Here is the mistake investors make with Nebius:
They see enormous contracts with Microsoft and Meta, compare those contracts with today's revenue, and conclude that the market must still be underestimating the opportunity.
But contract value is not market capitalization.
And market capitalization is not the entire capital structure.
Nebius is building one of the fastest-growing AI cloud platforms in the world. Its revenue growth, pricing power and customer validation are extraordinary.
The valuation is extraordinary too.
At approximately $226.59 per share as of September 23, 2026, the question is no longer whether Nebius has found product-market fit.
It has.
The question is how much of the future has already been pulled into the present price.
Nebius is not simply renting GPUs
A basic GPU cloud buys chips and rents them by the hour.
Nebius is trying to build something more valuable.
Its platform combines:
- GPU compute
- High-performance storage
- Networking
- Cluster orchestration
- Training infrastructure
- Managed inference
- Model optimization
- Enterprise security and compliance
That distinction matters.
Hardware rental is vulnerable to falling prices and increasingly standardized competition.
A full-stack platform can create higher utilization, better customer retention and more pricing power because customers build workflows around the software, not merely around access to a particular chip.
IREN's advantage begins with power and land, then moves upward into the cloud.
Nebius begins with cloud engineering and software, then builds outward into physical capacity.
Different starting points.
Similar capital requirements.
The growth is no longer theoretical
Nebius generated $582 million of revenue in the second quarter of 2026, up 454% year over year.
Its AI cloud business produced approximately $575 million of that revenue and reached $3 billion of annualized run-rate revenue.
More importantly, the AI cloud segment produced approximately $286 million of adjusted EBITDA, representing a margin close to 50%. Nebius Q2 results
Why does that matter?
Because Nebius is not merely buying revenue through uneconomic pricing.
At least at the segment level, the operating model is demonstrating substantial leverage as capacity fills.
Management continues to guide toward $3 billion to $3.4 billion of 2026 revenue and $7 billion to $9 billion of exit ARR. Nebius guidance
That would represent a remarkable ramp.
But exit ARR is not full-year revenue.
A company ending December at an $8 billion run rate does not necessarily generate $8 billion during the year.
Timing matters.
The Meta contract needs to be understood correctly
Nebius announced an AI infrastructure agreement with Meta worth up to approximately $27 billion over five years.
The first component is $12 billion of dedicated capacity based on NVIDIA Vera Rubin systems, with deployment beginning in early 2027.
The second component is an additional commitment of up to $15 billion. Nebius intends to offer that capacity to third-party cloud customers first, with Meta purchasing the remaining available capacity. Nebius and Meta agreement
That structure is attractive.
It gives Nebius potential upside if third-party demand remains strong while providing additional demand support from Meta.
But investors should not treat the full $27 billion as revenue already earned or as $27 billion of guaranteed high-margin profit.
It will be recognized over several years.
Capacity must still be built.
The associated hardware must still be purchased.
And the final margin depends on power, financing, depreciation and operating costs.
The headline is real.
So is the capital burden beneath it.
Microsoft and Meta solve one problem while creating another
Large customers improve revenue visibility and financing access.
Nebius recently raised $775 million of secured financing at SOFR plus 2.5%, backed by deployed GPUs and contracted cash flows from an investment-grade customer. The company said the financing and associated customer cash flows cover more than 100% of the capital expenditure for the underlying GPU deployment. Nebius secured financing
That is a powerful model.
Customer contracts make the assets financeable. Financing releases capital for further expansion. Expansion creates more revenue.
The flywheel is obvious.
But so is the risk.
Microsoft and Meta are customers today because AI infrastructure is scarce and difficult to deploy quickly.
They also operate some of the world's largest cloud and data-center platforms.
They are partners because they need capacity now.
They could become tougher competitors when that scarcity eases.
Nebius must use today's shortage to build a platform customers continue choosing after GPUs become easier to obtain.
That is why its software matters.
Power secured is not power producing revenue
Nebius expects to finish 2026 with approximately 5 GW of contracted power and plans to deploy more than 1 GW of capacity annually beginning in 2027. Nebius Q2 shareholder letter
Those numbers are enormous.
But investors must keep three definitions separate:
Contracted power is power Nebius has secured through agreements.
Connected power is power available at a completed facility.
Active power is power being consumed by revenue-generating computing equipment.
Only the third category pays the bills.
A signed power agreement is valuable because interconnection capacity is scarce.
It is not equivalent to an operating AI factory.
Nebius is consuming capital at hyperscaler scale
Nebius spent approximately $5.7 billion on capital expenditures during the second quarter alone.
Capital expenditure during the first six months of 2026 reached approximately $8.1 billion.
The company expects more than $9 billion of customer prepayments during 2026, which helps fund this construction. But those prepayments are not free cash flow. They are payments for services Nebius must still deliver.
That distinction matters.
Nebius reported $4.5 billion of first-half operating cash flow, but approximately $4.4 billion of that came from increased deferred revenue, largely reflecting customer advances.
The cash is real.
So is the future obligation.
Dilution and convertible debt are part of the valuation
At June 30, Nebius had approximately 272 million issued shares.
NVIDIA also owned pre-funded warrants covering approximately 21.1 million shares, purchased for $2 billion. Those warrants have a nearly nominal exercise price and should therefore be treated as economic shares. Nebius NVIDIA investment
Then, in August, Nebius exchanged $800 million of older convertible notes for approximately 15.8 million additional shares.
That brings the immediate economic share count to at least approximately 309 million, before considering other employee awards, remaining ATM capacity and potential future conversions.
Nebius also raised another $5 billion through convertible notes in August. Those notes initially convert at prices above $313 and $324, so they do not belong in today's basic share count. But they could create additional dilution if the stock performs well. August convertible financing
This is not necessarily destructive financing.
Nebius is raising capital at high share prices and relatively attractive debt terms to fund capacity supported by customer demand.
But an investor must value the company using the shares that economically exist.
Not the most convenient share count displayed by a finance website.
What does $227 require?
At $226.59 and approximately 309 million immediate economic shares, Nebius has an equity value close to $70 billion. NBIS price history
The August debt raise increased both debt and cash, leaving simplified net debt close to neutral before subsequent capital spending.
At an eventual enterprise-value-to-EBITDA multiple of 14 to 16 times, today's valuation requires approximately $4.4 billion to $5 billion of sustainable EBITDA.
At a 40% EBITDA margin, Nebius would need approximately $11 billion to $12.5 billion of annual revenue.
That is well above 2026 revenue guidance and above the company's projected 2026 exit ARR.
In other words, the stock already assumes that the 2026 ramp continues successfully into 2027 and probably 2028.
Nebius does not merely need to grow.
It needs to grow into expectations that are already enormous.
Three possible outcomes
These are analytical values, not price targets. I have not assigned a separate premium to ClickHouse, Toloka, Avride or TripleTen.
The bear case: roughly $47 per share
In this scenario:
- 2028 revenue reaches $8 billion.
- Pricing normalizes and the EBITDA margin settles near 30%.
- The market applies a 10 times EBITDA multiple.
- Net debt rises to approximately $8 billion.
- Diluted shares reach approximately 340 million.
The resulting value is roughly $47 per share.
This does not require Nebius to fail.
It only requires growth, margins and financing to fall substantially short of today's expectations.
The base case: roughly $205 per share
In this scenario:
- 2028 revenue reaches approximately $13 billion.
- The EBITDA margin reaches 40%.
- The market applies a 14 times multiple.
- Net debt is approximately $5 billion.
- Diluted shares reach approximately 330 million.
That produces an analytical value close to $205 per share.
The current stock price is already slightly above this outcome.
That tells us something important:
At $227, investors are not paying for average execution.
They are paying for excellent execution.
The bull case: roughly $414 per share
In this scenario:
- 2028 revenue reaches approximately $18 billion.
- Nebius maintains a 45% EBITDA margin.
- The market applies a 17 times multiple.
- Net debt remains around $3 billion.
- Diluted shares are contained near 325 million.
That produces an analytical value around $414 per share.
This requires Nebius to combine hyperscaler-scale infrastructure growth with cloud-software margins and disciplined financing.
Possible?
Yes.
Easy?
Absolutely not.
The accounting issue I would not ignore
Nebius increased the estimated useful life of its server and networking equipment from four years to five years beginning in 2026.
That reduces annual depreciation expense.
There may be a reasonable operational basis for the change, particularly when equipment is supported by multi-year customer commitments.
But investors should separate accounting life from economic life.
A GPU can remain physically functional for five years while losing much of its premium earning power as newer architectures arrive.
The true test is not how slowly Nebius depreciates the equipment.
It is how much revenue and cash flow the equipment generates before customers migrate to something better.
What to watch out for
Watch active power, not contracted power.
Watch realized revenue, not simply exit ARR.
Watch whether the 50% AI cloud EBITDA margin survives the next phase of expansion.
Watch depreciation, replacement capital and the residual earning power of older GPUs.
Watch customer prepayments relative to actual capital expenditure.
Watch the economic share count, including prefunded warrants and convertible securities.
Watch whether non-hyperscaler customers become a larger part of the business.
Most importantly, watch whether Nebius becomes a durable software-enabled cloud platform before AI infrastructure scarcity begins to normalize.
My conclusion
Nebius may be one of the strongest operators in the independent AI cloud market.
The revenue growth is real.
The margins are improving.
The customer validation is exceptional.
The software strategy is more differentiated than a simple GPU-rental business.
But $227 is not a price built around skepticism.
It is a price built around continued extraordinary execution.
I am more confident in the business than I am in the margin of safety at the current valuation.
That does not make Nebius a bad investment. It makes position sizing, entry price and continuous verification unusually important.
The principle is simple:
A wonderful AI business can still be a difficult investment when the stock price arrives at the destination before the cash flow does.