Nebius Group (NBIS): Expectations for Q2 Results

Next Wednesday, NBIS will report its Q2 2026 results.

In this article, I’ll share my expectations and the key things I’ll be watching closely.

Let’s dive in.

As usual, let’s start with the basics.

Analyst estimates:

Q2 Revenue: $574.7M
Q2 Adj. EBITDA: $173.1M
Q2 Adj. EBIT: ($202M)
2026 Revenue: $3.38B
2026 CapEx: $22.5B

Company guidance:

2026 Revenue: $3-3.4B
Group Adj. EBITDA margin: ~40%
YE2026 ARR: $7-9B
YE2026 Connected Power: 800MW-1GW
YE2026 Contracted Power: >4GW
2026 CapEx: $20-25B

Nebius ended Q1 at $1.92B of ARR. Since the company defines ARR as revenue from the final month of the quarter multiplied by 12, simply holding March’s revenue run rate flat throughout Q2 would imply roughly $480M of quarterly AI cloud revenue before accounting for any additional capacity brought online during the quarter. Even that number is arguably conservative, as it assumes no benefit from contracts being renewed or repriced at higher rates during the quarter, meaning the same installed capacity could potentially generate more revenue.

Obviously, the business didn’t stand still after March. Even though capacity deployment is expected to be more pronounced in the second half of the year, Nebius continued bringing new capacity online throughout Q2, while TripleTen should contribute at least another ~$10M of revenue and Avride will likely remain negligible at the group level for now.

For that reason, I think Nebius has good chances of delivering a revenue beat.

The biggest uncertainty is timing. There’s a meaningful difference between adding the majority of Q2 capacity in April vs. adding it during the second half of June. Both scenarios could leave Nebius in essentially the same position heading into Q3 and have virtually no impact on the year-end capacity target, but they would produce very different Q2 revenue numbers.

That’s why I care considerably more about the exit rate and the capacity available entering Q3 than about whether Q2 revenue lands a few million dollars above or below a particular number. Still, with consensus currently sitting at $574.7M, I like the setup.

The same logic applies to adj. EBITDA. Nebius delivered $129.5M of group adj. EBITDA in Q1, representing a margin of ~32%, while Nebius AI Cloud itself generated $174M at a 45% margin. Management continues to target 40% group Adj. EBITDA margins for the full year, but it also indicated during the Q1 call that group margins could come slightly lower in Q2 before returning to gradual expansion throughout the rest of the year.

Consensus is looking for $173.1M, which implies a margin of 30.1%. That looks consistent with what management communicated.

As a result, I think the EBITDA result will largely be a function of the top line. If revenue comes in ahead of consensus while margins behave as guided, adj. EBITDA should naturally follow.

On the $7-9B YE2026 ARR target, my base case is that management reiterates it.

Under different circumstances, I might actually be looking for a raise. Demand certainly doesn’t appear to be the problem, and Nebius entered Q2 with a record pipeline after Q1 pipeline generation increased roughly 3.5x sequentially.

The reason I’m more conservative is Vineland.

There’s some uncertainty around the timing of the remaining buildout there, so I don’t think management has any reason to get more aggressive with year-end guidance until it has greater visibility into exactly when that capacity becomes operational.

Management was recently asked directly about the issue and, according to DA Davidson, reiterated its existing guidance and confirmed that it remains on track to meet Microsoft’s contractual commitments.

That said, I also wouldn’t completely rule out Nebius providing its first meaningful target beyond 2026.

The attention currently being paid to whether the company can execute beyond this year’s capacity ramp has increased significantly, especially with well-known investors such as Michael Burry announcing a short position in the stock. In normal circumstances, I would’ve expected management to wait until Q3 or Q4 before giving investors a more concrete 2027 ARR framework. But with visibility into contracted power already extending well beyond the current year, I wouldn’t be shocked to see some kind of additional medium-term target.

I’m not entirely counting on it, though. This team has repeatedly shown that it doesn’t spend much time responding to every bearish narrative surrounding the company. Its preferred response is typically execution.

Finally, CapEx. Consensus currently sits at right around the midpoint of management’s guidance. Nebius spent ~$2.5B in Q1, meaning the spending ramp will accelerate substantially throughout the remainder of the year as new data centers are built out and filled with GPUs.

I personally expect full-year CapEx guidance to remain unchanged, especially after management indicated that component shortages, including memory, should have no more than a low-single-digit impact as a percentage of total spend. That said, I would welcome an increase if it reflected a stronger capacity deployment or funding outlook, rather than simply higher input costs.

Ultimately, what matters most is whether the company continues to execute against its key operational milestones and remains on track to meet its year-end targets, not whether it beats or misses quarterly estimates by a few million dollars. The bigger picture is much more important than that.

I say this every quarter, but capacity remains one of the most important variables to watch.

When the company had only seven operational data centers, I wrote in my Q4 earnings preview that I believed there were signs Nebius could be on a path toward 20+ locations over the following 18-24 months.

That thesis has played out faster than I expected.

Nebius has now formally announced 19 locations, and since Q1 I’ve found evidence pointing toward several additional deployments that haven’t yet been announced by the company.

The locations I’m watching most closely include:

• Wales

• Estonia

• India

• Singapore

• Additional capacity in London

Beyond those, there are also a few other possibilities I’m keeping an eye on, including Austria, potentially additional capacity in Paris (something I’ve been flagging for a few months without a formal confirmation yet), and further expansion in Israel.

And while I don’t have any concrete indication of another new U.S. location right now, I wouldn’t be surprised if Nebius is quietly working on one. Over the past few quarters, just when it seemed like the U.S. footprint was already largely mapped out, the company has continued announcing additional projects that weren’t expected by any member of the community. Given that pattern, I’m hesitant to assume the current list represents the full pipeline.

I’m far from suggesting all of these will necessarily be announced next week, or even that every one ultimately becomes a Nebius data center.

The broader point is that the deployment pipeline visible publicly continues to underestimate what appears to be happening behind the scenes.

We’ve seen this pattern repeatedly.

Nebius doesn’t announce every location the moment it begins working on it. Land, power, permitting and commercial negotiations can progress for months before a project is formally disclosed. That makes trying to model the entire capacity pipeline using only announced sites inherently backward-looking.

And this diversification matters even more today because of the discussion around Vineland.

I covered the Vineland situation in much more detail recently, so rather than repeat the entire argument here, feel free to check the full post on my X profile.

M. V. Cunha@mvcinvesting

DA Davidson lowered its PT on $NBIS from $250 to $175. The firm believes the current state of Vineland creates enough execution risk to question the year-end ramp and the company's valuation premium. I've read the entire report, so here are my overall thoughts on the issue: ⬇️

8:07 PM · Aug 7, 2026 · 118K Views

71 Replies · 57 Reposts · 785 Likes

Basically, there’s a legitimate possibility that Vineland is running behind the deployment schedule investors previously assumed.

DA Davidson’s concern isn’t really that Nebius won’t have access to power. The more relevant question is whether the physical data center infrastructure can be completed, commissioned and filled with GPUs quickly enough for all of that power to become active, revenue-generating capacity by year-end.

That’s a very big difference, but I don’t dismiss the risk.

Where I disagree with some of the conclusions being drawn is on what a potential delay would actually mean for Nebius as a whole.

First, Vineland isn’t binary. The site doesn’t have to go from roughly 50MW today to 328MW overnight on December 31. Capacity can be commissioned progressively, and a delay affecting some portion of the final buildout doesn’t automatically mean that the entire site has been delayed.

More importantly, Nebius is no longer dependent on a handful of locations. The whole point of building a geographically diversified portfolio across owned facilities, build-to-suit projects and colocations is that execution at one facility doesn’t have to determine execution for the entire company.

If Vineland is delayed by a few weeks or months but capacity brought online elsewhere compensates for it, Nebius still reaches 800MW-1GW of connected power, still delivers the $7-9B December ARR target and still fulfills Microsoft’s contractual commitments, then I simply don’t think the delay is particularly important to the investment thesis.

That’s the threshold I’m watching.

Does Vineland cause Nebius to miss company-wide guidance or contractual commitments? So far, we have no evidence that it does.

In fact, when specifically asked about the situation, management reiterated both its guidance and its ability to meet Microsoft’s commitments.

As I’ve explained before, Nebius’ virtualization and cloud orchestration layers give the company flexibility to manage workloads across multiple data centers rather than treating each location as a completely isolated pool of capacity. In theory, that should provide additional room to mitigate the impact of delays at any single site by directing workloads toward capacity becoming available elsewhere.

And this is where the growing number of locations becomes particularly important, not only the sites already being deployed, but also the much broader pipeline of potential locations the team is evaluating. Management has been quite clear that it’s continuously assessing dozens of potential sites, meaning the locations announced so far represent only a fraction of the opportunities being studied.

The larger that portfolio becomes, both operationally and in the development pipeline, the less dependent Nebius should be on the precise timing of any single project.

That doesn’t eliminate execution risk, of course, but it diversifies it.

So next week, I’ll be watching very closely for any update on Vineland, but even more importantly, I want to know whether management still expects 800MW-1GW of connected power by year-end, $7-9B of ARR, and full delivery against Microsoft’s contractual requirements.

I see any potential delay as a short-term execution risk worth monitoring, not something that changes the thesis.

Finally, last quarter, I said I expected Nebius to announce another major deal sooner rather than later. That hasn’t changed.

At Nebius Inflection, management appeared open to signing another agreement similar to the ones with Microsoft and Meta, and I continue to think another large customer commitment is only a matter of time.

Missouri remains the most obvious candidate, but it’s far from the only option given how broad the company’s deployment pipeline has become and how far in advance customers are now willing to contract for capacity that has yet to be deployed.

Nebius has also said it plans to deploy hundreds of thousands of Vera Rubin GPUs over the coming years. Given the demand frontier labs and hyperscalers have for next-generation NVIDIA infrastructure, I think those future deployments naturally create opportunities for additional large-scale contracts.

Also, given how much GPU pricing has increased over the past few quarters, I wouldn’t be surprised if the next one becomes the largest contract in the company’s history.

At the same time, it could be one of the last truly massive bare metal agreements we see from Nebius for some time. The long-term strategy remains focused on serving AI-native companies and enterprises through Nebius AI Cloud, Token Factory and higher layers of the software stack, where the customer base is more diversified and the economics are significantly more attractive.

Funding remains one of the most important pieces of the story for any company in this sector, as how the buildout is financed can make a significant difference to the ultimate ROIC of the business.

The question is where the capital comes from, at what cost, and how much dilution is required to fund the growth.

So far, I think the strategy is playing out pretty much as management had laid it out.

The first, and increasingly important, source is customer prepayments.

In Q4 2025, deferred revenue increased by roughly $1.56B, primarily due to upfront payments under customer agreements. Then, in Q1 alone, deferred revenue increased by another $3.2B, again primarily driven by customer prepayments.

Nebius ended Q1 with ~$9.3B in cash and cash equivalents, and I expect customer prepayments to remain a very meaningful source of funding.

As AI infrastructure remains capacity-constrained and customers increasingly want to lock in future supply, upfront commitments are becoming a bigger part of how the overall industry is financed. Nebius has explicitly said that it’s focused on maximizing prepayments from both existing and future customers to reduce the amount of external debt and equity it needs to raise, so I wouldn’t be surprised to see another few billion dollars come onto the balance sheet through prepayments this quarter.

The second piece of the strategy is debt backed by the contracts themselves.

Recently, Nebius completed its first senior secured debt financing, raising $775M. The facility matures on October 31, 2030, is priced at SOFR + 2.5%, and is backed by deployed GPU infrastructure together with contracted cash flows from an investment-grade customer. Nebius also said that the facility, combined with the cash flows from the underlying customer agreement, covers more than 100% of the CapEx required to deploy that infrastructure.

This is exactly the type of financing management has been talking about for several quarters: deploy infrastructure against a long-term contract with a highly creditworthy counterparty, then use those contracted cash flows and the underlying GPUs as collateral to obtain debt on attractive terms.

Once the infrastructure is operational, Nebius can effectively recycle some of that capital back into the broader AI cloud buildout.

This is also why I’ve always viewed the structure of the large hyperscaler contracts as much more important than simply looking at their headline revenue contribution.

Take the recent Meta agreement. In addition to the $12B of dedicated capacity, Meta agreed to purchase up to $15B of otherwise-unsold capacity across certain upcoming Nebius clusters over five years. Nebius intends to sell that capacity to its core AI cloud customers first, with Meta effectively acting as the backstop for what remains (which, given the current demand environment, will likely be nothing).

That kind of commitment dramatically changes the company’s financing profile, and it’s what makes these hyperscaler contracts such powerful financing tools.

I saw some criticism of the $775M raise because, relative to Nebius’s enormous capital requirements, it looks small.

That’s obviously true, but in my opinion, it’s also missing the point.

Why would Nebius raise every dollar it may need over the next several quarters today?

Capital has a cost. Raising debt too far before the capital is needed means paying interest on cash that’s sitting on the balance sheet. The much more capital-efficient approach is to raise money progressively as projects move through the development cycle and the capital is actually needed.

Importantly, the $775M transaction was significantly oversubscribed. Nebius described it as the first application of a repeatable financing framework and explicitly said it expects to raise additional capital on similar terms against more than $40B of contracted revenue from investment-grade customers (Microsoft and Meta).

So I don’t view $775M as some indication of the maximum amount Nebius can borrow. I view it as the first proof point that this financing channel works, and I expect more transactions like this soon.

In fact, at the Q1 earnings call, management had already said it was working on both asset-backed and corporate-level debt financing and expected to begin raising mid-single-digit billions of dollars through these channels in the near term.

Then there’s the potential impact of another large customer contract.

As I’ve said before, I expect Nebius to announce another major contract sooner rather than later. If that happens, the headline contract value will obviously matter, but I’ll be paying just as much attention to the payment structure.

A sufficiently large upfront payment, similar to the $7B paid by Microsoft, could fund a meaningful portion of the associated infrastructure before Nebius needs to raise a dollar externally. And if the counterparty is another investment-grade hyperscaler, the remaining contracted cash flows could potentially support another round of asset-backed financing.

That has always been one of the pillars of the strategy.

The hyperscalers aren’t the customers Nebius ultimately wants dominating its revenue mix. The long-term ambition is still to build a diversified AI cloud serving thousands of AI-native companies and enterprises.

But hyperscaler contracts can help finance the infrastructure that eventually serves those customers. They provide scale, visibility, prepayments and highly bankable contracted cash flows while Nebius builds out a much larger global capacity footprint.

Finally, there’s the ATM program.

Nebius established an at-the-market program covering up to 25M Class A shares in November, and as of the Q1 report management confirmed that it hadn’t used it.

I don’t expect Nebius to have used it during Q2 either.

However, that will inevitably change at some point. I continue to expect management to eventually tap the ATM, particularly when the share price is materially higher, potentially following another major contract announcement. And I’m perfectly fine with that.

Dilution has always been part of the thesis and something I’ve incorporated into every valuation model I’ve built for the company. You simply cannot model one of the fastest infrastructure buildouts in the industry and simultaneously assume that the share count will never increase.

What matters is the amount of dilution relative to the value created with that capital.

Considering how dramatically Nebius has accelerated its capacity ambitions compared with the plans it had when I first invested, I actually think dilution has remained quite controlled so far.

The company has increasingly diversified its funding stack, and that diversification matters for shareholders’ returns.

So, heading into Q2, there are a few things I will be watching particularly closely: the size of additional customer prepayments, any update on the secured-debt pipeline, whether the ATM remains untouched, and whether management provides any additional color on how much of the 2026 and early-2027 buildout is already funded.

There’s still an enormous amount of capital left to raise as Nebius moves toward multi-gigawatt scale. I don’t think anyone should pretend otherwise.

But the strategy has never been to raise all of that capital upfront. It has been to sequence the funding alongside the buildout, use large contracts to lower the cost of capital, maximize customer prepayments, recycle deployed assets into new growth capital and use equity selectively rather than as the default funding source.

So far, the team continues to execute on that playbook. If the strategy keeps playing out as intended, I think Nebius has one of the strongest combinations of customer demand, contractual backing and financing optionality in the sector to make the expansion happen.

On July 15, Nebius introduced an additional business line designed to scale its AI cloud globally through infrastructure partnerships. Importantly, this isn’t a replacement for its existing infrastructure strategy, but rather another way to bring capacity online and monetize demand without Nebius having to fund every dollar of the underlying infrastructure itself.

The easiest way to understand the difference is to compare it with colocation.

With colocation, a third party owns and operates the data center facility, while Nebius rents the space and power but still acquires and deploys its own GPUs and related hardware. In other words, Nebius avoids the cost and time required to build the physical data center, but it still has to fund the compute infrastructure that goes inside it.

Under this new asset-light partnership model, Nebius goes one step further. The partner finances and owns both the infrastructure and the hardware, while Nebius provides its systems architecture, hardware design, software and services stack, supply-chain access, and customer demand through its global sales organization. The partner operates the facility and hardware, while Nebius remains responsible for the cloud software and service levels.

This announcement created quite a bit of confusion, with some interpreting it as Nebius “pivoting its business model” because its previous strategy wasn’t working.

That interpretation misses the point entirely.

The core strategy remains unchanged. Nebius continues to aggressively expand its own capacity through owned data centers, build-to-suit facilities and colocations. This new model simply adds another avenue for growth, one that can expand the amount of compute Nebius can sell with minimal incremental capital requirements.

Given the current demand environment, the logic is pretty straightforward.

Nebius has been capacity-constrained for several quarters. CRO Marc Boroditsky has even discussed the unusual position of having to say “no” to customers because there simply wasn’t enough capacity available to satisfy all the demand.

That’s exactly where this model becomes interesting.

Nebius has spent heavily building out its go-to-market organization, but a great sales team can only monetize the capacity it actually has available. If customer demand is running ahead of Nebius’ ability to finance, construct and deploy its own infrastructure, bringing third-party capital into the equation provides another way to convert that demand into revenue rather than turning customers away.

The potential economics are also attractive. Nebius said these partnerships could take several forms, including revenue-sharing agreements, licensing fees, commissions and committed-capacity arrangements. That means this doesn’t look like traditional AI infrastructure revenue, where Nebius first deploys billions of dollars of capital into GPUs and data centers before generating a return. Instead, portions of this business could generate high-margin revenue while requiring comparatively very little capital from Nebius.

So no, this isn’t Nebius abandoning or pivoting away from its core strategy. It’s an additional lever to capitalize on a demand environment in which its ability to sell AI compute currently exceeds its ability to bring capacity online fast enough.

In the announcement, Nebius noted that the company had “already entered into initial arrangements under this asset-light model.”

I’d be particularly interested in hearing more about those arrangements during next week’s earnings call: their size, expected timing, economics, potential revenue contribution and, ultimately, how meaningful this asset-light model could become alongside Nebius’ traditional infrastructure expansion.

This section will be a little broader than expected, as quite a few people have been asking for my thoughts on the recent acceleration in open-source models, what it could mean for compute demand, and how I think Nebius is positioned within that shift.

With all the recent frenzy around open models, and the debate over whether falling inference costs could ultimately reduce the demand for compute, I expect management to double down on a message it has been consistent about from the beginning: the future of AI won’t be frontier models or open models. It’ll be both.

Roman Chernin has been particularly clear on this point.

The way he describes the evolution is pretty intuitive. When companies are first trying to crack a new use case, it often makes sense to start with the most capable frontier models from providers like OpenAI, Anthropic or Google. But once that use case starts working and usage begins to scale, economics become much more important. At that point, companies can increasingly look toward open and specialized models that can be tuned around their own data and workloads, sometimes delivering similar or even better performance for that specific application at a substantially lower cost.

That transition isn’t something Nebius views as a threat. Quite the opposite.

Roman recently recalled that during the original DeepSeek panic, when investors were worried that dramatically cheaper models would mean dramatically less infrastructure demand, Nebius actually had what he described as its best commercial week up to that point. The reason was simple: workloads that previously didn’t make economic sense suddenly became viable. In his words, when the same unit of intelligence becomes cheaper, customers don’t necessarily consume less intelligence, they can simply do more with the same budget.

I think this is one of the most important pieces of the Nebius thesis as inference becomes a larger portion of AI spending.

Cheaper tokens don’t automatically mean less compute demand. They can make entirely new workloads economical.

And we’re still very early in terms of how many of those workloads actually exist. Roman has repeatedly pointed out that, even at relatively advanced companies, AI is still being deployed across only a small fraction of potential use cases. Coding has been the first major application to reach meaningful scale, but there are countless other workflows that either aren’t technically possible yet or simply don’t make economic sense at current costs. As models improve and the cost of intelligence continues to fall, I expect many of those use cases to become viable, which should expand the overall market for inference rather than simply make today’s workloads cheaper.

This is exactly where Token Factory comes in.

Downloading the weights of an open model is easy. Running that model reliably and efficiently in production at meaningful scale is much harder.

Companies need to optimize how the model is served, orchestrate GPUs, manage caching and routing, maintain observability and reliability, continuously benchmark new models and, ultimately, squeeze as many useful tokens as possible out of every dollar of infrastructure. Token Factory is designed to abstract that complexity away and let customers buy the outcome, optimized inference, rather than worry about the GPUs underneath it.

Roman gave a great real-world example of this in one of the interviews. One customer was already running a fairly sophisticated inference workload on Nebius GPUs and managing the software stack internally. Nebius showed that Token Factory could improve the workload’s performance by roughly 20-25%, allowing the customer to run the same workload with around 20% fewer GPUs. But the interesting part is what happened next: the customer didn’t give the GPUs back. It used the efficiency gains to grow.

That’s probably the cleanest illustration of the opportunity here.

If Nebius can continuously lower the cost per token, customers can reinvest those savings into generating more tokens, serving more users and deploying AI into use cases that previously didn’t work economically. In that scenario, software optimization doesn’t cannibalize infrastructure demand, it can actually accelerate it.

We’re already seeing Nebius move aggressively to keep Token Factory at the frontier of the open-model ecosystem. Kimi K3, for example, was made available through Token Factory as an official Day-0 partner following its release. GLM-5.2 also reached Token Factory essentially at launch, while DeepSeek V4 Pro and many other models are already part of its current offering.

That release velocity matters because models are evolving incredibly quickly. An enterprise doesn’t want to rebuild its inference stack every time a new Kimi, DeepSeek, GLM, Qwen or MiniMax model comes out. It wants an infrastructure partner capable of testing, optimizing and deploying the best model for each workload while making those changes as seamless as possible.

And as we know, Nebius has been investing heavily to strengthen exactly that capability.

The acquisition of Eigen AI added model-level inference and post-training optimization expertise directly into Token Factory. Meanwhile, Clarifai’s core engineering and research team joined Nebius, with the company also licensing Clarifai’s inference and compute-orchestration technology. Nebius itself describes the combination quite neatly: Eigen optimizes at the model level, while Clarifai strengthens optimization at the system level.

And then there’s Tavily, which pushes Nebius another layer higher into the stack.

Nebius acquired Tavily to bring real-time agentic search directly into its platform, giving developers another building block for production AI agents. Roman describes the progression of the platform as something like bare metal → managed cloud → tokens → agents. At each step, Nebius abstracts more complexity away from the customer and opens its platform to a much larger universe of potential users.

I think that last point is particularly important.

Moving up the stack is obviously interesting from a margin perspective, but Roman argues that the bigger opportunity is actually market expansion. There are only a limited number of companies in the world capable of consuming hundreds of megawatts of bare metal infrastructure. There are far more companies that need managed cloud infrastructure, even more that need inference, and potentially tens of thousands of developers and enterprises that will ultimately consume AI through higher-level agentic services.

This is why I continue to view Nebius’ software investments as much more than some nice additional products sitting on top of the GPU business.

The software stack is what allows Nebius to turn the same underlying infrastructure into a broader, more diversified and potentially more valuable business.

There’s an economic angle to watch here as well. At the Nebius Inflection event, management highlighted that deals incorporating more of the software stack carried significantly better margins than bare metal contracts, generating payback periods of less than three years. That’s very similar to what some hyperscalers have been emphasizing recently.

So, during Q2 earnings, I’ll be listening closely for any additional color around Token Factory adoption, customer growth, model onboarding, and most importantly, whether Nebius is seeing a meaningful increase in the portion of customers consuming higher layers of its stack.

Capacity will remain the headline number for the foreseeable future. But over the long run, I believe how much value Nebius can extract from each unit of that capacity could be just as important as how many GPUs it deploys.

I’m obviously biased here. Nebius is my largest position, so keep that in mind when reading my expectations.

That said, I also think my optimism has been earned by the company’s execution.

Quarter after quarter, I’ve gone into earnings with what I thought were already ambitious expectations, and somehow this team has continued to surpass them. I’ve repeatedly had to update my assumptions, increase the scale of what I thought was possible, and rebuild my valuation models around a company that has been growing and expanding significantly faster than I originally expected.

The capacity roadmap is a perfect example. The funding strategy is another. The size of the contracts, the speed at which new locations have been added, the buildout of the software stack… there have been multiple points over the past year where the company simply forced me to rethink what I thought the opportunity could become.

So, after that track record, it wouldn’t make much sense for me to approach next week’s report without being optimistic.

That doesn’t mean I expect everything to be perfect. Far from that. There might be quarterly noise, execution challenges, and likely plenty of things I get wrong in this preview. This is one of the most ambitious infrastructure buildouts in the industry, and the bigger Nebius gets, the harder execution becomes.

But so far, execution has been outstanding, and until the company gives me a reason to believe otherwise, I expect that strong performance to continue.

Now let’s see what they’ve been cooking in the background.

As always, these are simply my personal expectations and opinions based on the information available today. I could be wrong on several of them, and nothing in this article should be considered financial advice.

Best regards,
M. V. Cunha

Disclaimer: The views expressed in this article are solely my own and are based on my personal research and analysis. This content is for informational purposes only and should not be considered financial, investment, or legal advice. Always conduct your own research before making investment decisions.