we transition more towards inference, we're anticipating higher margins.

Great. Thanks, Roman. You mentioned that Q1 ARR was ahead of what you discussed on the last earnings call. What drove that strength? And how are you feeling about the full year?

Overall, the demand environment in Q1 was strong. Customers wanted access to GPUs, and we saw that demand strengthen month by month. Customers recognized the value of our infrastructure and software. We provided reliable and scalable service, enabling customers to access clusters with thousands of GPUs in a matter of days, not weeks. We received positive feedback from our core customers regarding this. Additionally, our sales team ramped up significantly, with investments in our presales and solution architects and customer success teams, which allowed us to provide 24/7 white-glove support. This has significantly contributed to improving our sales process and overall customer success. Our brand awareness is growing, and we have put significant efforts into that, aided by industry recognition like SemiAnalysis cluster ranks Gold status that Arkady mentioned, which helped deepen our sales pipeline. Additionally, our approach to bringing the newest chips online as promptly as possible—like not just responding to specific contracts but in a more cloud-like manner—has been paying off. A great example is the DeepSeek moment in February when we quickly responded to significant demand for NVIDIA H200 chips, which we had deployed in greater volumes than other players at that time. All these factors resulted in strong growth, and we reached a record high number of managed customers during Q1. Regarding the full year, we continue to see a solid start in Q2. Demand remains robust, with an annualized run rate revenue of $310 million in April, confirming our momentum continuing into May. In the second half of the year, we expect to bring Blackwells to customers, which should provide further support for our revenue profile and reinforces our confidence in delivering on our guidance of $750 million to $1 billion in annualized revenue by the end of Q4 2025.

So you discussed getting to positive adjusted EBITDA margins by the end of the year. Can you provide an update on when you think that will happen? Maybe we'll go to Tom for this.

Yes, sure. Achieving positive adjusted EBITDA is a crucial milestone for us. It highlights our focus on profitability. Specifically, we intend to reach positive adjusted EBITDA at some point during the second half of the year. I would note that breaking it down to our core infrastructure business, we expect to reach positive adjusted EBITDA probably sometime in the third quarter. Next, our goal is to achieve positive adjusted EBIT, and we are working diligently towards that objective.

Great. And Tom, maybe sticking with you, there's a question here about CapEx. We've raised the CapEx guidance. Can you provide updates on the reasons for this?

Our primary business model is predicated on building capacity for demand. Up to now, we’ve been fortunate to finance much of our CapEx with cash on hand. Looking at this year specifically, we had CapEx spend from late Q4 pushed into Q1 due to typical quarter-to-quarter fluctuations related to data center build-outs. We want to be opportunistic. The targets we've set out are base cases, but we see opportunities to do more aggressively if they present value to our shareholders. For example, when we identify an opportunity to ramp up capacity quickly based on existing demand we aim to pursue it. The Israel data center is one such opportunity that wasn't initially in our roadmap, but we believe it's a good one to pursue.

Great. Maybe keeping with the theme of CapEx, Tom, you touched a little bit on how we're going to finance our future growth. How do we expect to finance the CapEx expansion given that the cash balance now is below what we're planning to spend?

In Q1, we spent $544 million towards that overall $2 billion CapEx. At the end of the quarter, we had $1.44 billion of cash remaining on our balance sheet, so we feel good about funding that CapEx. More generally, the equity stakes we hold in our noncore businesses should serve as significant funding sources to continue ramping up and scaling beyond this year while minimizing dilution to shareholders and maintaining a disciplined approach to debt. Obviously, as a public company, we have more traditional funding sources, and we will consider those when they make sense and are value-accretive.

Great. We're receiving higher-level questions here about future growth. Arkady, can you share where you're seeing growth in this business?

The majority of our current customer base consists of new AI companies that have emerged in the last couple of years, which we call AI natives. They are very advanced in technology, smart, and quickly growing. We like them, and they like us. These companies, mostly venture-backed, are primarily in the U.S., which is why we’re focused on building our data center capacity in the U.S. right now. Most of our revenue currently comes from this market. The second promising sector is large AI labs. We haven't tapped this market yet, but we're preparing to serve them faster and better. To accommodate these customers, we will need larger data centers, and our pipeline aims to get us to more than 100 gigawatts of data center capacity. The third promising sector includes enterprise-level clients, which are not yet fully utilizing AI technology. This is where AI is expected to create the most value, and our full-stack solution is very relevant here. This market is global due to its nature, and our presence across Europe and globally will be in high demand. Finally, there’s potential in national AI projects where we plan to support and build AI factories worldwide. Overall, these four sectors represent the tip of the iceberg for AI business opportunities, and AI infrastructure demand is expected to surge globally. Nebius is poised to meet this demand.

All right. What impact does Toloka's deconsolidation have on your business? I can probably take this. Toloka is our AI data solutions provider. They've done a commendable job building their business with high-quality customers like Amazon, Anthropic, Microsoft, Poolside, Recraft, and Shopify. We believe Toloka has strong growth prospects, validated by their investments from Bezos Expeditions and Mikhail Parakhin, the CTO of Shopify. We're happy to retain a significant majority economic stake in Toloka. As our voting shares dropped below 50%, we'll be deconsolidating Toloka and updating our financials and guidance excluding Toloka in our Q2 earnings report. Andrey, can you provide an update on your capacity expansion plans for this year?

Sure, Neil. We are aggressively building and acquiring data center capacity. The New Jersey data center is a built-to-suit project designed according to our specifications, which is crucial for delivering power usage efficiency and cooling efficiency. We expect the first capacity in New Jersey to be operational in late summer, rolling out periodically in conjunction with demand. The Kansas City site is fully operational, having deployed our Hoppers generation for us, while the second part of Kansas City is currently deploying Blackwells, which will be available shortly. Iceland is fully operational, and our expansion in Finland is progressing well, with expectations for the first phase operational by late Q3 and the second phase closer to year-end. We expect to deploy over 100 megawatts of capacity this year.

Can you share more about the new site in Israel and discuss your expansion strategy beyond the EU and U.S.?

The Israel site is significant as it opens up another market for us. While we continue to focus on scaling our capacity in Europe and the U.S., we don’t want to be limited to just those markets. Israel has a vibrant AI market with numerous AI startups, enterprises, and global corporate R&D centers, providing us significant customer opportunities. Additionally, this is a step towards supporting national AI factories, and we hope to be involved in more initiatives across Europe, the Middle East, and the rest of the world.

Andrey, can you share an update on your GPU rollout plan for this year?

We are on track with our GPU rollout plans. We deployed Hoppers generation in Q1 and are currently rolling out Blackwell, which will be available on the platform shortly. We also started deploying the Grace Blackwell family, and we expect to begin rolling out the Blackwell Ultra generation in Q3, with the majority of the year's deployments focused on Blackwells.

Any thoughts on the impact of tariffs on our data center expansion plans? How are you thinking about the cost to our business?

While the situation around global tariffs is very dynamic, we currently do not believe that it will necessitate major changes to our expansion plans. We also believe we can navigate the current tariff environment without significant impacts on our costs.

Daniel, tell us more about Nebius' customers. Why are they choosing Nebius over other providers?

Our customers choose us for high-performance, resilient, and scalable alternatives to other cloud providers. However, what truly differentiates us is our deep expertise in hyperscale infrastructure and our hands-on approach alongside our customers. We are not just another platform vendor; we strive to drive greater returns for every AI dollar spent by our customers. For example, in Q1, we saw significant wins in industries like healthcare, life sciences, media and entertainment, and financial services. One of our customers, Captions, is a leading AI video platform that partnered with us to scale GPU training for their next-generation audio-to-video model, enabling them to accelerate their time to market. Similarly, Quantori, a top biopharma partner, has utilized Nebius to build a 3D molecular generation framework which has sped up their R&D and innovation. Regarding the future, we are committed to verticalizing AI solutions across various enterprises as they incorporate AI into their core operations.

Roman, can you give us an update on the types of contracts we're seeing in the market, in terms of structure and duration?

A significant benefit of coming to Nebius is our flexibility, which allows us to support and grow with native AI tech startups. Contract terms can span from several months to over a year. As we bring the Blackwells fleet to market, discussions are developing around longer-term contracts. We are eager to expand our offerings with the new generation of high-interest GB200s and GB300s, which will drive demand and increase structure diversification across the contracts.

Can you elaborate on our relationship with NVIDIA? How is that progressing?

We have a longstanding collaboration with NVIDIA, as they were part of our capital raise last December. This relationship is essential as we build our go-to-market strategy. In Q1, we became one of the first vendors to offer the Blackwell Ultra AI Factory platform, and our partnership allows us to utilize NVIDIA Dynamo, which is open-source for inference scaling. As we scale our cloud based on NVIDIA architecture, we can achieve a dollar-for-dollar performance and maintain a strong relationship as they roll out their offerings. This is crucial for our growth as more opportunities arise with the startup community and enterprise customers to monetize AI effectively.

Andrey, we've launched several products in Q1 on the software side. How does our software stack compare with our competitors, and what were the greatest launches?

Nebius has been designed to build a full-stack AI cloud from day one, focusing on software specifically optimized for AI workloads. Our stack has three layers. First is the hardware management layer that we designed to monitor and optimize performance. The second is our full cloud platform, providing flexibility and stability, similar to large hyperscalers. The third layer comprises pre-configured third-party AI tools simplifying the AI development process. In Q1, we launched about 50 products across our AI cloud and AI Studio. Notably, we launched the Slurm-based cluster upgrade, which includes automatic recovery, proactive system checks, and issue detection before actual jobs fail. These updates have significantly reduced customer downtime and improved recovery times. We’ve made substantial improvements to our object storage, enhancing read and write speed, ensuring data sets can be accessed quickly and supporting training runs with decreased waiting time. Our partnerships with leading storage companies are providing more flexibility to customers. Integrations with existing AI platforms such as Metaflow, dstack, and SkyPilot permit a smooth transition for customers onto our ecosystem.

Tom, how does our software stack drive revenue and margins?

Our software stack is a critical part of our overall offering. Many customers rely on our stack to manage and execute their workloads. We’re relatively unique within the Neocloud space due to our full-stack capability. This makes us particularly attractive and sticky with clients, who can start large GPU clusters efficiently. We’ve created tools to help clients manage their data models effectively. Revenue contributions from our software stack are currently modest but significantly enhance our overall revenue, and we plan to build on that further.

Concerning funding, we know we'll need it for this year and for future years. How are we approaching financing options?

As we think about funding growth, we want to do this in a way that minimizes shareholder dilution while remaining prudent with debt. We have a solid position to achieve this with our available cash and the potential to monetize equity stakes in our noncore businesses. We will consider more traditional opportunities in the capital markets when they are sensible and value-accretive. For now, we feel positive about our ability to fund growth with current and potential capital sources.

What exciting developments are ongoing with Avride? What strategic options are you exploring?

There are only a few independent autonomous vehicle platforms capable of competing effectively in the U.S. market. As you know, the market is enthusiastic about Waymo, while we view Avride's capabilities as competitive as well. Recent partnerships with Uber, Hyundai, and others underscore the potential for rapid growth. Recognizing the capital-intensive nature of such businesses, we are actively discussing potential collaborations with strategic partners to drive Avride's immense growth.

Can you clarify what you mean by midterm, Tom?

We believe this business can scale quickly, reaching mid-single-digit billions in revenue over a few years—our definition of midterm. However, we are committed to pursuing aggressive growth if achievable. Much of our existing revenue forecast revolves around our AI-native customer base, though as Arkady mentioned, potential growth sources include enterprise customers and large labs.

Thank you all for participating in our first quarter 2025 earnings call. We'll see you again on our Q2 call. Thanks.

Thank you.