AI Infrastructure play growing ARR by 684%!
Nebius Investment Case!
We are currently in the midst of an AI boom that OpenAI started after stunning the world with the release of ChatGPT.
Many predict that this AI boom will have a bigger impact on the economy and the stock market than the internet!
Nvidia has been the poster child of this new AI revolution, as AI start-ups and cloud hyperscalers are spending hundreds of billions of dollars on GPUs that run AI workloads. This unprecedented demand sent Nvidia’s market cap to $3.5T, making it the second largest company in the world, only behind Microsoft.
While Nvidia became rich by selling AI picks and shovels, Nebius aims to become rich by being an AI refinery and processor.
In gold mining, after miners have used picks and shovels, before gold can be of any use, it must be refined and processed.
Similarly, raw GPU power (picks and shovels) is not enough. To deliver great AI services, companies need to refine AI models to make them better, more efficient, and cheaper.
This is what Nebius does.
In a sense, Nebius is hoping to become the “AWS of AI”!
1. Brief History
2. Business Model
3. Subsidiaries
4. The Opportunity
5. Unit Economics
6. Financials
7. Valuation
8. Conclusion
1. Brief History
Yandex is one of Europe’s largest internet technology companies. It was founded in Russia in the 1990s and became extremely popular, and is essentially Russia’s answer to Google.
In the 2010s, the company began expanding outside Russia and built various internet businesses in Europe and the US.
However, if you haven’t lived under a rock, you know that in 2022, Russia invaded Ukraine.
Yandex founder and CEO, Arkady Volozh, didn’t agree with this decision and openly spoke out against the invasion. Despite that, his company, Yandex Group, was put on the sanction list by the US, and thus it was delisted from the Nasdaq.
Shortly after, in part due to pressure and in part from his own desire, he sold all Yandex’s Russian assets and kept the foreign assets.
He renamed the remaining company Nebius and announced an ambitious goal to create a large AI infrastructure company.
After the split was completed, Nebius was removed from the sanction list and listed back on the Nasdaq.
2. Business Model
In the simplest of terms, Nebius is a data center business, similar to Amazon Web Services, Google Cloud, and Microsoft Azure. The company makes money by renting access to its data centers.
However, unlike the cloud hyperscalers, which mostly handle general computing, Nebius data centers specialize in handling AI workloads!
Nebius has tailored everything in its data centers to the very specific requirements that AI workloads have.
What is the difference between AI workloads and general Cloud computing?
Well, AI workloads demand extremely high performance, efficiency, and scalability, which Nebius delivers in a few key ways:
Purpose-built AI servers
Customized networking gear
Proprietary server racks
Innovative liquid cooling
Instead of buying generic off-the-shelf servers from Supermicro, Nebius designs its own server infrastructure.
An AI server is basically a Lego. It contains GPUs, CPUs, memory chips, networking equipment, cooling systems, and more.
Nebius engineers decide how many Nvidia GPUs, AMD CPUs, Micron memory chips, and Broadcom networking chips each server needs, where they will be positioned, and how the liquid cooling system will reduce temperature.
The actual equipment assembly is done by contractors such as Foxconn.
Nebius engineers learn from each installation and modify designs to improve output and reduce costs!
This approach could potentially allow Nebius to have lower operating costs. The company claims that its servers use 20% less electricity than off-the-shelf servers.
Also, the liquid cooling system enables servers to operate at higher temperatures without damaging the system. Heat is a big issue for data centers, as “hot” servers are much slower than “cold” ones.
Having powerful AI servers is only half of the equation. Without great software that effectively uses the computing assets at its disposal, these servers are useless.
Thus, Nebius extended the same vertically integrated approach to software, with it handling key aspects of the AI training process, such as:
Scheduling
Virtual networking
Storage
Data telemetry
and more
Scheduling refers to the process of managing and distributing different AI workloads across the cloud network. Smart and efficient scheduling is crucial to get the most out of thousands of very expensive Nvidia GPUs.
Custom virtual networks ensure isolated, secure, and low-latency GPU-to-GPU communication. This is a must when handling multi-tenant traffic in a large cloud network.
AI training involves reading petabytes of data quickly and repeatedly. This means that data must be moved from storage quickly and without errors.
Lastly, data telemetry is crucial for the automatic collection of operational data from the Nebius cloud platform for analysis. Customers must see the progress their AI workloads are making and how much they are paying.
This in-house development approach to hardware and software enables Nebius to manage costs and deliver better service quality.
3. Subsidiaries
As I mentioned in the history section, Nebius kept the non-Russian assets after splitting from Yandex.
Thus, they have a few interesting subsidiaries, and some of them might become billion-dollar companies in their own right, or could be sold off to raise capital for the core AI infrastructure business.
AVRIDE
AVRIDE is Nebius autonomous driving division, and in my opinion, this is the most promising subsidiary.
The company specializes in developing autonomous driving technology for self-driving cars and delivery robots.
Per the 2024 Investor Presentation, Avride has 200 software engineers exclusively working on developing this technology.
Their food delivery robot, visible in the picture above, works within the Ohio State University compound to make over 1,200 deliveries a day. Students can use the Grubhub app to order from a few participating restaurants and get a meal delivered straight to their dorm.
Avride has signed deals with Uber Eats to test their delivery robot in more locations across the US. Meanwhile, Rakuten was chosen as a partner in Tokyo.
The small robot is purposefully designed to serve short-distance, high-density environments for small-scale logistics deliveries, such as restaurant meals, groceries, documents, packages, and more.
Moving on to the self-driving vehicle.
Avride has signed a long-term partnership agreement with Hyundai to develop self-driving technology to be used in Robotaxis in Dallas, Texas.
The company expects to spend around $150M in total on this division in 2024 and 2025.
It is highly likely that Avride will have to be split from Nebius, as the company will need billions of dollars in funding to scale its operations.
However, it is also possible for Avride to be sold off completely, to raise capital for the main AI data center business. Avride could be valued at a few billion dollars.
Others
Toloka is a data labeling company that helps AI models identify data. Recently, Toloka received significant investment from Jeff Bezos, the billionaire founder of Amazon.
As a result of new investors, Nebius no longer owns the majority of Toloka’s shares, so it will be deconsolidated from their financial statements.
Recently, a company in a similar niche, ScaleAI, received a $14.8B investment from Meta. While the scope of ScaleAI is much larger, nevertheless this is a signal that Toloka’s value is potentially underestimated by the market and Nebius investors.
Beyond Toloka, Nebius also owns TripleTen and ClickHouse.
Tripleten is a tech-education platform, helping people transition careers.
ClickHouse is an open-source database company in which Nebius has a 28% stake.
In total, Nebius stake in these companies could be worth $2-4B.
4. The Opportunity
Nebius believes they have a massive opportunity to become one of the key AI technology suppliers, becoming the “AWS of AI”.
The company estimates that its TAM will reach $260B a year in 2030!
This expansion in TAM will be driven by changing AI computing dynamics.
Today, a large majority of AI computing is spent on training AI models, not inference. In the future, that is expected to flip, with analysts predicting that already by 2027, 64% of AI spending will go to inference.
Currently, AI applications haven’t seen large corporate adoption, as they require a lot of computing power and consume a lot of energy, thus making them quite expensive. For instance, OpenAI is projected to spend over $46B in the next 4 years.
However, it is expected that in the next few years, AI models will get much better!
Not only will models get more advanced, but they will also become more efficient. This means that unique use cases will become economically viable. AI will get embedded deep into various parts of our daily lives.
This will drive demand for inference, and Nebius is already preparing for this future.
They are building data centers and software to support AI inference at a massive scale.
5. Unit Economics
As Nebius is a very capital-intensive business, A key question is what the unit economics are for what they do.
So, let’s model the economics of a $1B investment in a new AI data center project!
Nvidia GPUs are the largest expense category because AI workloads require the most advanced chips. Additionally, as the demand greatly outstrips supply, Nvidia and its foundry, TSMC, demand very high prices.
After GPUs, all supporting equipment must be purchased, including CPUs, memory chips, networking, and cooling equipment. Additionally, land must be acquired, and construction needs to take place.
Hyperscalers sometimes spend 20-30% more on supporting infrastructure than on GPUs, but as AI needs the best chips, this is unlikely.
For simplicity, let’s assume a 60:40 split for GPUs and all other expenses.
This leaves us $600M to spend on Nvidia GPUs!
Nvidia H200 GPUs cost around $45K, but Nvidia likely gives a volume discount, so assuming $37K per chip, we can buy around 16 thousand GPUs, which Nebius plans to depreciate over 4 years.
Let’s assume that the other $400M gets depreciated over 10 years.
Nebius likely charges its customers a per-hour access fee, and from what I could find online, the price is around $3 an hour.
If we assume an average 80% utilization rate over these 4 years, Nebius could generate $1.63B in revenue over the 4 years!
But Nebius must also pay for electricity, maintenance, all the software licenses, and other direct costs.
Let’s assume that these expenses will be 20% of revenue, leaving Nebius with an 80% gross margin.
That leaves us with a gross profit of $1.1B!
After accounting for the depreciation of GPUs and other equipment, we arrive at a project operating profit of around $331M after 4 years.
That is a pre-tax ROI of 8.3% per year!
However, this doesn’t take into account corporate overhead expenses such as legal, accounting, sales, marketing, taxes, R&D, and others. It’s very hard to estimate these expenses at this time, but they are likely to be significant.
There are a few ways in which this analysis could be incorrect!
If the utilization rate drops from 80% to 60%, the project becomes unprofitable.
Additionally, AI GPUs are advancing rapidly, and competition for AI workloads is intensifying. Whilst customers are willing to pay $3 per hour today, in 2 years the price could drop to $2 or $1.
With 80% utilization and $2 per hour, this investment wouldn’t be profitable.
Furthermore, what if energy costs spike? That could cause gross profit to fall to 60%, again, making the project unprofitable.
Most crucially, some experts find the 4-year depreciation assumption questionable. AI workloads are extremely intensive, and there have been some reports from Nvidia customers that intense usage causes them to depreciate much faster, possibly even after 2 years.
However, there is also potential for this calculation to be wrong to the upside!
If the Nebius is correct, the demand for AI services could be much higher than I model.
For instance, if the utilization rate increases to 90%, and the price per hour to $4. Project revenue could reach $2B, almost 23% higher than the earlier estimate.
Furthermore, while the company depreciates these servers over 4 years, it might be that there is some value that can still be generated for an additional 1-3 years. If Nebius could achieve a 20% utilization in year 5, that would generate an additional $85M in revenues.
Additionally, gross margin could be higher than 80% if demand for their services is indeed as strong as the company hopes.
Furthermore, this model assumes that after 10 years, the entire $1B investment is depreciated, while in reality, the building, the land, cooling systems, and other equipment have higher useful lives.
If, after the GPUs are depreciated, Nebius would invest another $1B to replace them, this $1B investment would likely generate a higher ROI than the previous. The company would discover new and creative ways to generate additional value from existing infrastructure.
Lastly, this model doesn’t take into account any additional software revenue that Nebius might earn from these GPUs. These revenue sources will be crucial to make this a viable business, because as my example demonstrates, just renting the GPUs by the hour doesn’t seem to generate compelling returns.
If Nebius can generate 20% extra from these additional fees, the business suddenly seems much more compelling.
6. Financials
Nebius finances are not great in the conventional sense. The company has the finances of a venture-backed start-up, and if not for its origin history of splitting from Yandex, we probably wouldn’t even have the opportunity to invest in this company.
In Q1 2025, Nebius had revenues of $55M, an increase of 390% Y/Y.
Meanwhile, annualized ARR grew 684% Y/Y to $249BM, the company expects to close the year with ARR of $750M to $1B, indicating that there is huge potential for future growth.
While such growth is impressive, it came at a significant expense, with operating loss increasing 60% to $129.5M.
It will probably take many years before the company is anywhere near profitability, as the management has outlined aggressive expansion plans.
The company recently increased its capex guidance for 2025 from $1.5B to $2B, with significant investments planned in Israel, Finland, and the Middle East!
In its 2024 Investor Presentation, they stated a mid-term target of a few billion ARR. To achieve this goal, Nebius will require significant capex expenditures, likely on par or above the $2B 2025 level.
Fortunately, the split from Yandex left Nebius with billions in cash, however, a majority of that has already been spent.
As of Q1 2025, Nebius has $1.45B in cash, not enough to cover its 2025 and 2026 capex plans.
This means that the company will have to raise significant amounts of capital!
While some dilution is inevitable, the company’s debt-free balance sheet gives it flexibility to raise capital through debt rather than equity.
Just 2 weeks ago, Nebius announced a raise of $1B in convertible notes. $500M of which are due in 2029 and have a 2% coupon, and the other $500M are due in 2031 and have a 3% coupon.
Bond conversion prices are $61.74 in 2029 and $64.31 in 2031, meaning that bondholders will get 19.44M shares. As there are 238M shares outstanding as of Q1 2025, this amounts to a dilution of 8%.
This doesn’t seem that dilutive as it could have been. The low-interest rate payments mean that the company will have enough cash flow to invest in the business.
However, the bonds can also be repaid in cash, and bondholders would be unlikely to convert to shares if Nebius trades below the conversion price. Logically speaking, if Nebius stock price is below the concession price, that means that they are underperforming expectations, meaning they are unlikely to have the cash to pay the bond principal.
In any case, this is just the first of likely many debt issuances.
If market conditions are not as favorable as the company hopes, the next issuance might come at much worse terms than the current one.
7. Valuation
What investors need to understand is that Nebius is a very unique investment opportunity. Rarely do public market investors have the chance to invest in early-stage growth companies.
Nebius is essentially a start-up that venture capital investors would normally invest in!
Today, the financial performance of the company is atrocious, as we discussed in the financials, the losses are just stunning.
However, growth is stunning as well!
Whilst TAM seems to be large and quickly expanding.
But investors can’t have the cake and eat it too.
A venture-like investment comes with venture-like risks.
The reason why venture capital investors sometimes generate astronomical returns is that they invest in many early-stage companies that have the potential to go to zero or generate 1000x returns.
It is also funny how both VC failures and successes are always so obvious in hindsight!
But the reality is that it is much harder to identify a potential winner than many industry experts would like to admit.
Because of these reasons, valuing Nebius using conventional metrics is just pointless.
The company trades at a P/S of 70. Such a multiple would make Wall Street investors heads spin, but it is common for venture capital investors.
Valuation Model
This is the hardest valuation model that I have ever done, because there are so many variables, so I would like to preface that I have no confidence in it, it’s just a thought experiment, not a forecast.
Today, Nebius has around 30K GPUs, let’s assume that they have 300K by 2030.
Utilization of 70%.
$3.5 per hour.
20% other revenue on top of GPU revenue.
We get to revenues of $7.7B!
To get this GPU capacity, Nebius would likely need to invest more than $17B in capex, so let’s model $3B in depreciation in 2030, 18% of that.
Taking into account R&D, other operating expenses, and taxes, we get to a net income of $531M, a net income margin of around 7%.
Assuming a 6% dilution and an exit P/E of 40, we get to a 2030 share price of $63.89.
In this scenario, there is a 32.5% upside to Nebius.
These growth assumptions seem quite aggressive to me, but I could be wrong.
There are companies that are easier to understand by the average investor and easier to model that offer better returns!
Furthermore, as this industry is moving so rapidly, I can’t with any certainty determine if Nebius strategy will indeed work out.
AWS, Google Cloud have way more resources than Nebius, and they can get better prices from chip suppliers to have lower operating costs.
While Nebius is exclusively focused on serving the AI niche, what stops hyperscalers from customizing older data centers to meet the needs of AI workloads?
8. Conclusion
Nebius is building an extremely capital-intensive business in an industry where the economics are not fully known yet!
While on paper, the demand for AI workloads is huge, the economics of this business seem too uncertain to me.
This is a VC-like investment opportunity.
But it comes with VC-like risks.
I don’t think that this is a stock that the average investor should invest in.
There are too many unknowns, and unit economics don’t seem that attractive to me.
Of course, it is entirely possible that I am wrong and Nebius will become a $100B company as certain analysts suggest.
But investors need to be aware that Nebius is an extremely risky investment, with huge potential upside and downside.
Nebius is an extremely unprofitable company, competing with larger companies with significantly more capital!
However, as we are in the middle of a raging AI bull market, this stock could reach stratospheric highs. But that would be on the basis of extreme speculation, not operational results.
Thank you for reading Global Equity Briefing!
Global Equity Briefing is an investing newsletter with a focus on analysing global companies. I have written highly detailed Deep Dives on Nu Bank, Ferrari, Palantir, Grab, Celsius, Mercado Libre and Hello Fresh!
Additionally, I have written Investment Cases on Meta, Amazon and Google! and comparisons of Visa vs Mastercard and Eli Lilly vs Novo Nordisk!
My goal for 2025 is to write around 4-6 articles per month!
Subscribe to get all my articles as soon as they are released!
Support my work by becoming a paid subscriber!
You can follow me on Social Media below:
X(Twitter): TheRayMyers
Threads: @global_equity_briefing
LinkedIn: TheRayMyers
Disclaimer: Global Equity Briefing by Ray Myers
The information provided in the "Global Equity Briefing" newsletter is for informational purposes only and does not constitute financial advice, investment recommendations, or an offer or solicitation to buy or sell any securities. Ray Myers, as the author, is not a registered financial advisor, and readers should consult with their own financial advisors before making any investment decisions.
The content presented in this newsletter is based on publicly available information and sources believed to be reliable. However, Ray Myers does not guarantee the accuracy, completeness, or timeliness of the information provided. The author assumes no responsibility or liability for any errors or omissions in the content or for any actions taken in reliance on the information presented.
Readers should be aware that investing involves risks, and past performance is not indicative of future results. The author may or may not hold positions in the companies mentioned in the "Global Equity Briefing" report. Any investment decisions made based on the information in this newsletter are at the sole discretion of the reader, and they assume full responsibility for their own investment activities.
Great article thanks. Very insightful.
I took a small position a few months back at ~$25. Took another look at the financials and bailed out at 35. Of course it's gone higher since then but like you say this will either be hero or zero. Might pick up again if it drops but otherwise will leave well alone.
Thanks again
Your numbers are way off based upon company projections and other analyst reports. They hope to be EBITA break even in later half of 2025. They should have 1 GW of capacity by 2030 at least. I could go on …….