Picture supply: The Motley Idiot.
DATE
Tuesday, Might 20, 2025 at 8 a.m. ET
CALL PARTICIPANTS
Chief Govt Officer — Arkady Volozh
Chief Product Officer — Andrey Korolenko
Chief Enterprise Improvement Officer — Daniel Bounds
Chief Operations Officer — Roman Chernin
Chief Monetary Officer — Tom Blackwell
Want a quote from one in all our analysts? E-mail [email protected]
TAKEAWAYS
Income Progress: Income elevated almost 400% year-over-year in Q1 2025. Annualized run charge income elevated almost 700% year-over-year.
Money Stability: Ended Q1 2025 with $1.44 billion in money.
Capability Enlargement: Added three new places; Israel was highlighted as a key new market, with Iceland and Finland additionally coming on-line or increasing.
Product Launches: Launched roughly 50 new merchandise within the AI cloud and AI Studio platforms, together with the Slurm-based cluster improve, enhanced object storage, and common availability of MLflow and JupyterLab pocket book providers.
Operational Effectivity: Achieved a 5% enchancment in out there compute nodes for industrial use by platform reliability upgrades in Q1 2025.
Partnerships: Introduced new and deepened partnerships with NVIDIA, together with launch companion standing for NVIDIA Blackwell Extremely AI Manufacturing unit, changing into one in all 5 NVIDIA Cloud Companions, and supporting the DGX Cloud Lepton market at launch.
Buyer Base: Grew to lots of of managed and self-service clients throughout sectors akin to expertise, media, leisure, and life sciences.
Quarterly CapEx: Spent $544 million in Q1 2025 towards an up to date full-year CapEx plan of roughly $2 billion. This compares to the prior $1.5 billion CapEx steering for 2025. The rise in CapEx for 2025 is attributed to shifted spending and opportunistic growth, together with the Israel information heart.
Income Steerage: Reiterated full-year group income steering of $500 million to $700 million for 2025. Annualized run charge income (“ARR”) steering is $750 million to $1 billion by year-end 2025.
Profitability Outlook: Expects detrimental adjusted EBITDA for the total yr however plans to attain constructive adjusted EBITDA in the course of the second half of 2025. The core infrastructure enterprise may doubtlessly attain this goal in Q3 2025.
Medium-Time period Outlook: Targets mid-single-digit billions of {dollars} in income and EBIT margins of 20%-30% within the midterm (inside a number of years), with potential upside if enterprise adoption accelerates.
Depreciation Coverage: Maintains a full-year depreciation schedule, extra conservative than the 5- or 6-year schedules typical within the trade.
Non-Core Asset Monetization: Holds a 28% minority stake in ClickHouse (reportedly valued at ~$6 billion), a big majority financial curiosity in Toloka, and a stake in Avride, all recognized as future capital sources for core AI infrastructure funding.
Toloka Deconsolidation: Will deconsolidate Toloka as voting management fell under 50%, with up to date financials and steering to be offered within the Q2 report.
Market Segments: Core buyer base stays AI-native startups; enterprise and frontier AI lab markets are focused for progress, with ambitions to enter world and nationwide AI challenge sectors.
Information Heart Roadmap: Deploying over 100 megawatts of capability in 2025, with a pipeline that might doubtlessly exceed 1 gigawatt within the midterm (outlined as a number of years).
GPU Rollout: Executing deployments of NVIDIA H200, Blackwell, and Grace Blackwell (GB200) GPU generations, with Blackwell Extremely slated for Q3.
Contract Construction: Contract durations vary from a number of months to over a yr, with the introduction of Blackwell and newer GPU generations is facilitating longer-term commitments, as mentioned in Q1 2025.
Tariff Threat: Administration acknowledged the present tariff surroundings isn’t anticipated to have important influence on prices or growth plans however acknowledged ongoing monitoring as a result of potential volatility.
SUMMARY
Nebius Group N.V. (NBIS) Income elevated almost 400% year-over-year in Q1 2025, and expanded its AI infrastructure capability with a number of new web site launches and aggressive product rollout within the first quarter. The corporate bolstered its monetary place and operational flexibility by elevated CapEx funding, with roughly $2 billion in CapEx deliberate for 2025, a large money steadiness, and strategic monetization plans for non-core property. Aggressive positioning was strengthened by landmark partnerships with NVIDIA and differentiated software program choices now integral to its full-stack AI cloud answer.
Chief Monetary Officer Tom Blackwell emphasised, “we anticipate adjusted EBITDA to be detrimental for the total yr, we plan to show constructive sooner or later within the second half of 2025.”
Current product advances included integration with exterior AI platforms akin to Metaflow, dstack, and SkyPilot, enabling clients to transition workloads with minimal friction.
Roman Chernin acknowledged, “April’s annualized run charge income was $310 million, and we’re persevering with to expertise sturdy demand into Might.”
One of many 5 NVIDIA Cloud Companions, Nebius receives trade validation through inclusion within the SemiAnalysis GPU Cloud ClusterMAX rating system.
The deconsolidation of Toloka, triggered by the lack of majority voting management, will lead to future Nebius financials excluding Toloka’s direct contribution after Might, with up to date financials and steering excluding Toloka to be offered within the Q2 2025 report, with new reporting to be issued within the subsequent quarter.
Methods for future capital deployment depend on retaining “important majority financial curiosity” in non-core holdings following fairness funding occasions, as specified in the course of the name.
INDUSTRY GLOSSARY
Slurm-based cluster: A sort of useful resource administration and scheduling system for high-performance computing workloads, usually used for managing massive GPU clusters for AI coaching.
GPU Cloud ClusterMAX: An trade score system evaluating the efficiency, scale, and availability of GPU-enabled cloud clusters for AI workloads.
AI Manufacturing unit: Refers to large-scale, purpose-built information heart infrastructure optimized for coaching and serving AI fashions, usually utilizing the newest GPU expertise.
Blackwell Extremely: NVIDIA’s subsequent era of high-performance GPUs, meant for large-scale AI mannequin coaching and inference in cloud environments.
DGX Cloud Lepton: An NVIDIA AI cloud infrastructure providing that permits speedy deployment of high-end GPU clusters for demanding AI purposes.
ARR (Annualized Run Fee Income): The income run charge calculated by annualizing the newest month’s income, used as a forward-looking indicator for subscription-based or usage-driven companies.
Neocloud: Business time period denoting new-generation, AI-centric cloud service suppliers with infrastructure, software program, and providers optimized particularly for large-scale AI operations.
Full Convention Name Transcript
Arkady Volozh: Sure. Thanks. Thanks, Neil. Thanks, everybody, for becoming a member of our Q1 2025 outcomes name. I’ll begin with saying that demand for AI compute was very sturdy in first quarter. And really, our outcomes present it. Our income grew almost 400% year-over-year, and our annualized run charge income grew almost 700%. We noticed nice momentum in our core infrastructure enterprise. We ended the quarter with a strong money steadiness of $1.4 billion and we truly proceed to spend money on our infrastructure. To that time, we’re quickly constructing our capability to serve clients world wide.
It is a world race as you perceive, and we’re nicely positioned with our footprint within the U.S., Europe and now within the Center East. As you possibly can see right here on the slide, we added three new places not too long ago, and there’s extra to return. We’re exploring new places for capability build-out, and we hope to share extra with information with you very quickly. We additionally introduced some new partnerships this quarter to construct our sturdy relationship with NVIDIA in addition to Meta and Llama. And eventually, we had a really productive quarter with respect to constructing out our expertise stack. And we’re getting trade recognition for our AI cloud providing.
Right here, for example, SemiAnalysis ranked us in GPU Cloud ClusterMAX score system. And now I will hand over to Andrey Korolenko to debate a number of the key merchandise we launched in Q1.
Andrey Korolenko: Thanks, Arkady, and whats up, everybody. We consider the debt and the standard of our care is considerably differentiate in opposition to the opposite Neoclouds. We made an awesome progress in Q1 in additional growing our AI cloud providing and had various notable product launches. We — to start with, we launched the Slurm-based cluster upgrades akin to computerized restoration for the failed nodes proactive system well being checks, digital points earlier than jobs truly fail. These modifications reduces downtime for purchasers. And improved capability availability on our infrastructure, which led to round 5% enchancment on out there nodes for the industrial use, which is sort of important.
A number of platform providers had been launched and moved from the beta part to the final availability, MLflow and JupyterLab pocket book as an examples, however there was way more. We additionally invested quite a lot of time and efforts in reliability and efficiency of the platform. Notably, we launched an enhanced object storage, and this ensures that giant information units might be assessed and saved shortly throughout mannequin coaching runs, lowering shopper outcomes. In constructing on that basis of our homegrown storage capabilities, we have now additionally partnered with three main storage suppliers akin to DDN, VAST and WEKA. And that permits us to ship the very best expertise for all buyer situations going ahead with the Blackwell era clusters.
Final, however not least, we expanded integrations with exterior AI platforms akin to Metaflow, dstack, SkyPilot and that enables clients to carry the present objectives into our ecosystem with minimal friction. And companions, Daniel will speak about our companions.
Daniel Bounds: Thanks, Andrey. Along with strengthening our product in Q1, we additionally made important progress in the direction of increasing our companion ecosystem. From additional constructing out our information storage answer portfolio, as Andrey talked about, with trade leaders, we prolonged our core AI cloud capabilities to the ISV panorama with tight technical integration and we made bulletins enabling clients to eat Nebius infrastructure throughout a large section of the trade. Equally necessary are the relationships we have now with the total vary of AI marketplaces, established channel companions that assist us meet the client demand for our AI infrastructure throughout the globe. I would additionally like to speak about NVIDIA. As you recognize, NVIDIA is an investor in our firm.
We’ve got a protracted historical past of working with the NVIDIA staff and we need to proceed to construct on that relationship. In Q1, we made a number of bulletins with them. Level one, within the Q1 time-frame, Nebius and NVIDIA introduced that Nebius can be one of many first AI clouds to supply the NVIDIA Blackwell Extremely AI Manufacturing unit platform. We additionally turned a launch companion for NVIDIA Dynamo, probably the most environment friendly options for scale and compute throughout inference. And Nebius was additionally named 1 of 5 reference platform NVIDIA Cloud Companions, on this time-frame, serving to us as we specialize and ship AI accelerated providers constructed on NCP reference architectures.
And eventually, some breaking information, Nebius will assist the NVIDIA DGX Cloud Lepton market at launch. We could not be extra excited with our partnerships, not simply with NVIDIA, however throughout the panorama. So with that, I wish to hand it over to Roman.
Roman Chernin: Sure. Thanks, Daniel. Let’s converse somewhat bit about clients. Our technique is to serve all kinds of consumers with our sturdy platform. We’ve got lots of of consumers each managed and self-service, who use Nebius Cloud platform for coaching and inference workloads throughout numerous industries akin to tech, media and leisure, life science and extra. With our increasing capability footprint and world gross sales assist, we at the moment are in a position to serve clients 24×7 with really tailor-made method of our high-level consultants on either side of Atlantic that along with the superior software program platform goes past commoditized GPU as a service choices.
This highlights our flexibility and skill to quickly adapt to the evolving wants of our numerous buyer base whereas delivering high-quality options powered by our tech stack. That is what our clients worth probably the most. They acknowledge that we’re constructing an AI specialised cloud with completely happy scalar degree of capabilities. Really, all these elements contributed to our sturdy Q1 outcomes. And going ahead, the demand surroundings for AI compute stays sturdy. And our gross sales momentum has continued into Q2. April’s annualized run charge income was $310 million, and we’re persevering with to expertise sturdy demand into Might. And now I will cross it to Tom Blackwell to stroll by our steering.
Tom Blackwell: Sure. Thanks very a lot, Roman. And in order Roman stated, we have had an awesome begin to the yr, a really, very sturdy first quarter. And we have carried in — we’re carrying in sturdy momentum into the second quarter. So we really feel very assured in our capacity to attain the ARR steering for the entire yr that we gave, which was $750 million to $1 billion. We’re nicely on observe to attain this. So we’re additionally reiterating our total income steering for the group, which is within the vary of $500 million to $700 million. So considering — we’re turning to profitability right here. So we’re sustaining our adjusted EBITDA steering for the total yr.
So simply to elaborate on {that a} bit, we anticipate — whereas we anticipate adjusted EBITDA to be detrimental for the total yr, we plan to show constructive sooner or later within the second half of 2025. On CapEx, we’re presently planning CapEx of roughly $2 billion for 2025. And it is a bit up from the earlier steering of $1.5 billion as a result of a few elements. First, we had some CapEx spend that had been deliberate for late This fall, which truly fell in early Q1. So a few of that — that might — results in the rise in the direction of $2 billion.
And in addition, as we have at all times stated that we need to be opportunistic in the case of actually ramping up our infrastructure capability as we see demand, and so we would like to have the ability to form of chase demand — and safe that demand nicely. And so we have had — we have thought of some further investments past the preliminary information heart growth plan, for instance, you could have seen some protection not too long ago across the information heart in Israel, which we predict is a superb alternative. It is an awesome market, and truly — will come on and provides some extra colour round that at afterward the decision. So seeking to the midterm, it is a nice enterprise.
It is in an awesome trade, and we predict the long run alternative is immense. Once we have a look at the midterm, we predict — we consider that this enterprise will obtain mid-single-digit billions of {dollars} in income and we’re actively constructing out our capability pipeline to assist that scale of income progress. The fact is that there are additionally situations the place we may develop extra aggressively. And so Andrey and his staff are very targeted on actually constructing out the entire infrastructure potential pipeline that might allow us to ship doubtlessly greater than 1 gigawatt of capability in Midtown.
So if we try this, that might clearly — that might permit us to attain considerably extra income than the form of midterm steering that we’re speaking about right here. So we’ll be opportunistic, and we’ll go after alternatives as we see them. Sure. I feel a number of the elements that might drive that further incremental progress on prime of the midterm steering as we see extra adoption from enterprise-level clients and likewise potential form of bigger, longer-term contracts. And once more, we’ll give — Arkady will give a bit extra colour on that at a later stage.
By way of profitability, it is a enterprise that we will develop profitably and we anticipate medium-term EBIT margins to vary within the form of 20% to 30% vary. So this shall be supported by our AI cloud enterprise reaching scale. We even have — we have now an necessary differentiator, which is the total stack and significantly the software program on the prime finish of the stack.
And the software program is — it is a vital a part of our enterprise mannequin, what makes us engaging to shoppers, sticky to shoppers, and in the end, we predict it is what is going on to permit us to attain greater margin, create higher-margin enterprise fashions and actually service clients in several methods and a wider vary of consumers that permit us to mainly get enhance that the impact of income per GPU. So not simply the GPU as a service mannequin, nevertheless it’s a broader vary of form of income sources. So we additionally — I feel it is also necessary to notice that we truly take a really conservative view on depreciation.
So truly, with all of those numbers, we apply right here a full yr depreciation schedule whereas others, I feel, use usually use extra of a 5- or a 6-year depreciation schedule inside our trade. Long run, I feel whereas we see 20% and 30% is the EBITDA margins within the midterm, long term, we may transcend that. I feel there’s various situations as we proceed to scale up and increase the enterprise the place we may go nicely north of 30% in the long term. So simply to wrap up, we’re constructing AI infrastructure efficiently and at scale.
I feel as you have heard Arkady speak about on earlier calls, basically, we predict our differentiation and what units us aside actually comes down to 2 issues: Above all, it’s the high quality of our expertise. There’s additionally our entry to capital that to permit us to make the most of that expertise and to ramp up and to scale up shortly. So briefly on the expertise. We’ve got an incredible staff of engineers. We’re constructing superb {hardware}, software program and providers. These engineers, they’re actually — they’re one of the best of one of the best within the trade. It could take years to construct a staff at that high quality, and we’re actually about to have them. They’re constructing nice tech.
We’re constructing out our native AI cloud, and we’re increasing the vary of AI native clients that we’re in a position to service — and actually, it is the AI cloud that we construct, it goes nicely past what you would possibly name a traditional naked metallic providing. We’re constructing out sturdy partnerships, as Daniel talked about throughout the ecosystem and all of that is permitting us to succeed in and repair a broader vary of consumers. By way of capital, so we predict that we’re truly in a really favorable place and truly fairly a singular place amongst Neoclouds to actually finance this future progress in an environment friendly method.
So we have now important capital funding potential for the core enterprise, which truly comes from our numerous possession and fairness stakes of noncore companies. And these — the monetization of those potential fairness stakes can actually translate effectively into backside line outcomes of the core enterprise. So simply to offer some examples of what we’re referring to right here. You will have seen ClickHouse within the information recently, we have now a 28% or roughly 28% minority stake within the enterprise, and this will doubtlessly be a vital supply of future capital.
So in line with a number of the latest press stories, there is a fundraising spherical underway in the intervening time, which might doubtlessly worth the enterprise at round $6 billion, and we consider that enterprise will proceed to carry out extraordinarily nicely and develop considerably from present ranges. We’ve got Toloka and we’re extraordinarily happy to announce that they arrive on the technique of strategic buyers from Jeff Bezos and Mikhail Parakhin coming into the construction. And we predict that their funding involvement within the enterprise is basically going to assist Toloka to scale up among the many prime tier of AI information firms globally, with nice backing from these buyers.
And however what’s necessary for us, we predict that is nice for Toloka, it is nice for us as nicely as a result of we and for our shareholders as a result of we preserve a big majority financial curiosity in Toloka. So we’ll profit from all of the upside. We even have Avride. It is among the best autonomous car groups on the planet. They’re doing nice this yr. Within the final quarter, we have introduced they’ve entered into partnerships with gamers like Uber, Hyundai, Grubhub, Rakuten these partnerships actually underscore, I feel, the energy of tech and the staff and locations them actually amongst a choose group of worldwide leaders in that area.
A quick observe on Avride, as we have talked about beforehand, we’re truly — we’re in pretty lively talks with potential third-party buyers and strategic buyers that might come into the enterprise that we consider would actually assist them to scale up even quicker and actually construct their companies. However once more, whereas we might at all times look to retain important financial curiosity within the upside. So it is actually our capacity to make use of these property in these states, which supplies us a extremely a really engaging supply of financing.
So once we take into consideration the way forward for billions and {dollars} of funding within the core enterprise, will be capable of very successfully monetize these companies and to develop extraordinarily effectively in a method that actually minimizes any dilution to current shareholders, whereas permitting us to remain very disciplined by way of debt. So simply — once more, simply to form of summarize, as soon as we obtain adjusted EBITDA profitability, our sturdy steadiness sheet and continued low curiosity burden, we consider will permit income progress to translate very effectively into backside line outcomes. So I will cease there, and Neil, I will hand again over to you for Q&A.
Neil Doshi: [Operator Instructions] Nice. Let’s begin with our first query. You simply guided to midterm income and margins. What do you imply by midterm? And what are the constructing blocks to get in there. Roman?
Roman Chernin: Sure, Thanks, Neil. Our base case plan requires a number of billion {dollars} of income within the midterm over the following few years. Whereas our base case assumes that we develop our capability to assist the sort of income progress from 2025 ranges of 100 megawatt. Our ambition is to develop a lot bigger and far quicker. For that, we’re constructing an information heart pipeline to supply scalability to greater than 1 gigawatt of energy. Additionally, as Tom stated earlier, how shortly we get there shall be a operate of how briskly we will scale and seize demand by extra enterprise-level clients and longer-term contracts.
And in addition a number of phrases in regards to the margins, our goal of [ 20% to 30% ] EBIT margins is a operate of two elements. Simply larger mixture of workloads the place we will run our GPU fleet with a excessive degree of utilization for an extended time frame. Second, I might say is software program. We put quite a lot of efforts into growing our software program, which permit us assumed contribution from high-margin software program and providers income over the long run. And value to say, as well as, we take a extra conservative view on depreciation the place we use a full yr depreciation schedule whereas others use 5 or 6 years.
So when extra shut shift to inference, this can come to greater margins for us as nicely.
Neil Doshi: Nice. Thanks, Roman. Roman, perhaps you would take the following query, too, which is round Q1 ARR was forward of what you mentioned on the final earnings name, what actually drove that energy? And the way are you feeling in regards to the full yr?
Roman Chernin: Sure. As I stated, total demand surroundings in Q1 was sturdy. Prospects who need to entry to GPUs. And we see that demand strengthen every month. Prospects, I consider, acknowledge the worth of our infrastructure and software program. We had been in a position to present dependable and scalable service. Our software program allows clients to start out accessing clusters with 1000’s of GPUs simply inside a matter of days and never weeks. And we heard from a few of our core clients’ recognition of that. We additionally noticed the advantages of our gross sales staff ramping up and particularly the investments in our presales and answer architects and buyer success staff. Now we will present 24/7 white glove assist.
And I consider it is like considerably contributed to enhance our gross sales course of and clearly, wholesale buyer success. Our model consciousness can also be rising. We put quite a lot of effort there. And in addition because of trade recognition. For instance, SemiAnalysis cluster remarks Gold standing that Arkady talked about contributed and we see that our pipeline comes extra deep and robust. Additionally, we see that our method to carry the most recent chips on-line as early as potential like not responding to particular contracts, however on this like extra cloud method. And our flexibility to supply the true cloud phrases, the mix of pay-as-you-go or reservations of various land is paying again as nicely.
A very good instance when — was the DeepSeek second in February, we may in a short time reply to the massive demand to NVIDIA H200 chips that we have now deployed in additional volumes that perhaps another gamers at that second. All that resulted in a powerful progress and we reached a file excessive variety of our managed clients throughout Q1. Few phrases about like the total yr, we’re seeing — we proceed to see a strong begin of Q2. The demand stays sturdy. April annualized run charge income of $310 million confirms that, and we’re seeing the sturdy momentum continued to Might.
And within the second half of the yr, we anticipate to carry Blackwells for purchasers which ought to present additional assist to our income profile and provides us confidence that we will ship on our steering of $750 million to $1 billion annualized income by the top of This fall 2025.
Neil Doshi: See — so that you mentioned attending to constructive adjusted EBITDA margins by the top of the yr. Are you able to present an replace if you suppose that may occur? Perhaps we’ll go to Tom for this.
Tom Blackwell: Sure, positive. So I suppose, I touched on this briefly within the presentation, however simply to choose up. So to start with, attaining constructive adjusted EBITDA is a crucial milestone for us, and it actually highlights that we’re very targeted on attending to profitability. And as we set out in a number of the midterm targets, we consider it is a enterprise that may publish actually sturdy profitability going ahead. So particularly, once more, with respect to adjusted EBITDA, we intend to succeed in constructive territory sooner or later in the course of the second half of the yr.
One factor I might observe is definitely if we break it down and have a look at the core infrastructure enterprise, then we’ll get — we’ll transfer in quicker there and we’ll get to constructive adjusted EBITDA in all probability someday within the third quarter. The subsequent objective will clearly be to then deal with reaching constructive adjusted EBIT, and we’re working full steam in the direction of that objective.
Neil Doshi: Nice. And Tom, perhaps sticking with you, there is a query right here about CapEx. We have raised the CapEx steering. Can we offer any replace on the explanations for this.
Tom Blackwell: Sure. So I imply — so look, our major enterprise mannequin relies on constructing capability for demand. And we have been very lucky to have the ability to finance quite a lot of our CapEx with our money available up till now. So this yr, so to start with, by way of the form of the particular steering for this yr. As I discussed earlier, we had some CapEx spend that had been keyed up for the top of the fourth quarter final yr, which acquired pushed into the primary quarter, there was — that is simply all the way down to form of typical quarter-to-quarter fluctuations based mostly on form of numerous elements associated to information heart build-outs.
However we — once more, we need to be opportunistic. We view the targets that we have set out as base circumstances, however there are quite a lot of situations the place we will do extra and go extra aggressively. And the place we see a possibility to take action, that is in a method that is worth accretive to our shareholders, we would like to have the ability to achieve this. So for instance, once more, once we see a possibility to ramp up capability quicker round current demand that we will see. We would like to have the ability to achieve this.
So once more, the Israel information heart is one which we have not initially had in our street map, nevertheless it was a possibility that got here alongside and we thought that was an excellent one for us to go for. And so we’re very happy. And I feel in all probability at a late stage, I will ask maybe Roman to offer a bit extra colour round that. But it surely’s an awesome market, and we’re very excited to be moving into that market. So — and that form of — it is a new geography on prime of a number of the earlier geographies that we have been targeted on.
So — however by way of these form of incremental information heart build-outs like Israel, from a income standpoint, we’ll be investing in placing the capability in place later this yr. And the income shall be — we’ll see extra contributing to 2026, holding us very a lot on that path in the direction of the form of mid-single-digit billions of income that we spoke about earlier within the presentation.
Neil Doshi: Nice. Perhaps holding with the theme of CapEx, Tom, I do know you touched somewhat bit within the slides on how we will finance our future progress. So how will we anticipate to finance the CapEx growth, provided that the money steadiness now could be under what we’re planning to spend.
Tom Blackwell: Sure, positive. So — So simply to form of recap. So in actual fact, if we have a look at Q1, so we have already spent $544 million within the first quarter in the direction of that total $2 billion CapEx, so discuss you about. And — on the finish of the quarter, we had $1.44 billion of money remaining on the steadiness sheet. So we be ok with our capacity to finance that CapEx. And in addition, once more, I might simply come again so far.
So form of going past that and searching additional afield, the fairness stakes that we have now in these noncore companies once more, we consider will present us a really important funding sources for future in opposition to it, we will proceed to ramp up and scale past this yr in ways in which actually reduce once more dilution to shareholders and permit us to remain very disciplined at me I feel it is a actually necessary level. Clearly, as a public firm, we have now entry to extra conventional funding sources, and we are going to have a look at these every so often once we consider they make sense and worth accretive.
And I suppose, simply one other level that I might make is that, once more, provided that we have now — proper now, we’re very fortunate — there was no debt, and we anticipate persevering with to have comparatively low ranges of debt. In order that implies that we’re going to have the ability to reinvest a big quantity of our income again into driving worth creation in our core AI infrastructure enterprise.
Neil Doshi: Nice. Appears to be like like we’re getting type of a better degree of query right here is simply round future progress. So perhaps Arkady, perhaps you possibly can inform us the place you are seeing the expansion sooner or later on this enterprise?
Arkady Volozh: Effectively, our present buyer base, the vast majority of them. They’re all these new AI firms that emerged within the latest couple of years and truly, they’ll proceed coming to the market each month. They’re very superior within the expertise — we name them AI native. They’re sensible. They’re very quick rising, truly we’re like them and so they like us. These firms are often venture-backed and understandably, the vast majority of them are within the U.S. That is why we’re so targeted on constructing our information heart capability within the U.S. proper now.
All the expansion you presently see in Nebius all this quarter outcomes, yr outcomes, most of our revenues, they arrive principally from this market now. The second very promising sector, which, by the way in which, isn’t in our income but is all these observe frontier AI labs, massive clients. We’ve not tapped this market but, however we’re doing quite a bit to be able to serve them and assist them to develop quicker. In an effort to serve this — these clients we are going to want way more and far greater information facilities. And we are literally preparing for this. Because of this we have now our a pipeline to get to this greater than 100 gigawatts of knowledge heart capability.
So we’re not there but, however we shall be there quickly. That is the second sector. The third sector, perhaps probably the most promising sector by way of progress is the enterprises. AI expertise right now has reached only a small fraction of the company shoppers. Everyone talks about it. However on the identical time, that is the place the world expects the vast majority of added worth, which AI shall be creating. And by the way in which, our full stack answer and higher-level providers, which we offer could be very a lot related precisely right here. This market is way more world by nature as a result of the true trade actually enterprises are in all places on the planet, in lots of international locations.
And that is the place our European and world infrastructure presence shall be in excessive demand, I feel. And by the way in which, I might say of simply a number of hyperscalers, we’re one in all only a few AI cloud suppliers that really can serve company shoppers in a number of geographies. So that is probably the most promising sector for us, I consider, within the close to future. And there’s additionally the fourth sector, which we’re additionally watching fastidiously its potential market in nationwide AI tasks. We hear an increasing number of about them. And right here, once more, we see an enormous alternative for us, and we plan to construct our AI factories in several international locations and geographies in U.S., in Europe and Center East and truly elsewhere.
And — however all in all, these 4 sectors and the entire market is only the start for AI expertise or AI enterprise. AI infrastructure shall be in excessive demand in lots of indices and in lots of geographies and Nebius shall be there to serve this demand.
Neil Doshi: All proper. Let’s have a look at. Subsequent query, how does Toloka deconsolidation influence your online business? I can in all probability take this. So Toloka is our AI information options supplier. They’ve performed a extremely good job by way of constructing their enterprise. They’ve high-quality clients like Amazon, Anthropic, Microsoft, Poolside, Recraft and Shopify and we consider Toloka actually has nice progress prospects. And that is actually validated by their investments from the Bezos Expeditions and Mikhail Parakhin, the CTO of Shopify. Given these progress prospects, we’re completely happy to retain a big majority financial stake in Toloka. And as we — now our voting shares dropped under of fifty%. We shall be deconsolidating Toloka.
For the reason that transaction closed in Might, we shall be updating our financials and steering ex Toloka in our Q2 earnings report. All proper. Appears to be like like we have now a number of questions round infrastructure. Perhaps we will begin with Andrey, on this one. Andrey, are you able to present us an replace in your capability growth plans for this yr?
Andrey Korolenko: Positive, Neil. We’re aggressively constructing and buying typically our information heart capability and increasing the footprint, what we introduced in February was the New Jersey information heart, and it is a build-to-suit challenge constructed by companion in line with our specs and design and that is fairly necessary due to — it helps us to ship the effectivity of the ability utilization and the cooling effectivity, once more, what we demand from ourselves. We anticipate that the primary capability in New Jersey to be operational in late summer time, after which it is going to proceed to roll out on periodical foundation together with the demand. We additionally introduced the Kansas Metropolis.
And the primary a part of it’s already absolutely operational, and that was first — was the final deployment of the Hopper GPU era for us. And in the intervening time, there are Blackwells deployed within the second a part of the Kansas Metropolis, and they are going to be out there on the platform a bit later in second quarter. We additionally introduced and truly launched Iceland, which is absolutely operational in the intervening time and our build-out in Finland goes fairly nicely. And precisely on observe, and we anticipate the primary part of the growth shall be operational in late in Q3, and the second part shall be — will observe nearer to the year-end.
And anticipate that operationally, we can have over 100 megawatts of capability deployment this yr.
Neil Doshi: Nice. Thanks, Andrey. So are you able to share extra in regards to the new web site in Israel and may you talk about your growth technique past the EU and U.S. Tom, I do know you form of alluded to this, perhaps Roman, you possibly can discuss somewhat bit extra about this and elaborate on Israel.
Andrey Korolenko: Sure. Thanks, Neil. So to start with, one factor in Israel is means for us that we’ll opened in yet another market. We stated like we had been very a lot and proceed to be targeted to scale our capability in Europe and U.S., however we do not need to be restricted solely by these markets. So to start with, it is a new marketplace for Nebius. Israel has nice AI market with quite a lot of go AI native start-ups, enterprises and R&D facilities of the worldwide company. It is a nice gate for us to quite a lot of clients. However what’s additionally necessary that that is our first however in all probability not the final step in supporting nationwide AI factories.
And we hope to assist and construct extra nationwide AI factories world wide. And we’ll look to see how we will plug into these initiatives throughout Europe, Center East and the remainder of the world. So we open like opportunistic and seeking to this market as a carry talked about.
Neil Doshi: All proper. Are you able to share an replace in your GPU rollout plan for this yr, Andrey?
Andrey Korolenko: So except for the Israel capability, we’re very a lot on observe with the rollout that we deliberate earlier. We — this yr, we deployed — in Q1, particularly, we deployed the Hoppers era H200 particularly. For the time being, we’re rolling out the Blackwell, as I already talked about and they are going to be out there within the platform shortly. And we additionally begin to deploy the Grace Blackwell household. So the GB200 household. We anticipate that in Q3, the Blackwell Extremely era will begin with the primary deployment. And the vast majority of this yr, we shall be truly deploying the Blackwells.
Neil Doshi: Nice. Thanks, Andrey. Andrey, perhaps sticking with you once more, and we have now a query round regulatory points round tariffs — any ideas on form of the influence of tariffs on our information heart growth plans? And in addition simply how are you serious about the associated fee to our enterprise?
Andrey Korolenko: Good query. Sure, there’s undoubtedly was in these sure — I imply, it is not clear round world tariffs however based mostly on the place we stand now, we do not consider that the present standing would lead to main modifications to our growth plans. We additionally consider that we will navigate by the present surroundings of tariffs with out important influence to our prices. I might, nonetheless, it is only a very dynamic state of affairs and issues can change fairly shortly as we already noticed in the course of the Q1. And we’re actively monitoring the state of affairs.
Neil Doshi: Thanks Andrey. So it appears to be like like we have now some questions on clients. So perhaps I will give this to Daniel. Daniel inform us extra about Nebius’ clients? And why are they selecting Nebius over different suppliers?
Daniel Bounds: Nice. Thanks, Neil, and thanks for the query. To start with, our clients select us as a result of we provide high-performance, resilient and scalable different for different cloud suppliers, however what actually makes a distinction. Our differentiation lies in our deep experience and hyperscale infrastructure and our function as a hands-on practitioner together with our clients. So we’re not simply one other platform vendor. What this does is in the end allow us to drive a larger return for each AI greenback our clients spend. Some examples of that, in Q1, we noticed nice momentum and new wins in vertical industries like well being care and life sciences, like media and leisure and monetary providers. One buyer of ours Captions is a number one AI video platform.
They partnered with us to scale GPU coaching for the next-generation audio to video mannequin, Mirage. And so by leveraging our infrastructure, they accelerated their time to market. They empower their creators to ship emotionally compelling and story-driven content material and in the end push the boundaries of AI-powered storytelling. So an awesome instance within the media and leisure trade. One other instance can be Quantori. They are a prime biopharma companion of ours. They use Nebius to construct a framework for 3D molecular era. And in the end, by rising the quantity of molecules that they may mannequin, they achieved chemically legitimate buildings and actually enabled quicker, scalable R&D and speed up the innovation that they’ve in drug and supplies discovery.
So actually monetizing and unlocking the ability for AI for these clients. And that is only the start. So wanting forward, we’re doubling down on the verticalization of AI options throughout the enterprise. Prospects starting from retail to robotics as they embed AI deeper into their core operations, we need to be proper there with them to drive measurable outcomes.
Neil Doshi: Nice. Thanks, Daniel. We’re getting a number of questions round contracts. So perhaps, Roman, are you able to inform us somewhat bit extra or give us an replace on what kind of contracts we’re seeing available in the market, perhaps by way of construction and period?
Roman Chernin: Sure. Thanks, Neil. The very first thing I need to spotlight that the advantages of coming to Nebius is our flexibility that permit us to assist and develop with native AI tech startups and meet their wants and suppleness. Contract plans are likely to go from a number of months right into a yr and past. As well as, as we’re simply beginning to carry the fleet of Blackwells that opening up extra discussions about longer-term contracts, new era, excessive curiosity and GB200s and GB300s, we increase this to drive extra demand and provides us flexibility on the kinds of the contracts, we will safe.
Neil Doshi: All proper. Shortly on NVIDIA. Are you able to discuss somewhat bit extra in regards to the NVIDIA relationship? How is that progressing? Daniel, I do know you shared some ideas in your slide, however something extra you need to elaborate there?
Daniel Bounds: Sure. I feel between Andrey and I, we have lined quite a bit. I will do some little bit of reiterating right here simply in case anyone missed a number of particulars. Clearly, we have now not only a tight collaboration however a long-standing collaboration with NVIDIA, they’ve been an investor and a capital elevate final December with us. And we have now a really sturdy go-to-market that we have constructed with them. In Q1, particularly, like I discussed earlier than, throughout the Blackwell household, however significantly as we introduced the Blackwell Extremely AI manufacturing unit platform, we’re going to be one of many first distributors to face on the GB300, NVL72-powered situations.
And we predict that is going to be an actual sport changer available in the market, and we’re proper there with NVIDIA as these roll out. We additionally talked in regards to the ecosystem and NVIDIA Dynamo. This open supply inference framework. And in order we proceed to roll out the size and the number of the AI factories which might be wanted available in the market we’re proper there with them in actual time.
After which the opposite factor that I discussed earlier, nevertheless it’s nonetheless necessary is — the flexibility to face up a cloud based mostly off of NVIDIA structure that really performs to spec and delivers a minimum of $1 for each greenback invested, if no more, is what the NVIDIA Cloud Accomplice program is all about and what the reference architectures that we’re 1 of 5 companions actually delivers for purchasers. It is a clear validation of our technical management and that simply rolls over into {the marketplace} that they are standing up with DGX Cloud Lepton and plenty of different alternatives, whether or not it’s with the startup neighborhood or increasing out and serving to enterprises monetize AI.
We have been very a lot in lockstep with NVIDIA for a really very long time and look to have a really brilliant future.
Neil Doshi: Nice. We’re undoubtedly getting some questions round our software program stack. And we additionally get these questions fairly a bit when talking with shareholders and buyers. So it looks like we have launched quite a lot of merchandise in Q1 on the software program facet? And the way does our software program stack examine with our opponents? And what actually had been form of the largest launches, Andrey, perhaps you need to take this.
Andrey Korolenko: Sure, Neil, we began Nebius with a transparent objective to construct a full-stack AI cloud. Meaning from the day 1, our focus has been to create a software program stack that’s particularly constructed for the AI workloads. And our stack is mainly three layers. And the primary one is the layer that manages our {hardware}. And since we design our {hardware}, we can also provide the instruments to watch its efficiency and optimize the utilization. The second layer is that we construct a full cloud platform. It is fairly much like the massive hyperscalers. It is a virtualized surroundings, so clients get extra flexibility and higher stability total.
And the third layer, there’s an software the place you possibly can ship the pre-configured third-party AI instruments and that simplify all the the AI growth course of. So we shipped numerous merchandise in Q1. I feel it is round 50 merchandise throughout the AI cloud and the AI Studio. Notably as I used to be saying within the — on the opening, we launched the Slurm-based cluster improve and akin to computerized restoration, proactive system checks, issued detection earlier than the precise jobs fail. And these modifications considerably scale back the downtime for the purchasers and improves the time to get well on our facet. We made quite a lot of efforts and made quite a lot of enhancements on the — our object storage.
Now simply we enhance the pace of learn and write for compute node that once more ensures that information units might be assessed and say shortly, shortly sufficient for the coaching runs, and that improves the time to outcomes throughout coaching. Additionally the partnership with the main storage firms helps us to supply extra flexibility to our clients. And as I stated earlier as nicely, the integrations, we consider that it is crucial to combine with the present AI platforms akin to Metaflow, dstack and SkyPilot, and that simply permits the purchasers to carry their jobs and their instruments, current instruments on us with minimal friction.
Neil Doshi: Nice. Thanks, Andrey. What’s it — so by way of monetary efficiency, tying software program to our — again to our finance, how does that drive income and margins. Tom, perhaps you possibly can take a stab at this query.
Tom Blackwell: Sure, positive. I imply — so I feel, look, it is necessary to grasp that our software program stack, it is a important a part of the providing, proper? So a lot of our clients depend on the stack to assist them handle and execute their workloads. And once more, we’re comparatively distinctive in having that full stack providing inside — among the many Neocloud in our house. Makes — it additionally simply makes us very sticky with clients. So a number of the issues that, for instance, I feel it permits us to do — it permits us to provision massive clusters of GPUs shortly, so clients can begin their jobs with out ready.
We have created numerous instruments to assist them handle their information fashions, and observe their progress. And so I imply, once we take into consideration income contribution, I imply, I suppose the income contribution that you would get away as of right now, it is comparatively small, nevertheless it’s actually very a lot an added worth a part of the providing. And once more, it simply drives total clients coming to us and drive the general income and so it is about of the broader providing. And we will be very targeted on constructing it out, constructing our use circumstances and persevering with to make the merchandise extra sticky with clients form of going ahead.
So I feel over time, it may be — in all probability it could change into even probably the most important standalone driver of higher-margin income, nevertheless it’s actually about serving to us entry a wider vary of consumers provide higher-margin providers, higher-margin merchandise, holding them on the platform. And so it is an necessary a part of our total income progress form of going into the midterm.
Neil Doshi: Nice. And coming again to the funding query, simply perhaps somewhat bit extra pointed right here. So the query is, you will want funding for this yr, but in addition for the approaching years? And the way are you serious about financing choices, I do know, Tom, you talked somewhat bit about that on the slides and form of within the CapEx query, however — anything you need to form of reiterate on this level.
Tom Blackwell: I will simply truly form of reiterate fairly briefly on this. Once more, it is necessary for us as we take into consideration funding the expansion, we need to do that in a method that minimizes shareholder dilution and permits us to be prudent by way of debt. We’re in an awesome place to try this with the money we have now on the steadiness sheet and with the potential to monetize these numerous stakes within the noncore companies. Effectively, after all, in the end, we’ll have a look at — we’ll be contemplating different extra classical alternatives to the capital markets, and we’ll replace us on when we have now extra.
However once more, we really feel excellent about our capacity to proceed to fund this progress based mostly on the out there sources of capital that we have now to us.
Neil Doshi: Nice. There is a query on the opposite enterprise, particularly Avride. You stated chances are you’ll discover strategic choices. Are you able to perhaps share somewhat bit extra about why you are enthusiastic about Avride and perhaps what these choices may doubtlessly be, Arkady?
Arkady Volozh: Effectively, sure, we lined it a number of instances. There’s just some impartial autonomous car platforms that may compete in on the U.S. market right now. Positively, the market is worked up about what Waymo has achieved, sees it. And one in all only a few gamers that may truly construct a platform corresponding to that’s like we did not influence. And as you possibly can see, different market gamers, massive market gamers truly acknowledge this, have a look at the latest announcement of Avride partnerships with Uber, with Hyundai, with different massive gamers. And as we stated, they should develop. They should develop a lot quicker. It is yet one more capital intense companies in our portfolio, and we’re in lively discussions.
We will verify that we’re in lively discussions with potential strategic companions. And who can truly actually assist to drive progress to this bold challenge.
Neil Doshi: Perhaps one final query, a clarifying query. Are you able to clarify precisely what you imply by midterm? Tom, perhaps you need to take this one?
Tom Blackwell: Sure, positive. So once more, simply to recap on that. In order we glance into the midterm, we actually consider this enterprise can scale up shortly and obtain form of single — mid-single-digit billions of {dollars} of revenues. So what we imply by midterm, I imply, with successfully a number of years. And — however on the identical time, as we have tried to form of define, we are going to — we’re working very arduous to go as aggressively as we will and we’ll get there — to get there as quickly as potential. And we have now I feel form of form of body how we’re serious about the long run progress.
And so once more, quite a lot of our current income form of forecast stems round this type of AI native buyer base. However I feel Arkady set out the opposite form of incremental sources of progress across the enterprise clients, the massive form of the massive labs and so forth and so forth. So in our thoughts, we predict midterm is a number of years and we’ll go as shortly as we will.
Neil Doshi: Nice. Thanks, everybody, for collaborating on our first quarter 2025 earnings name, and we are going to see you once more on our Q2 name. Thanks.
Arkady Volozh: Thanks.
