Nebius at the UBS conference today

Investment Analysis Clubs Saul’s Investing Discussions
1

I’ve just read the script on Seeking Alpha and listened to the call with Nebius at the UBS conference today. Marc Boroditsky (CRO) and Neil Doshi (VP and head of IR) were the speakers.

Here are some of the main points from this 30 minute interview:

  • Nebius was founded by ex-Yandex leadership and started with a large, highly experienced engineering team with 1,000 engineers.
  • Nebius is positioning itself as a full hyperscaler rather than a simple GPU-rental provider.
  • End-to-end engineering spans data center construction, custom racks and servers, cooling systems, and power delivery.
  • There is a strong cultural focus on cost optimization, efficiency, and margin scalability.
  • Customer support is staffed by AI engineers, providing unusually deep technical assistance around the clock.
  • The platform includes a full cloud software layer: virtualization, managed Kubernetes, and a robust control plane.
  • Nebius aims to enable seamless training, optimization, and inference on a single vertically integrated stack.
  • Increasing investment in vertical-specific tooling (e.g., healthcare, life sciences, physical AI).
  • Strategy includes both proprietary capabilities and a broad integrated partner ecosystem.
  • No defined limit on how far up the software and tooling stack Nebius plans to expand.
  • Customer strategy targets foundation-model builders, AI-native startups, and traditional enterprises.
  • Management expects long-term demand to be dominated by enterprise adoption.
  • Early enterprise traction is appearing in areas such as finance and high-value vertical use cases.
  • Microsoft has signed a large, multiyear agreement for a massive Nebius cluster deployment. Microsoft can use Nebius capacity for Copilot growth, internal model training, and other compute-heavy initiatives.
  • Nebius reports a sharp acceleration in compute demand across all segments.
  • The company is seeing far more demand than supply, with pipeline growth significantly exceeding available capacity.
  • AI startups are exceptionally well-funded and consuming large volumes of compute early.
  • Many scaling AI companies are shifting from relying on foundation models to building their own specialized models.
  • Software vendors are redesigning products to become fully AI-first.
  • Inference workloads are expanding extremely rapidly and are outpacing industry-wide infrastructure build-out.
  • Nebius argues that the industry is not overbuilding; supply is still lagging true demand. “The demand is there. I mean we’re seeing it in our pipeline. We’re experiencing it with our customers. We’re watching customers doubling every 6 to 8 weeks in some cases. We’re watching the uptake take place.”
  • Infrastructure footprint expanding globally from a single initial site.
  • Key bottleneck is the physical complexity and speed of constructing data center capacity, not chips alone.
  • Nebius uses a parallelized, agile approach to building infrastructure rather than a standard cookie-cutter model.
  • The company has raised substantial capital and maintains a healthy balance sheet.
  • Additional liquidity can be unlocked from stakes in Avride, ClickHouse, and Toloka.
  • Future funding strategies include asset-backed financing, corporate debt, and opportunistic equity.
  • The core business has recently turned EBITDA profitable, with the full group nearing breakeven EBITDA.
  • Nebius targets strong medium-term EBIT margins driven by scale, premium cloud revenue, operating leverage, and in-house hardware design efficiencies.
  • Company maintains conservative financial practices, including stricter depreciation schedules than peers. (4 years rather than 6 at Coreweave).

For those of us who follow Nebius closely there was not much we didn’t already know - but it is always good for us to hear these things afresh. Perhaps the biggest takeaway for me was the fact that Nebius said that demand for compute is multiples higher than the capacity they can currently bring online, and that the market is still underbuilt, not overbuilt.

We are still in the early innings.

Jonathan

62 Likes
2

Not if you look at the percentage of revenue earned, however:

Nebius is providing over $17B worth of “dedicated capacity” to Microsoft and another $3B to Meta.

The guidance for 2025 revenue was between $0.5B and $0.55B, so what’s coming from Microsoft far outweighs everything else the company may do - at least for several years. In 2026 those Microsoft and Meta contracts start kicking in, and revenue will jump to between $7B and $9B.

So, being a “full hyperscaler” will amount to less than 10% of the company’s business.

Meanwhile, the company is spending billions in Capex to ramp up the contracted capacity. And, if succesful, will have just given major legs up to its legacy competition. That would be like the NetFlix of 2001 providing DVD delivery services to BlockBuster in exchange for money.

27 Likes
3

Nice summary. I think the most important take away from the conference was that Nebius reaffirmed their mid-term EBIT margin guidance of 20~30% and clarified that the Microsoft / Meta deals would be structured in that range as well. This, if achieved, will be a strong validation of their business model.

Cheers,

Luffy

18 Likes
4

Hello,

I use AI to help me better understand the companies I invest in. For Nebius, I’ve compiled a lot of data and combined my knowledge and understanding with AI to address the implication that hyperscaler revenue as a percentage of total revenue is too high.

Subject: Investment Research Update: Nebius Group (NBIS) – Analysis of Revenue Concentration
Date: December 6, 2025
Focus: Revenue Mix Dynamics (Hyperscaler vs. Organic)

  1. Executive Summary: The “Booster Rocket” Structural Thesis
    A prevailing concern in the market centers on Nebius’s revenue mix, with the Microsoft and Meta contracts currently representing an estimated ~85-90% of forward revenue. This concentration has led to the characterization of the company as effectively a captive entity of its hyperscaler partners.
    Strategic Perspective: Analysis suggests this concentration may function less as a dependency risk and more as a strategic financing mechanism. The “Booster Rocket” framework posits that the high-volume, lower-margin hyperscaler revenue serves to subsidize the massive CapEx required to build the infrastructure, which ultimately supports the “Orbiter”—the high-margin Organic Platform.
    Key Observation: The market currently applies a conglomerate discount for this concentration. However, the hyperscaler contracts effectively underwrite the hardware costs, allowing the organic platform to scale on infrastructure that is pre-financed by third-party commitments.
  2. Revenue Bifurcation: “Scale” vs. “Margin”
    To accurately assess valuation, the revenue stack requires bifurcation. The topline guidance ($7B–$9B ARR) comprises two distinct business units with divergent economic profiles.
    Projected 2026 Revenue Mix Analysis
    | Segment | Revenue Share (Est.) | Gross Margin Profile | Customer Concentration | Comparable Valuation Multiple |
    |—|—|—|—|—|
    | Hyperscaler (MSFT/Meta) | ~90% | ~35-40% | Extremely High | 3x - 5x (Utility / REIT) |
    | Organic (Token Factory) | ~10% | ~55-60% | Diversified | 12x - 15x (SaaS) |
    The Bear Case: The argument posits that Nebius functions primarily as a “landlord” for Microsoft, rendering the organic business negligible by comparison.
    The Counter-Thesis:
  • Absolute Scale of Organic Growth: Despite representing a small percentage of the total, the organic business ($551M ARR) is growing at 355% YoY. In absolute terms, this segment exceeds the revenue of many standalone AI platforms (e.g., comparable to or larger than Hugging Face’s estimated revenue). Its perceived smallness is an optical effect of the Microsoft contract’s magnitude.
  • Margin Impact: As the Token Factory scales, it exerts a disproportionate impact on profitability. Projections indicate that by 2027, the Organic segment—while potentially only ~20% of total revenue—could contribute 35-40% of Gross Profit due to superior software-like economics.
  1. The “Capex Shield”: Concentration as a Feature
    While high concentration typically signals risk (contract renewal uncertainty), in the context of AI infrastructure, it acts as a financing bridge.
  • Subsidized Capital Efficiency: Constructing a GPU cloud requires billions in upfront capital. Whereas competitors often rely on equity dilution, Nebius leverages the credit-grade Microsoft contract to secure debt financing for H100/Blackwell clusters.
  • Infrastructure Lifecycle: Should a hyperscaler contract wind down in the medium term (3-5 years), Nebius retains the amortized infrastructure. This dynamic would allow the redeployment of fully paid-off compute assets into the high-margin Organic business, theoretically lowering the cost basis significantly.
    Market Implication: The Microsoft revenue stream de-risks the hardware obsolescence cycle, providing a “Capex Shield” that protects the Organic business from the capital intensity faced by pure-play competitors.
  1. Structural Shift: The “Inference Flip”
    Critiques of the revenue mix often overlook the structural shift in AI workloads. Hyperscaler deals are largely predicated on Training capacity (cluster rentals), whereas the organic growth engine is driven by Inference (Token Factory).
  • Growth Velocity: Data indicates that Organic Inference volume (tokens generated) is growing approximately 4x faster than Hyperscaler capacity reservation.
  • Diversification Trajectory: In Q3 2025, the Token Factory added over 500 enterprise accounts. While individual account revenue is minimal compared to the Microsoft contract, the collective adoption rate points to a path toward diversification over the 2026–2028 horizon.
  1. Updated Valuation Framework
    Applying a blended multiple to the entire revenue base likely distorts the intrinsic value. A “Sum-of-the-Parts” (SOTP) methodology offers a clearer assessment:
  • Part A (Hyperscaler): Valued as a Data Center REIT (approx. 20x P/E) given the stable, contract-backed nature of the cash flow.
  • Part B (Organic): Valued as a High-Growth SaaS entity (10-15x Revenue) based on the $551M ARR and triple-digit growth rate.
    Conclusion: Even if the Hyperscaler revenue is discounted as “low quality” due to concentration, the Organic business embedded within the topline represents a significant, high-growth asset. The “concentration risk” is evident, but currently appears priced as a fundamental flaw rather than a strategic mechanism for asset financing.

Even if one was fortunate enough to buy shares at significantly lower prices, today’s price is very appealing.

I believe that the stock will significantly appreciate over the next year or two and have been buying more shares on recent weakness.

DJ

32 Likes
5

The “counter-thesis” only reinforces my bear case, at least in my mind. It’s not “an optical effect,” it’s the reality of where Nebius’ revenue comes from. And - this is the kicker - that such a small portion of revenue will contribute an out-sized portion of profitability is even worse! This shows that the deals with Microsoft and Meta are not very profitable, yet they are the more than lion’s share of revenue. Add in that these deals only serve to strengthen Nebius’ competition and I’ll continue to stay away.

The only counter to my bear take is that in a few years Nebius will have data centers stood up, paid mostly by these deals. However, these will be with out-dated chips that will suck up more power per compute than what Nvidia will be offering in those few yeas, and the data centers will also have power/cooling systems that need major updating to be used with those new chips.

18 Likes
6

Absolutely! In the end, whenever that is, I don’t expect Nebius to continuously provide great returns. However, I think we’re still pretty early in the hype-cycle, especially if they execute on their strategy of differentiating their services. I just hope I can hop off the locomotive at the right time.

DJ

13 Likes
7

Just to say that Bert has published another really bullish article on Nebius. I know that Saul was always a fan of Bert’s writings. It can be read on Seeking Alpha right now.

Bert’s final sentence is “I believe its valuation is quite modest for its highly visible opportunities and that is why I have made it my largest single holding in my high growth portfolio.”

Needless to say, I strongly agree.

Jonathan

27 Likes
8

For the new members here, “Bert” can be found as “Bert Hochfeld”. He has his own website and subscription service and frequently contributes to Seeking Alpha.

12 Likes
9

I’ll continue to play the role of Nebius bear, even (especially?) after reading Bert’s article:

To repeat, its gross margins, excluding depreciation are 71% and are on a rising trajectory so far this year.

• Bert is going to be surprised when Nebius’ gross margins decline significantly as the Microsoft and Meta deals ramps up and swamp Nebius’ other busines revenue..

I think the basic question to address is if it is reasonable to make a $5 billion/year of capital commitments in order to increase ARR by $8 billion in a single year with a gross margin in the low 70% range.

No, the basic question to address is whether it is reasonable to make a deal with one of your main competitors (Microsoft), that helps that competitor out while you take on the Capex, loans to support that Capex etc. In other words, while you have to contend with the non-high value business of supporting data centers, your competition gets to grab the high-value of providing AI services. Which, ironically, is what you want to do.

Bert quotes Nebius management saying that their “main focus is still to build our own core AI cloud business.” With their main revenue coming from bare metal deals, just exactly how will that work? They have to keep MS and Meta happy or else.

Bert cites 4 customers of Nebius services: Cursor, Black Forest Labs, Mistral AI, and Higgsfield AI. He claims Cursor “uses Nebius infrastructure.” That is probably true since Cursor uses a variety of AI cloud services:

“AWS is the backbone of Cursor’s infrastructure…Microsoft Azure hosts parts of Cursor’s secondary infrastructure…Google Cloud Platform (GCP) is used for a smaller set of backend systems…OpenAI provides Cursor with access to GPT-based models for various tasks…Anthropic supplies Claude models, which Cursor uses for general-purpose chat and code reasoning…Google Cloud Vertex AI is used to access Gemini models…xAI offers Grok models…”

But, what’s important to remember is that these customers of Nebius’ cloud services pale in size compared to the Microsoft+Meta juggernauts.

When investors think about why Nebius has an exceptional growth ramp, one part of the answer is that it has customers whose usage of Nebius infrastructure is being driven by their own hyper growth.

But, not the main part, which is, again, the Microsoft & Meta deals.

Bert then goes on to claim that the chips Nebius is purchasing today will “continue to be deployed and functioning for most customers.” I don’t know how he determined “most,” but the Microsofts and Metas and other cutting edge AI service providers will always want the latest and greatest. And as Nvidia continues to improve the compute per watt, older chips will actually be more expensive to run.

Let’s see what happens when the Microsoft and Meta businesses start scaling. If Nebius is able to continue to grow their profitability as Bert claims, I’ll have to change my tune.

41 Likes
10

Just to say thank you for doing this. You have been consistently bearish on Nebius as much as I have been consistently bullish. But I always appreciate your comments and they always make me examine my confidence in Nebius. I appreciate opposing views. It’s what makes Saul’s board so great.

Jonathan

38 Likes
11

Well, I appreciate your appreciation, but I’m curious as to why my arguments seem to have no effect, as NBIS remains your #1 holding.

What do you say to the overarching view that Nebius is essentially taking money today to help their legacy competition put them out of business long-term? It’s like a prize fighter taking a bribe to throw a fight, hoping he can use that money to train better and get another shot later.

What do you say to Nebius’ future profit outlook being quite different now that the MS and Meta deals will swamp all other potential revenue for years? Revenue going from $0.5B to $8B (+/-$1B) is almost entirely (90%) because of MS and Meta, not because of organic growth of their AI software infrastructure. Even at triple-digit growth, that organic business will remain small in effect on the company’s bottom line for years.

What do you say to the fact that Nebius can’t use the MS & Meta infrastructure for their own higher profit business until 2030/2032, by which time today’s chips won’t be worth the power to run them? The deal with Microsoft is for 5 years, with MS having the option to extend it to 7 years. The Meta deal is a 5 year term, too.

Bert keeps repeating 70% gross margin, but the company has only recently turned EBITDA profitable. Nevertheless, it seems very unlikely that the company can sustain EBIT margins when the gross margins for running a data center are half or less than for their AI software platform business. Even if the 90/10 split becomes 80/20 or 70/30 in a few years.

While Bert and others spend much time (too much in my opinion) on trying to counter the “AI is a bubble” thesis, note that my arguments don’t assume that at all - quite the opposite actually. NBIS is not a good investment not because AI is in a bubble, but because AI is now an arm’s race and NBIS has decided to make guns for their adversaries (90% of revenue), not use their guns for their own business (10% of revenue).

This will probably be my last post on NBIS - I feel like I’m mostly repeating my arguments and not seeing any counters. As always, time will tell.

23 Likes
12

I think these are legitimate concerns, but I don’t think they invalidate Nebius strategy. You ask why Nebius is still my number one holding - well it is important to say again that I first purchased it back in January at much lower levels than today. I made it my number 1 stock in the first part of this year because the more I discovered about the company the more I liked. It is as simple as that.

I really liked their CEO and their back story. I liked how it was operating like a start up and yet they had already built the biggest Cloud company in Russia (maybe Europe too). Everything I read at the time convinced me that this was a company that I really wanted as my top holding. I focussed on their ARR numbers and on their hyper growth and I enjoyed watching it grow from the low 30’s when I first bought it all the way up to 130+. And I’ve continued to hold it as it has dropped down to the 80’s where it is today.

After one of your posts last month I did actually decide to sell out of Nebius completely (as I stated in my last portfolio review). Your arguments got through to me - but then I really regretted it and so I bought back in again just 2 hours later. So that was just a blip for me.

I don’t pretend to understand everything about the company - and I am always learning - but I took further encouragement from the fact that Bert said yesterday he had also made it his top stock in his growth portfolio. I also took a lot of encouragement from the UBS conference I posted at the top of this thread.

In terms of the specific points you made in your post… You said that because of the MS/ Meta contracts that Nebius is technically barred from running its own platform till then - but I don’t think this is the case. What they’ve done is lock in long-term utilization with hyperscalers to de-risk capex during the most capital-intensive and uncertain phase of the AI build-out. Given how fast GPUs depreciate, idle capacity would be far more damaging than selling early-cycle capacity at hyperscaler pricing. In other words, they sold utilization risk, not strategic optionality.

In terms of chips being obsolete by the end of the contracts - but isn’t that true regardless of who the customer is. Monetizing it aggressively now, while scarcity pricing still exists, is rational. The alternative would have been holding capacity for internal demand that didn’t yet exist and absorbing depreciation. These contracts effectively fund the transition to next-gen builds (Blackwell/Rubin) rather than trapping Nebius in obsolete hardware.

It’s correct that infrastructure margins are much much lower than pure software. But Nebius isn’t a colocation business either—it’s a managed AI compute platform with a software layer, and EBITDA profitability has only just been reached. At this stage, margins are more about utilization and operating leverage than steady-state mix
The argument is not “Nebius will magically have software margins,” but rather “it’s too early to judge margin sustainability before utilization and operating leverage have played out.

I don’t think that selling compute to hyperscalers is arming enemies—it’s monetizing shovels at peak demand while building operational capability. Hyperscalers will not want to keep all AI capacity on their own balance sheets indefinitely, and so there is a place for the so called neo-clouds.

I continue to read as much as I can on Nebius and to follow them closely - and I will get out as fast as I can if the numbers do change (like Saul has taught us). But for the moment I still feel confident to stay in it as my number 1 holding.

Jonathan

23 Likes
13

No, that’s not what I said.
What I said was that Nebius can’t use the Microsoft-leased servers for their own use until the contract ends. Neither MS nor Meta are leasing server time, they’re leasing bare metal hardware.

And should the hardware not be obsolete in 2030, Microsoft, at their option, can extend the contract another 2 years. Nebius is taking on all the risk of depreciation.

You say: “these contracts effectively fund the transition to next-gen builds rather than trapping Nebius in obsolete hardware,” but that’s not really true as Nebius doesn’t get to use these builds until they’re obsolete. And with the scarcity of chips and power like you said, Nebius has to give priority to their 800lb revenue customers first. It’s now even harder for them to build out capacity for their own use.

Neo-Clouds were supposed to be providing high-value, AI-specific software infrastructure that the legacy hyperscalers hadn’t gotten around to yet because they were focused on their existing non-AI cloud service business. What you just tacitly admitted is that in reality the Neo-Clouds have mostly become data center builders and operators. So, I think they should be valued as such.

Your claim that Nebius is “a managed AI compute platform with a software layer” is not really true anymore - that’s only 10% of their revenue, and only 20% of their profits.

It’s totally arming their enemies. Microsoft gets to lock in AI cloud service customers today. Microsoft gets the first mover advantage. The shovels analogy only works when the vast majority of miners didn’t make money. That’s not the situation with AI - lots of companies are going to make money on AI, and lock-in to software platforms will remain a thing. Nvidia isn’t making shovels for Nebius to sell. Nvidia is minting new guaranteed gold-laden land grants. Nebius is now mostly just the hired help.

9 Likes
14

Thanks for clarifying — you’re right that Microsoft and Meta are leasing bare metal hardware, so Nebius can’t use those specific GPUs for its own platform until the contracts end. They are taking depreciation risk on that hardware. I understand your point.

Where we differ is in what that means strategically. Locking in long-term demand converts uncertain, rapidly depreciating GPUs into cash flow. Holding them for internal use before demand was proven would have been far riskier. Scarcity of chips and power affects everyone, so prioritizing large customers is a rational way to survive and fund future builds.

I agree that Nebius’ software/platform business is small today and that neo-clouds have become mostly infrastructure operators. That said, AI-optimized infrastructure is still different from generic data centers, and the software option hasn’t been abandoned.

Re your point on “arming competitors,” Microsoft gains lock-in, but Nebius likely never had a realistic path to compete at that level anyway. Monetizing compute at peak demand is a good way to survive, even if it limits short-term upside.

In short, the disagreement is about the alternative path: more risk for potential upside versus the route Nebius chose, trading some upside for financial stability and survival.

13 Likes
15

I still don’t understand this. They wouldn’t be buying all these chips/servers and setting up the data centers (or this large) without MS/Meta. They would just setup what infrastructure was necessary for their own cloud platform customers.

It is true that Nebius will make money on servicing MS and Meta, and they can use those profits to help fund expansion for their own customers, but due to the size of the MS/Meta deals, I think Nebius will have a harder time focusing.

This seems exactly backwards - since chips and power are scarce, why waste what they can actually get on lower margin businesses?

The whole raison d’être for the NeoClouds was that they could be more nimble and more focused on supporting AI workloads than the legacy cloud providers like Amazon, Microsoft, Google, and Oracle. If they were never going to be able to compete with them in the first place, why own them via stock? And if they’re now, for the next few to several years anyway, mostly going to be data center providers, what is the appropriate multiple for that kind of company?

6 Likes
16

In their recent call Nebius said the following…

As busy as we are with these mega deals, our main focus is still to build our own core AI cloud business. We made great progress here with AI native start-ups like Cursor, Black Forest Labs and others. The economics and the cash flow of mega deals are attractive in their own right, but they also enable us to build our core AI cloud business faster. This is our real future opportunity.

Yes, they are currently getting most of their revenue from MS and Meta. But this is providing them cash flow to build their core AI business.

10 Likes
17

Well, of course they’re going to say that. Whether the company can actually focus on the 10% of their business that brings in 20% of their revenue is something they’ll need to prove.

And yes, it is providing them with money to fund this, assuming their data center build out and operations (including electricity) are profitable.

At best, however, it’s still giving their competition a major leg up. Microsoft has locked in their costs (and capacity) for 5 years and can focus on their AI software stack instead.

8 Likes
18

Hi @Smorgasbord1

I decided to put your questions to Bert directly, and he was kind enough to reply at length. I asked him if I could share his response to you here and he also said I could, so I am reposting it here.

This is how I phrased your points.

1. How do you respond to the view that Nebius’ current strategy risks weakening its long-term competitive position by using today’s capital to build infrastructure primarily for Microsoft and Meta—effectively empowering legacy competitors rather than scaling its own higher-margin AI platform?

2. Given that roughly 90% of projected revenue growth appears tied to these two counterparties rather than organic expansion, how should investors think about the true earnings power of Nebius’ core software business over the next several years?

3. What is your assessment of the contractual constraints that prevent Nebius from repurposing this infrastructure for its own use until 2030–2032, especially in light of rapid chip obsolescence?

4. How realistic are management’s margin aspirations when data center operations typically generate materially lower gross and $EBIT margins than AI software, even if the revenue mix gradually shifts?

5. Finally, in an AI environment that resembles an arms race rather than a bubble, does Nebius’ choice to be a large-scale supplier to competitors represent a sound long-term strategy, or a structural disadvantage for shareholders?

This was Bert’s reply, which I’ve edited slightly….

to make clear-it is totally unfactual that 90% of Nebius growth is coming from the two hyperscaler contracts. A simple reading of the ARR guidance and the terms of those contacts will show you that.

Bert then quotes the same quote from Arkady on their recent call… the one I posted higher up this thread.

“As busy as we are with these mega deals, our main focus is still to build our own core AI cloud business. We made great progress here with AI native start-ups like Cursor, Black Forest Labs and others. The economics and the cash flow of mega deals are attractive in their own right, but they also enable us to build our core AI cloud business faster. This is our real future opportunity.”

Obviously, judging by the share price today and Friday this is not understood or believed. I strongly suggest that you take a look-at Cursor and Black Forest.

You can further go to the transcript of the Nebius CRO at the UBS conference a couple of weeks ago. The customer that is growing 40% in inference every two months is almost certainly Cursor.

I strongly suggest a quick glance at the UBS transcript. You have asked about margins. The AI natives pay based on usage. I an sure there are volume discounts. We all can remember how Twilio gave away its deal with Uber and Arkady certainly remembers. I imagine the incremental gross margins on usage above minimums is very high. The last thing I would worry about for this company is contribution margins from customers whose usage is growing 40% every two months.

The concept that Nebius is enabling Microsoft or Meta or whoever else they might sell to is an odd one for me. I think the strategy is crystal clear-in the dim past I used to be a VP of Plans and Strategies for some IT companies. I wish I could have thought of something as elegant as this.

Nebius wants to grow at an unprecedented cadence without overleveraging its balance sheet. The company has sold hyper-scalers $20 billion of compute-presumably at a reasonable, fully burdened margin. It is taking that revenue and using it as collateral for the capital it needs to build the data centers it is in process of building. The datacenters have the capacity for the $20 billion of hyperscaler revenue, but they also have the capacity for AI natives to run their apps. The Motley Fool individual has this not just wrong-but exactly wrong. It is Nebius who is using the hyperscalers to grab share with capacity and functionality from AI natives.

With the UBS transcript at hand, I could go further and comment about software and support and functionality-it really is all there, but I suppose maybe the ink is too cold after this latest artic blast and it may have disappeared. Really, the transcript is replete with all the answers one might reasonably expect.

36 Likes
19

To those curious what Cursor is, its one of the premier AI IDE services (Tool for writing code). They are competing directly with Microsoft Github Copilot. Cursor hit 1B in ARR within 24 months of launching their product. Lots of developers argue Cursor is better than Copilot. I think cursor has a slight edge at the moment but with the AI field who has the better product can shift multiple times in a month. Cursor is a private company and thus not previously discussed on this board.

Saying Nebius growth is being driven by Cursor growth is its own risk. Cursor has to be getting close to hitting its growth potential ceiling and it will be hard for it to continue to grow at those rates.

Drew

8 Likes
20

Well, you didn’t phrase my points correctly, as Bert pointed out:

Bert is right, and that’s not what I said. I said 90% of Nebius’s revenue (not growth) is coming from the MS+Meta deals. That remains a fact.

Yeah, like I said: “Well, of course they’re going to say that.” Whether the company can actually focus on the 10% of their revenue portion (potentially 20% of the profits) is something they’ll need to prove.

Again, you didn’t state the “helping competitors” point as clearly as I have. It’s very simple and inarguable: Microsoft is today offering its “Azure AI Foundry” as an AI cloud service, and it’s likely it will run, at least partially, on the Nebius servers being leased to Microsoft. Nebius’ own “Token Foundry” directly competes with that:

No, Bert is completely wrong here.

Again, your indirect wording of the issue may be at play here, but even so it appears Bert doesn’t understand the difference between leasing AI bare metal to Microsoft and providing value-added AI cloud services to end customers, and furthermore, appears to be misapplying the “AI native” terminology.

As an example, Samsung is both supplier to Apple and a competitor. Apple phones often have displays and memory chips from Samsung, yet Samsung also makes its own branded phones (thanks to Android). Samsung is a large, diversified company and is OK with this, but no-one goes about valuing Samsung solely as a smartphone vendor.

End users buying iPhones don’t know nor care that there are Samsung devices inside them - they’re buying the iPhone experience and eco-system. And as established and large as Samsung is, they’re happy to have that hardware supplier business. But, Nebius isn’t in that same position. They’re small and still trying to establish themselves. While Samsung regularly pulls in around 20% of the world’s smartphone business, I suspect Nebius’ portion of the AI Software Platform business is under 1%.

Customers using Microsoft’s AI services won’t know whether they’re being hosted on Nebius’ or CoreWeave’s or Microsoft’s own servers. Since, as you pointed out, chips and power are in limited supply now, Nebius is giving a major leg up to Microsoft by providing these to it. Apple can easily switch from Samsung displays to LG displays, and its customers will mostly never know.

With regards to “AI natives,” Bert has it exactly backwards. With the Nebius deal, Microsoft IS the AI native supplier with its Azure AI Cloud Services. Nebius is simply a supplier of Nvidia servers to help power it, and Microsoft can switch from Nebius to CoreWeave to its own servers without customer knowing nor caring.

The amount that this may happen is unclear, since those data centers haven’t been built (Bert talks like they’re built already) and Nebius hasn’t provided any details as to what additional capacity beyond what MS & Meta have contracted for. And, of course, building out additional capacity is an additional Capex expense, reducing the EBITDA value of those contracts. Granted, Nebius potentially gets some economies of scale for this additional build-out.

So sure, if you want to say that the MS/Meta deals with Nebius help them build out their own server capacity at a cheaper price than they would be able to otherwise, fine. But, that’s quite a bit different than saying the profits from those deals will be at the same (or greater!) gross margins than what Nebius experiences today, or that somehow these deals increase customer demand for Nebius’ separate AI Software Platform. They mostly enable Microsoft to offer its AI Software Platform more quickly and with a guaranteed cost.

17 Likes