Nebius Group (NBIS): Updated Valuation Model Post-Q3 Results
Earlier this month, NBIS reported its Q3 results.
You can find my detailed review here:
As you’ll notice, the company significantly raised its capacity expansion estimates, so it’s time to update my Valuation Model based on the new information available.
It’s always a great sign of execution when you have to repeatedly raise your previous assumptions to reflect reality, and that’s exactly what has been happening with Nebius ever since I started covering it in January.
Note: I’ve always followed the mantra: “If I need to build a fancy valuation model to prove a stock is cheap, then it’s not cheap enough.” Nebius easily passes that test in my view. That said, it’s been my largest position for a long time, and thousands of you regularly ask for deeper context. So even though I don’t believe a model is necessary to justify the investment, I still push myself to update it every quarter for the sake of transparency and consistency.
Before we begin, it’s worth revisiting my previous valuation model, which focused on YE2025.
At the time, many people said my assumptions were overly conservative or even disconnected from reality. But in hindsight, that wasn’t the case. NBIS will likely close the year within the range I projected. It’s also interesting that the stock’s recent high at $141/share landed very close to my prior bull-case valuation before pulling back sharply.
As I said back then, I don’t view these valuation ranges as Price Targets, but rather as a way to keep expectations grounded. That’s why I consistently choose prudence over optimism in my assumptions, especially given how quickly market conditions can change. In this case, that caution proved justified, at least for now (of course, I’d be very happy to be wrong if NBIS ends up closing the year at much higher levels).
What Changed After Q3?
Let’s discuss what has changed.
1. Updated Capacity Guidance
The company increased its outlook for both connected and contracted power, as well as issued YE2026 ARR guidance. This shift affects how we project active power deployment over the next years.
2. Financing Structure & ATM Program
Nebius has reinforced its balanced financing strategy through a mix of corporate debt, asset-backed financing, and equity. As part of that approach, the company announced an ATM equity program of up to 25M Class A shares, launched this month, which gives it flexibility to raise capital when needed. While dilution usually raises concerns, I’m not worried in this case, as long as Nebius maintains the same capital efficiency and discipline. That said, it’s important to account for the dilution in the model, as it is required to support the company’s capacity expansion goals.
3. Extending the Forecast Horizon
My prior model used 850 MW of active power by YE2026. With the improved visibility into contracted power, it now makes sense to extend projections and develop a YE2026 valuation using expectations for YE2027 active power instead.
And before moving forward, it’s important to remember a key operational distinction the company emphasized:
Contracted power refers to capacity legally secured from suppliers, committed on paper but not yet physically available.
Connected power is capacity that is fully built, energized, and ready to power GPUs immediately.
This difference matters for modeling because Nebius can contract enormous amounts of future capacity, but revenue growth only ramps once that power is physically connected and populated with GPUs.
My Updated Valuation Model
1. Core Business
Instead of using 850 MW of active power by the end of 2026, I’m now assuming 1.2-1.4 GW of active power by the end of 2027 to estimate a potential valuation at YE2026.
I’ve already explained the difference between contracted and connected power, which is why we can’t treat those targets as the same thing.
I do believe NBIS could eventually exceed this new range, but considering the enormous amount of capital required to turn contracted power into active, GPU-running capacity, I prefer to stay relatively conservative based on the information available today. Yes, the demand is clearly there and could support an even more aggressive expansion, but it’s not as simple as snapping your fingers and having data centers built and fully populated with GPUs.
Based on recent evidence, it’s also fair to say that my $10M per MW per year assumption is also conservative, but I’ll continue using it given the uncertainty around long-term competitive pressures and the monetization of Nebius’ software layer (which is another factor that provides additional upside).
Under that assumption, the ARR range by the end of 2027 would be $12B-14B.
If Nebius stopped deploying new capacity in 2028, that would likely be its revenue range. However, I don’t think that will happen by any means, so I’m applying a 10% uplift to each scenario to account for continued deployment. Again, I believe this 10% assumption is fairly low, but I don’t want to be overly optimistic considering how uncertain this market becomes further out.
On profitability, management has guided to 20-30% EBIT margins mid-term, based on a 4-year depreciation schedule for GPUs and equipment (compared to peers like CoreWeave using six years). I previously used the lower end of that range, but that guidance was based on mid-single-digit billions of revenue, with the potential to reach 30%+ according to management. Since I’m now using double-digit billions, sticking with 20% feels too conservative. At the same time, I don’t want to assume the high end given the uncertainty around the long-term unit economics of the sector, so I’ll use 25%.
Regarding multiples, in my last update I raised my previous 15-25x EBIT multiple range to 20-30x. This was my justification:
“Because predictability of revenue and cash flows is, in my opinion, one of the biggest drivers of valuation multiples. With this Microsoft contract, 5x larger than what I expected, Nebius now enjoys far greater visibility, utilization guarantees, and external validation.”
We now have even more visibility following the agreement with Meta and the overall demand dynamics, but I’ll keep the 20-30x range because I don’t want to go higher than that for a capital-intensive business.
Valuation Scenarios (Core Business Only)
• Bear Case (15% discount rate): $49.9B PV at YE2026
• Base Case (15% discount rate): $67.6B PV at YE2026
• Bull Case (15% discount rate): $87.3B PV at YE2026
2. Subsidiaries
Regarding the company’s non-core business units, I will keep most assumptions unchanged, but I’ve updated the present value calculations to the end of 2026 instead of the end of this year.
Avride
Valuing Avride remains inherently difficult due to its dual focus on autonomous ride-hailing and sidewalk delivery robotics, as well as the early stage of its monetization curve.
In previous versions of this model, I used Motional as the nearest proxy. Motional was valued at $4.1B after Hyundai increased its stake in 2024, and the comparison made sense at the time (especially considering my preference for conservative assumptions). Avride was founded earlier, has completed far more autonomous rides, and, critically, operates across both autonomous ride-hailing and delivery robotics, while Motional has yet to enter the latter category. These operational advantages made Motional’s valuation a sensible conservative baseline.
However, after reading Citizens’ research on Avride, I realized that Nuro is actually a closer operational peer. Nuro also runs a hybrid model combining autonomous last-mile delivery with purpose-built robotic vehicles, which more closely resembles Avride’s two-pronged approach than Motional’s narrower focus. Nuro’s most recent valuation was ~$6B (August 2025), making it a more relevant comp in several respects.
That said, I’m not adjusting my valuation range upward based on Nuro’s higher benchmark, especially since it was tied to expectations for 2028 and not today. Also, Uber and Nebius jointly announced up to $375M in strategic investment and related commitments to support Avride’s expansion. This aligns with what I’ve always expected: Nebius is not aiming to keep a majority stake in Avride long term. Over the long term, the company intends to reduce its capital exposure in non-core units, and the Uber partnership confirms that external funding, and therefore dilution, is likely underway.
Previous and Current Assumptions
• Avride IPOs (or reaches another exit event) in early 2028 at a valuation of $10-20B, driven by the expectation that it will have hundreds of autonomous vehicles and thousands of delivery robots in operation
• NBIS’ stake is diluted to just 30% in the process
Discounted Scenarios (using a 15% discount rate):
• Bear Case: $10B valuation × 30% stake → $3.0B stake → $2.6B PV at YE2026
• Base Case: $15B valuation × 30% stake → $4.5B stake → $3.9B PV at YE2026
• Bull Case: $20B valuation × 30% stake → $6.0B stake → $5.2B PV at YE2026
TripleTen
While TripleTen is the least significant of Nebius’ non-core business units in terms of near-term valuation impact, it remains a strategically valuable asset, particularly given its potential to supply skilled talent into Nebius’ broader ecosystem.
TripleTen is projected to more than double its revenue in 2025, following strong growth of 144% YoY in Q1 and 251% YoY for full-year 2024, with estimated top-line revenue reaching $40-60M. Comparable private edtech companies typically trade at 1-2x revenue multiples, despite growing at a much slower pace.
To stay conservative, I’m applying that same range of multiples and assuming revenue growth moderates to 50%, 60%, and 70% in 2026 (based on the 2025 midpoint of $50M).
This results in:
• Bear Case: $75M (1x 2026 revenue, assuming 50% growth)
• Base Case: $120M (1.5x 2026 revenue, assuming 60% growth)
• Bull Case: $170M (2x 2026 revenue, assuming 70% growth)
Toloka
Toloka is Nebius’ AI data solutions subsidiary (with NBIS now holding only a non-controlling stake), focused on providing high-quality training data for modern AI models. Previously consolidated within NBIS, the company was deconsolidated after Nebius ceded majority voting rights during a strategic investment round led by Bezos Expeditions (Jeff Bezos’s investment arm), with participation from Mikhail Parakhin, Shopify’s CTO and now Chairman of Toloka.
Although the valuation of that transaction was not disclosed, Toloka’s operational momentum has become increasingly evident. FY2024 revenue grew 138% YoY to $26.4M, and Q1 2025 revenue more than doubled YoY (we have no update after that since its numbers were deconsolidated from NBIS from Q2 onwards). Investor appetite for this segment also remains strong: META’s $14.3B investment for a 49% stake in Scale AI highlighted the strategic importance of high-quality data pipelines in the GenAI era.
Under new management, Toloka has started expanding beyond traditional annotation. Recently, it announced Tendem, which the company describes as the industry’s first hybrid AI + human agent. While the details go beyond the scope of this article, the launch signals a meaningful evolution in Toloka’s product strategy and reflects Mikhail Parakhin’s clear focus on bringing the company into its next phase of growth.
That said, I’m keeping my valuation assumptions unchanged. Toloka remains relatively small in absolute revenue terms and operates in a very competitive market where differentiation takes time to fully materialize. Until the company demonstrates sustained scale and margin expansion, maintaining the prior conservative assumptions feels appropriate.
Previous and Current Assumptions
• Toloka IPOs (or reaches another exit event) by early 2028 at a valuation of $5-10B
• NBIS’ stake is diluted to 20% post further funding
Discounted Scenarios (15% discount rate):
• Bear Case: $5B valuation × 20% stake → $1.0B stake → $870M PV at YE2026
• Base Case: $7.5B valuation × 20% stake → $1.5B stake → $1.3B PV at YE2026
• Bull Case: $10B valuation × 20% stake → $2.0B stake → $1.74B PV at YE2026
ClickHouse
ClickHouse represents, in my opinion, the most promising non-core asset within NBIS’ portfolio (depending on how long the company retains its stake). The company’s growth trajectory, technological edge, and market validation suggest enormous upside.
Before ClickHouse’s most recent funding round, NBIS disclosed a 28% ownership stake. Although Nebius participated in that round, it did not disclose the updated stake in its Q2 or Q3 results. Given its participation, I assume only modest dilution.
ClickHouse is rapidly emerging as a category leader in modern data infrastructure. It’s gaining market share from giants like Snowflake, Databricks, MongoDB, and Elastic. Its recurring revenue has more than quadrupled in the past year, and it’s been validated by marquee clients such as Tesla, OpenAI, Anthropic, Microsoft, Meta, Cloudflare, and ServiceNow.
The product’s speed, scalability, and developer-centric approach give it a real edge. A Senior Staff Software Engineer at Tesla summed it up perfectly:
“Data in ClickHouse is better than data anywhere else. No other system lets you slice and dice your data, ask interesting questions, and get answers in an acceptable amount of time. There’s nothing out there that competes with ClickHouse.”
This user loyalty, product depth, and elite customer base strongly suggest ClickHouse is on track for a highly successful IPO in the coming years.
As Arkady (Nebius’ CEO) put it:
“The way we’re thinking about this stake is that there’s a lot of value to be created in the business. But if there were to be a liquidity event in the coming years at a significantly higher valuation, then that’s something we’d potentially consider as a source of several billion dollars.”
That implies Nebius intends to hold the stake through IPO. And given management’s consistent participation in funding rounds, it’s clear they’re working to minimize dilution and retain exposure to future upside.
Despite the inherent uncertainty in IPO timing and broader market conditions (same as with Avride and Toloka, but in this case, the CEO already confirmed the company’s intentions to go public), ClickHouse is well-positioned to benefit from structural tailwinds in AI and data analytics. Its trajectory is reminiscent of Snowflake, which IPOed in 2020 at a $33B valuation after being valued at $12.4B privately just months earlier, and ended its first trading day at over $68B.
Previous and Current Assumptions
• ClickHouse IPOs by early 2028 at a valuation of $15-25B
• NBIS’ stake is diluted to 15%
Discounted Scenarios (15% discount rate):
• Bear Case: $15B valuation × 15% stake → $2.25B stake → $1.96B at YE2026
• Base Case: $20B valuation × 15% stake → $3.0B stake → $2.6B at YE2026
• Bull Case: $25B valuation × 15% stake → $3.75B stake → $3.26B at YE2026
Total Non-Core Business Units
As mentioned throughout this analysis, valuing Nebius’ non-core assets involves a high degree of subjectivity. That’s precisely why I used a multi-scenario approach to capture a realistic range of potential outcomes.
Here’s the final valuation range for the four non-core business units:
• Bear Case: $5.5B PV at YE2026
• Base Case: $7.9B PV at YE2026
• Bull Case: $10.4B PV at YE2026
In my view, this range is reasonable, perhaps even conservative. I’d argue that the bull case could easily exceed $10.4B, while the bear case appears quite achievable based on current progress and external validation. Of course, these are my personal beliefs, and readers are encouraged to form their own conclusions.
Sum-of-the-Parts (& Sensitivity Analysis)
Summing everything up, we arrive at a valuation range between $55.4B and $97.7B, with $75.5B as the base case scenario.
However, as mentioned earlier, we need to account for dilution. Nebius ended Q3 with 251.8M shares outstanding. Given the scale of capital required to support this level of expansion, additional equity issuance is unavoidable, whether through the ATM program or through future convertible instruments (most likely both).
My previous model already assumed an adjusted share count of 279.3M. For this update, I’m adding the full 25M shares authorized under the ATM program, plus an additional 25M shares to reflect the likelihood of future convertibles or similar financing mechanisms.
If this seems overly aggressive, consider the following: Industry experts estimate that building 1 GW of AI infrastructure costs between $30-50B. Arkady has already said that roughly 80% of the total cost comes from GPUs. While Nebius may achieve better cost efficiency than competitors, there is very little flexibility when it comes to acquiring GPUs themselves (aside from potential leasing structures, for example). Even if the 25M share ATM were executed at the current all-time-high share price, it would raise only about $3.5B. Nebius ended Q3 with $4.8B in cash, but management guided for roughly $3B in capex for Q4 alone, meaning that cash will be deployed quickly. To reach management’s long-term power targets, and by extension the assumptions in this model, Nebius will need to raise significantly more capital. There are certainly multiple financing avenues available, and pre-payments on contracts like the ones with Microsoft and Meta will likely help, but given the current lack of clarity, I believe it is reasonable to assume even more dilution.
One might argue that selling stakes in non-core business units could offset some of this dilution, but in that case, I shouldn’t include their valuation contribution at all.
A business this capital-intensive cannot scale aggressively without substantial dilution, so it’s important to reflect that in the model.
This brings my adjusted share count to 329.3M (vs. 251.8M today).
Based on all the assumptions outlined throughout the article, here’s the final table showing the potential share-price scenarios for YE2026:
For reference, today’s closing price was $94.69/share.
Here, it’s important to clarify two points:
1) The “bear case” I’m presenting doesn’t mean the stock can’t finish the year well below that level. My scenarios assume a relatively stable market backdrop. A true bear case would require a broader disruption in AI compute demand, and if that happened, every AI-exposed company would miss expectations and sector-wide valuations would compress sharply. As I’ve said since the beginning, this is a high-risk, high-volatility business, and investors must be prepared for that.
2) Despite the subjectivity involved and the dependency on AI market conditions, I truly believe most of my base assumptions remain conservative, especially considering the quality of this team. The $235.39 figure is NOT a Price Target. Execution will evolve, new data will emerge, and assumptions will need to be updated accordingly (just like I’ve been doing throughout the year). It’s simply a benchmark to illustrate that, even under what I consider reasonable inputs, the potential upside remains meaningful, even after my position has appreciated nearly 300%.
That’s why I made NBIS a 30%+ position at $25.67/share: I saw asymmetric upside that I’d be comfortable holding even if the stock doubled or tripled, exactly what ended up happening.
Of course, the risk-reward profile is different today, but I actually think it’s just as compelling, if not more so, than it was in the $50-60s before the Microsoft deal was announced. We now have far greater visibility, which meaningfully de-risks the investment.
Finally, it’s once again important to acknowledge that NBIS is a very challenging company to value, which is why I always try to stay grounded in my assumptions (even if they may still feel aggressive to some). I have zero interest in misleading any of my subscribers.
I’m obviously biased, so please make sure to study the company yourself and understand all the dynamics behind this valuation, as well as the broader risks associated with this investment.
That’s it, thanks for reading!
Don’t forget to follow me on X (@mvcinvesting), where I share daily news about the company.
Disclaimer: As of this writing, M. V. Cunha holds a position in Nebius Group (NBIS) at $25.67/share.
thank you very much
Nice one. Thanks.