There are moments in a bull market when financing becomes a compliment.
A company announces billions of dollars of new debt, the stock falls, and then investors begin explaining why the debt is actually proof of demand. That is roughly where Nebius stock finds itself today.
Nebius has become one of the most fascinating public-market experiments in artificial intelligence. In less than two years, it has transformed from the residual shell of the old Yandex structure into an AI-cloud company growing at a speed that would sound fictional in most industries. Q2 2026 revenue reached $582.3 million, up 454% year over year. Nebius AI Cloud alone generated $574.9 million. Annualized run-rate revenue reached $3.0 billion by the end of June. Group adjusted EBITDA jumped to $236.2 million. The core cloud segment produced nearly a 50% adjusted EBITDA margin.
And then the balance sheet started growing almost as fast as the revenue.
In August, Nebius raised $5.75 billion through convertible senior notes, after originally proposing $4.5 billion. The deal was upsized because investor demand was strong. The company intends to use the capital for data-center buildout, GPUs and the infrastructure needed to satisfy more than $40 billion of customer commitments.
That is the contradiction investors now have to price. The financing looks aggressive because the business is capital intensive. The financing also looks rational because demand is outrunning supply.
I think the real question is no longer whether Nebius can find customers. It clearly can.
The question is whether the economics of building this much capacity this quickly remain attractive after debt, depreciation, dilution and hardware obsolescence are included.
The old question was demand. That question has mostly been answered.
For much of 2025, the central risk around Nebius was simple: could a relatively new AI-cloud provider win meaningful workloads against hyperscalers with vastly deeper pockets?
The answer, at least for now, is yes.
Nebius says capacity is effectively sold as quickly as it comes online. Its shareholder materials indicate more than $40 billion of customer commitments. Four large deals signed in Q2 averaged more than $1 billion each. Management expects over $9 billion in customer prepayments during 2026. The company raised its contracted-power ambition repeatedly and now talks about a footprint measured in multiple gigawatts.
That changes the entire character of the investment thesis.
When demand is uncertain, the risk is commercial.
When demand is overwhelming, the risk moves into execution and capital allocation.
That is where Nebius stock sits today.
The $5.75 billion convertible raise is not automatically bearish
Convertible debt frightens shareholders because it contains two unpleasant words: debt and dilution.
But a convertible is not simply expensive equity disguised as a bond. It can be an efficient way to fund capital-intensive expansion when a company has a high stock price, strong expected growth and cash flows that are likely to improve.
Nebius issued two series: $3.45 billion of 0.50% notes due 2030 and $2.3 billion of 4.50% notes due 2034. The total reached $5.75 billion after purchasers exercised their full additional allotments.
The low coupon on the 2030 paper is striking. Investors were willing to lend at only 0.50% because the conversion option has value. In other words, bond buyers accepted a low cash yield in exchange for potential participation in future equity upside.
That sounds attractive for Nebius today.
It can become expensive for shareholders later if the stock performs extremely well and the securities convert.
This is why dilution should never be analyzed as a static share-count question. It is a contingent claim on future equity.
The same principle appears in smaller companies all the time. My book How to Value Penny Stocks spends a large amount of time on dilution because investors routinely confuse company growth with per-share growth. Nebius is obviously not a penny stock, but the capital-structure discipline is identical.
Why Nebius needs so much money
AI-cloud infrastructure is one of the strangest business models in public markets because growth looks like software from the revenue side and heavy industry from the cash-flow side.
A software company can add customers without building a power station.
Nebius cannot.
It needs land, grid access, substations, data centers, cooling systems, networking equipment and enormous quantities of GPUs. In Q2 alone, capital expenditures were approximately $5.7 billion.
That number is larger than the company’s quarterly revenue by nearly ten times.
On first inspection, that sounds absurd.
Then the customer prepayments enter the picture.
Nebius says 50%–60% of capex on certain new deals can be effectively self-financed through customer prepayments. Management expects more than $9 billion in prepayments during 2026. It has also introduced asset-backed financing tied to deployed GPU infrastructure and contracted customer cash flows.
This matters because it changes who actually funds the data center.
If customers prepay for future capacity, Nebius receives financing before revenue recognition. If lenders finance GPUs against contracted cash flows, equity holders do not need to fund the entire buildout directly.
The dream is a capital flywheel: contracts generate prepayments, prepayments fund capacity, capacity generates revenue, revenue supports debt, and debt funds more capacity.
The nightmare is almost the same sentence with one assumption changed: contracts weaken before the assets are paid back.
The AI-cloud margin may be the most important number in the entire story
Nebius AI Cloud generated $285.7 million of adjusted EBITDA in Q2 on $574.9 million of segment revenue, implying an adjusted EBITDA margin of roughly 49.7%.
That is extraordinary for infrastructure.
It is also why investors are willing to finance so much growth.
If a data-center operator can generate close to 50% adjusted EBITDA margins once capacity is utilized, leverage can be rational because the assets throw off substantial operating profit.
But I would be careful with the word “margin.”
Adjusted EBITDA excludes depreciation. For a company whose economics depend on GPUs that cost billions and become technologically obsolete, depreciation is not an accounting nuisance. It represents a real economic consumption of assets.
Q2 depreciation and amortization was $259.7 million, equal to 45% of group revenue. Nebius extended the useful life assumption for server and network equipment to five years from four years beginning in 2026.
That does not mean the accounting is wrong. It means investors need to understand how sensitive reported profitability is to the assumed life of hardware.
A GPU can remain physically functional for five years and still lose economic relevance much faster if newer architectures offer dramatically better performance per dollar or watt.
This is the most important connection to our current Nvidia analysis. Nvidia’s rapid hardware cycle is wonderful for Nvidia. For cloud operators, it creates an asset-aging problem.
Nebius is not CoreWeave — and that comparison matters
The market naturally compares Nebius with CoreWeave because both sell GPU-intensive AI cloud capacity.
The comparison is useful but imperfect.
CoreWeave has shown how quickly AI infrastructure revenue can scale when a provider secures major hyperscaler and AI-lab customers. It has also shown how leverage can become a central part of the equity story.
Nebius is trying to build a somewhat broader platform, with full-stack cloud services, inference products, owned software and adjacent assets. It also still owns interests in companies such as ClickHouse and operates businesses like Avride and TripleTen.
Those assets can create optionality. They can also make the valuation harder to parse.
For investors who want the direct comparison, our CoreWeave deep dive examines exactly why rapid AI-infrastructure growth and aggressive financing belong in the same model rather than separate bullish and bearish narratives.
Customer concentration is the risk nobody can diversify away with more GPUs
Large AI infrastructure contracts are wonderful until one customer represents too much of the economics.
When a customer signs a multibillion-dollar agreement, it improves visibility and financing access. But it also creates dependency. The provider may build infrastructure around specific technical requirements. Debt can be secured against those contracts. Capex decisions become tied to a small number of counterparties.
This is why I care less about the headline $40 billion of commitments than about their structure.
How concentrated are they?
How much is take-or-pay?
What happens if model economics change?
Who bears the cost of next-generation GPU upgrades?
Can customers renegotiate if compute prices collapse?
These questions determine whether a contract backlog is truly bond-like or merely a long-duration commercial relationship.
The power advantage is becoming the real moat
In 2024, AI-cloud companies talked mostly about GPUs.
In 2026, the scarce asset increasingly looks like electricity.
Nebius has repeatedly raised its contracted-power targets and is building across multiple regions. The logic is simple: GPUs can eventually be purchased by anyone with capital. Multi-gigawatt power access near fiber, water and permitting is harder to replicate quickly.
This is also why AI infrastructure increasingly resembles an industrial business rather than a software story.
IREN, which we analyzed in our latest IREN article, is exploiting exactly the same scarcity from a different starting point. IREN owned power-heavy Bitcoin infrastructure before the AI boom. Nebius is building the cloud stack directly. Both ultimately compete for the same constrained input: energized data-center capacity.
What could break the flywheel?
The first risk is utilization.
AI-cloud economics look magnificent when expensive GPUs are fully utilized. Idle GPUs are brutal. They still depreciate. They still consume financing capacity. The data center still costs money.
The second risk is price compression.
If GPU supply improves faster than demand, cloud pricing can fall. A company that underwrites a project at today’s contract economics can discover that replacement contracts earn less.
The third risk is customer credit.
Not every AI lab will become a profitable software giant. Some of today’s customers are spending venture and debt capital at extraordinary rates. If the underlying AI economics disappoint, infrastructure providers eventually feel it.
The fourth risk is technology obsolescence.
If each new GPU generation dramatically changes performance-per-watt, operators must upgrade to remain competitive. That can shorten real economic asset lives even when accounting lives stay long.
The fifth risk is financing.
Nebius has excellent access to capital today. Access to capital is not permanent. It is a market condition.
Why the stock can still work even after the huge run
There is a mistake value investors make with growth companies: they assume a stock that has already risen dramatically must be expensive.
Price history tells us nothing about valuation.
The relevant question is what future cash flows the current enterprise value implies.
Nebius is difficult to value because the denominator is still moving. Capacity, prepayments, debt, capex and revenue are all growing too quickly for a traditional steady-state multiple to feel comfortable.
But one number gives us a useful anchor: $3.0 billion of annualized run-rate revenue at June 30, before a large amount of contracted capacity comes online.
If Nebius can compound that revenue while preserving high segment margins and funding 50%–60% of certain projects through customer prepayments, the equity can justify a large valuation.
If the market is capitalizing revenue while ignoring the cost of constantly refreshing the hardware base, the stock can look cheaper than it really is.
My NBIS valuation framework
I would not use a conventional P/E ratio because current net income is distorted by massive depreciation, stock compensation, acquisition costs and rapidly changing capacity.
I prefer a future EBITDA-and-capital-intensity framework.
In a bear case, growth decelerates sharply after the current contract wave, AI-cloud margins normalize into the 25%–30% range, and financing remains expensive. The market begins treating Nebius more like leveraged infrastructure than software. That could justify a dramatic multiple compression even if revenue continues growing.
In a base case, ARR moves materially above today’s $3 billion level, cloud margins remain strong, customer prepayments finance a significant portion of capex and the company demonstrates that depreciation is not masking weak economic returns. In that world, the current premium becomes defensible.
In a bull case, Nebius becomes one of the few scaled independent AI clouds in a market dominated by hyperscalers, contracts compound beyond $40 billion, and power access becomes a structural moat. The equity can continue to re-rate because the company would then be valued less like a data-center operator and more like an infrastructure platform.
The problem is that the spread between those outcomes is enormous.
The most important metric is not revenue growth
If I owned Nebius stock, I would track return on newly deployed capital more carefully than revenue growth.
The company can always grow revenue by spending more billions on GPUs.
The question is whether each new billion invested creates enough durable cash flow to compensate equity holders for the financing risk.
I would watch four things:
Annual contract value per megawatt. Management says economics on recent deals are improving. That needs to continue.
Cash payback periods. The shorter the payback, the less dangerous hardware obsolescence becomes.
Customer-funded capex percentage. Prepayments materially reduce equity risk.
Free cash flow after growth capex. Eventually the business has to demonstrate that the infrastructure can generate cash rather than permanently consume it.
Final view: the debt is not the problem. The return on the debt is.
I understand why Nebius stock sold off when the convertible financing was announced.
Investors had spent months paying for explosive AI growth. Suddenly they were reminded that explosive AI growth requires explosive amounts of money.
But I do not think $5.75 billion of new convertibles is automatically a bearish development.
The company is sold out of capacity. Revenue grew 454%. AI-cloud adjusted EBITDA margins approached 50%. Customer commitments exceed $40 billion. Investors were willing to buy more debt than Nebius originally planned to sell.
The market is not financing a business without demand.
It is financing a business whose demand may be growing faster than its ability to build.
That is a much better problem.
It is still a problem.
Nebius now has to prove that every new gigawatt, every rack and every GPU creates enough long-term value to justify the liabilities attached to it.
If it can, the August debt scare may eventually look like the moment the market misunderstood what scale required.
If it cannot, the same debt will be remembered as the moment investors mistook customer demand for shareholder returns.
Nebius stock FAQ
Why did Nebius raise $5.75 billion?
Nebius raised convertible debt to fund data-center capacity, GPU purchases and global AI-cloud expansion. The offering was upsized from the originally proposed amount because of strong investor demand.
How fast is Nebius growing?
Q2 2026 group revenue increased 454% year over year to $582.3 million. Annualized run-rate revenue reached $3.0 billion at the end of June.
Is Nebius profitable?
Group adjusted EBITDA was $236.2 million in Q2 2026, while the Nebius AI Cloud segment generated $285.7 million of adjusted EBITDA at approximately a 49.7% margin. Net accounting profitability is affected heavily by depreciation, stock compensation and other items.
What is the biggest risk to NBIS stock?
The central risk is capital intensity: Nebius must invest billions in GPUs and data-center capacity while ensuring contracts, utilization and pricing produce attractive returns before the hardware becomes obsolete.
Sources and data status
Data status: August 30, 2026.
- Nebius Q2 2026 financial results filed with the SEC
- Nebius Q2 2026 shareholder letter
- Nebius: $5.75 billion convertible-note closing
This article is independent financial analysis for educational purposes and does not constitute investment advice.
