Our previous company analysis of IREN “Company analysis: IREN Limited (IREN)” focused on a simple question: can a company building AI infrastructure turn massive capex into high-margin, high-ROIC free cash flow?
The same question applies to Nebius (NASDAQ: NBIS – $59.84 billions). But there is one difference: Nebius is already much further ahead in proving the business works.
Q2 2026 revenue reached $582.3M, up 454% year over year. Adjusted EBITDA was $236.2M, a 41% margin. The company spent $5.7B on capex during the quarter alone. So this is already a real business with real margins, but one that requires an extraordinary amount of capital to keep growing.
The following is our analysis of Nebius Group ($NBIS) (remember this is not financial advice):
Current Situation
As of August 20, NBIS closed at around $220, down sharply following the announcement of a new $5B convertible offering. Based on its latest share count, that puts the equity value at roughly $58B before adjusting for the latest dilution and potential future conversion.
The business is growing at a ridiculous pace:
Nebius also had $8.0B of cash at the end of Q2, around $8.5B of non-current debt, and $5.0B of non-current deferred revenue. Cash flow from operations was strongly positive, but that was heavily supported by customer prepayments and deferred revenue growth. Capex was still far above operating cash generation.
That distinction matters. Adjusted EBITDA looks excellent. Free cash flow does not.
Nebius is spending heavily today because it believes the return on deploying another MW of AI infrastructure is attractive enough to justify it. That is the entire investment case.
The Numbers That Matter
Nebius has already secured some of the strongest contracts in the neocloud sector.
Microsoft signed a multi-year agreement for dedicated capacity from Nebius’s Vineland data center. The company said the economics were attractive enough to support debt financing secured against the contract.
Meta is even more important. Nebius agreed to provide Meta with $12B of dedicated capacity over five years, beginning in early 2027. Meta also committed to purchase up to another $15B of available capacity across certain future clusters if Nebius does not sell that capacity elsewhere.
The total potential contract value is therefore approximately $27B. That gives Nebius something extremely valuable: visibility. Demand is no longer purely theoretical. Major customers have committed to capacity years before all of that infrastructure is operational. But the same warning from the IREN analysis applies here:
Contracted revenue is not the same thing as free cash flow.
Nebius still has to build the infrastructure, deploy the GPUs, finance the capex and maintain attractive returns on that capital.
The Real Economics
Management said its major Q2 AI Cloud deals generated roughly $20–25M of annual contract value per MW. Around 70% of those deals included customer prepayments covering approximately 50–60% of the associated capex.
Management also estimated a payback period of around 1 year and 10 months for those Q2 deals, compared with its previous two-to-three-year range.
If those economics are sustainable, they are extremely attractive. The company is effectively saying:
Deploy capacity → secure contracted revenue → recover a meaningful part of the investment through customer prepayments → generate EBITDA at roughly 40% margins → recycle capital into more capacity.
That is a much better model than simply buying GPUs and hoping someone rents them. The problem is obvious: we are currently operating in an AI infrastructure market where demand remains ahead of supply.
The real question is whether $20–25M of annual revenue per MW and sub-two-year paybacks survive once significantly more capacity enters the market.
Nebius is also testing shorter-term capacity contracts at more than $40M of annual contract value per MW, according to management. If those economics prove repeatable, there is significant upside to current assumptions.
I would not value the company based on that yet. Those returns may reflect a genuine pricing advantage. But they may also reflect the current shortage. The next few years will tell us which one it is.
The Moat
The more interesting parts of the business are:
secured power and data center capacity
speed of deployment
the ability to secure financing against high-quality contracts
proprietary software and infrastructure management
relationships with large AI customers
the ability to offer both dedicated and flexible capacity
customer prepayments that reduce Nebius’s own capital requirements
Nebius is also trying to build a more asset-light model where third parties finance infrastructure while Nebius provides the technology, systems and commercial layer. If that works, it could materially improve returns on capital. That could become the real upside.
Today, however, I would still describe Nebius’s moat as partly structural and partly cyclical. The structural part is the technology, deployment capability, customer relationships and financing engine. The cyclical part is the ability to charge exceptional prices for scarce AI capacity. Investors should not confuse the two.
The Capital Problem
Nebius has better operating economics than many neocloud peers. It also has one of the biggest capital appetites. The company expects approximately $20–25B of capex in 2026. It spent $8.1B during the first half alone, including $5.7B in Q2.
It is funding this through several channels:
customer prepayments
operating cash flow
asset-backed financing
convertible debt
equity issuance
warrants and other equity-linked instruments
During the first half of 2026, Nebius raised roughly $2.85B through treasury share sales, alongside $2B from pre-funded warrants and $4.34B from convertible notes.
Then came the recent financing news. Nebius priced an upsized $5B convertible offering, consisting of $3B of 0.50% notes due in 2030 and $2B of 4.50% notes due in 2034. It also agreed to exchange $800M of existing convertibles for approximately 15.8M Class A shares.
The company expects approximately $4.94B of net proceeds, potentially rising to $5.68B if the buyers exercise their full options. This is the biggest issue I have with the stock at current levels. The growth is real. The margins are real. But the capital requirements are also very real. For shareholders, the relevant question is:
How much value does each additional dollar of capital create per diluted share?
Nebius can become a much bigger company and still produce disappointing shareholder returns if debt and dilution grow faster than the underlying value per share.
What Is Already Priced In?
At around $220, I believe the market is already pricing in a major AI infrastructure winner. That does not mean the entire upside is gone. Nebius could still grow into a valuation much larger than today’s. But the current price requires several things to remain true:
compute demand stays strong
capacity remains valuable
pricing holds up
margins remain high
financing stays available
customer prepayments continue
dilution remains manageable
The stock is effectively being valued on what Nebius could earn in 2027 and 2028. Current revenue is almost irrelevant to the valuation.
A Simple Valuation Framework
These are our estimates, not management guidance. The purpose is to frame the outcomes that matter rather than pretend there is one precise price target.
These ranges assume additional dilution from current and future financing. The bear case is not a business collapse.
Nebius can still become a much larger company in the bear case. The problem would be lower pricing, lower utilization, weaker returns on incremental capex and a larger share count. That is enough to destroy returns from today’s valuation.
The base case assumes Nebius executes broadly as planned, reaches significant scale and sustains margins around the mid-30s.
The bull case requires something more important: Nebius has to prove that its current AI infrastructure economics are durable rather than temporary shortage economics.
Bull Case Scenario
Nebius continues securing high-quality contracts before building capacity. Customer prepayments cover a meaningful portion of capex. The company keeps financing infrastructure efficiently through contract-backed debt instead of excessive equity issuance.
Revenue per MW remains above $20M. The company sustains around 40% EBITDA margins. The asset-light model starts working. The market remains supply constrained for longer than expected, allowing Nebius to retain some capacity for higher-priced short-term contracts.
Under that scenario, Nebius becomes a large AI infrastructure platform with technology, financing and deployment capabilities that make each additional MW more valuable. That could justify a much higher valuation.
Potential value: $425–570/share.
Bear Case Scenario
The bear case starts with one thing: AI compute capacity stops being scarce. Nebius does not need a collapse in AI demand for the thesis to weaken. It only needs supply to catch up faster than expected.
Revenue per MW falls. Premium pricing disappears. Utilization becomes less consistent. Customers negotiate harder. GPU refresh cycles remain expensive. At the same time, Nebius has already committed enormous amounts of capital to the buildout. That combination is dangerous.
Lower returns on new capacity plus rising debt plus continued dilution would change the story quickly. The company could still report huge revenue growth. The stock could still fall substantially.
Potential value: $125–160/share.
What Could Make Nebius Much More Valuable?
Three things stand out:
The current unit economics hold: if Nebius can consistently generate $20M+ of annual revenue per MW while recovering capital in less than two years, the company deserves a much higher valuation. That is the single most important operational variable.
Financing becomes a competitive advantage: customer prepayments and contract-backed debt are a huge advantage if Nebius can keep using them.
Every MW funded by customers or low-cost infrastructure financing requires less shareholder capital. That directly improves returns per share.
The business becomes less capital intensive: the asset-light model is probably the most underappreciated upside. If Nebius can increasingly operate capacity financed by partners while earning attractive returns from its technology and commercial platform, FCF conversion could eventually look much better than the market currently expects.
What Could Break The Thesis?
I would focus on five things:
Revenue per MW materially falls below ~$20M without lower capex per MW: the current investment case depends heavily on strong unit economics.
AI Cloud EBITDA margins fall below ~30%: at that point, the enormous capital intensity becomes much harder to justify.
Customer prepayments become a smaller part of the funding model: Nebius would need to replace cheap customer financing with more debt or equity.
Dilution accelerates faster than underlying EBITDA and FCF per share: growth does not automatically create shareholder value.
Capex remains above $20B annually without a clear path to positive, sustainable FCF: this would tell us Nebius is building a larger business without proving the returns on the capital required.
Final View
Current Situation
Nebius is already proving much more than a future AI story. Revenue grew 454% year over year in Q2, Adjusted EBITDA reached a 41% margin and major contracts provide significant demand visibility.
The problem is capital intensity. Nebius spent $5.7B on capex in Q2, expects $20–25B for 2026 and has just raised another $5B through convertible debt.
Bull Case
Nebius maintains $20M+ revenue per MW, sustains around 40% EBITDA margins, finances a growing share of capex through customer prepayments and contract-backed debt, and proves that its asset-light model can reduce capital intensity.
Potential value: $425–570.
Bear Case
AI infrastructure supply catches up, pricing and utilization weaken, returns on incremental capex fall and debt plus dilution grow faster than shareholder value.
Potential value: $125–160.
What The Market Is Pricing In
At around $220, the market is already pricing Nebius as a major winner of the AI infrastructure buildout. The upside from here depends on Nebius proving that today’s exceptional unit economics can survive at much larger scale.
What To Watch
Revenue per MW
AI Cloud EBITDA margin
Capex per MW
Customer prepayments
Realized revenue versus commitments
Debt and cost of capital
Diluted share count
FCF conversion
Utilization
Asset-light capacity
Thesis Breakers
Revenue/MW falls materially below ~$20M. AI Cloud EBITDA margins fall below ~30%. Customer prepayments weaken. Dilution grows faster than per-share economics. Capex remains above $20B annually without a credible path to sustainable FCF.
Verdict
Nebius currently looks like one of the strongest businesses in the public neocloud sector. The growth is real, the margins are already showing up, and the contract base is far more substantial than most competitors. But at around $220, investors are already paying for a lot of that success. The next leg higher depends on proving that the current revenue per MW, margins and payback periods can survive as AI capacity expands across the industry. If they can, the stock has significant upside. If those economics are mostly a product of today’s shortage, the capital intensity, debt and dilution can become a serious problem. But as of right now, our opinion is that Nebius is fairly valued.
Disclaimer: This article is for informational and educational purposes only. It represents our analysis and opinions based on publicly available information and should not be considered financial, investment, legal, or tax advice. We are not financial advisers, and nothing in this article should be interpreted as a recommendation to buy, sell, or hold any security. Investing involves risk, including the potential loss of capital. Always do your own research and consider your personal circumstances before making any investment decision.