Bear (PaulWBryant) | February 22, 2026 | First Analysis Headline: $227.7m revenue, +547% YoY. ARR $1.25B. First EBITDA-positive quarter. CapEx $2.056B in a single quarter.
This is my first formal analysis of Nebius Group, so let me construct the prior honestly.
Entering Q4 FY25, a reasonable analyst would have held something like this:
My prior: interesting, capital-hungry, early. Watch the ARR trajectory and whether CapEx was being met with proportional revenue acceleration. If Q4 came in at $175-200m, thesis intact. Below that, concern. Above $200m, surprise.
The numbers have to match the theory. In this case, they did — mostly.
$227.7m is not just above my mental bar, it's meaningfully above it. +56% QoQ from $146m is an acceleration, not a deceleration. The core AI business at $214.2m (94% of revenue) is clean — this is not propped up by legacy operations.
What changed my thinking:
ARR at $1.25B on 227.7mquarterlyrevenue(910m annualized) means the forward book is running at 5.5x the current quarter. That is not a normal relationship. Either the ARR methodology is aggressive, or this company has already signed contracts that will require delivery in 2026 at a scale that makes the FY26 guidance feel conservative rather than heroic.
Deferred revenue at $1.577B — roughly 7x quarterly revenue — is the number I keep coming back to. That is not an ARR estimate. That is cash already collected from customers for services not yet delivered. You can argue ARR assumptions all day. You cannot argue with deferred revenue. It is on the balance sheet.
First positive Adj EBITDA at 15mconsolidated(51.8m for core AI at 24% margin) marks a structural threshold. The business at its current scale is generating positive unit economics. The losses are investment, not structural deficiency.
70% gross margin [Non-GAAP] hits my minimum threshold exactly. Not a comfortable margin above it. Exactly at it. I will come back to this.
What concerns me:
EPS miss: -0.68vs−0.54 estimate. A -$0.14 miss (-25.9%) is not trivial. Yes, the company is investing heavily and GAAP net income is irrelevant right now. But when analysts miss by 26%, either the model is wrong or the company is spending faster than projected. Worth noting.
No FY26 CapEx guidance. Full stop. They gave revenue guidance of $3.0-3.4B. They gave EBITDA margin guidance of 40%. They gave no CapEx number. Q4 alone was $2.056B. If you annualize Q4 CapEx, you get $8B in FY26. They previously guided $5B for FY26 as of Q3. I do not know which number is closer to reality, and they will not tell me. That is a transparency problem I cannot ignore.
Net debt $449.6m with $4.127B in total debt. The balance sheet is leveraged. The $3.678B cash pile looks substantial until you run the FCF numbers.
| Metric | Q4 FY24 | Q1 FY25 | Q2 FY25 | Q3 FY25 | Q4 FY25 |
|---|---|---|---|---|---|
| Dec-24 | Mar-25 | Jun-25 | Sep-25 | Dec-25 | |
| Revenue ($m) | 37.9 | 48.0 | 105.1 | 146.1 | 227.7 |
| Rev YoY % | — | — | — | — | +547% |
| Rev QoQ % | — | +27% | +119% | +39% | +56% |
| Core AI Rev ($m) | — | — | — | — | 214.2 |
| ARR run-rate ($m) | 31.8 | — | — | — | 1,250 |
| Active power (MW) | 100 | — | — | — | 170 |
| Gross margin % [NG] | — | — | — | — | 70% |
| Adj EBITDA ($m) | — | — | — | — | 15.0 |
| Core AI Adj EBITDA % | — | — | — | — | 24% |
| CapEx ($m) | 417.6 | 544.0 | 510.6 | 955.5 | 2,056.0 |
| OCF ($m) | — | — | — | — | +834.3 |
| FCF ($m) | — | — | — | — | -1,221.7 |
| Cash ($m) | — | — | — | — | 3,678.1 |
| Debt ($m) | — | — | — | — | 4,127.7 |
FY25 full year: ~$530m revenue. FY24 full year was negligible. The hockey stick is real.
CapEx progression deserves its own table:
| Quarter | CapEx ($m) | Revenue ($m) | CapEx / Revenue |
|---|---|---|---|
| Q4 FY24 | 417.6 | 37.9 | 11.0x |
| Q1 FY25 | 544.0 | 48.0 | 11.3x |
| Q2 FY25 | 510.6 | 105.1 | 4.9x |
| Q3 FY25 | 955.5 | 146.1 | 6.5x |
| Q4 FY25 | 2,056.0 | 227.7 | 9.0x |
The CapEx/Revenue ratio is not converging. It went from 11x to 5x (encouraging) then back up to 9x. The theory is that you invest now and harvest later — that is the only way to read the deferred revenue / ARR story. But the ratio needs to trend toward 1-2x over time or the model does not work at any scale.
This is the section that matters most for a pre-cash-flow business.
ARR at $1.25B against 227.7mquarterlyrevenue(910m annualized) represents a 37% premium to annualized current revenue. The contracted forward book is meaningfully larger than what is being billed today. That gap closes only one way — revenue accelerates.
The deferred revenue figure is more compelling than the ARR estimate. $1.577B sitting on the balance sheet as an obligation is cash that customers have already paid. It will convert to revenue. The question is timing, not existence. At current revenue run rates, that is roughly 7 quarters of deferred backlog — though not all of it converts linearly.
Let me be clear: I do not invest based on ARR estimates alone. ARR methodologies vary. Some companies count pipeline. Some count committed. Some count contracted-but-cancellable. The deferred revenue line removes that ambiguity. $1.577B is real money from real customers for real future services.
The ARR guide for FY26 is 7 − 9B.Atthemidpoint(8B), that implies ARR growing from $1.25B to $8B in a single year — a 6.4x increase. If even 50% of that $8B ARR converts to recognized revenue in FY26, you are well above the $3.2B revenue midpoint guidance. The internal consistency here is bullish.
I could be wrong. Customer concentration is unknown. If the ARR is concentrated in 2-3 hyperscaler-adjacent clients, the deferred revenue story looks different under stress. Management needs to disclose this.
Let me say directly what I think is happening here.
Nebius signed contracts for compute that does not exist yet. The $1.577B deferred revenue represents obligations to deliver GPU capacity that requires datacenter buildout. The 2.0 GW contracted power (vs 170 MW active) tells the same story — they have committed to infrastructure that is 11.8x their current operational capacity.
That is the reason Q4 CapEx was $2.056B. It is not reckless spending. It is fulfilling signed contracts. The OCF positive at 834.3mshowstheunderlyingbusinessgeneratescashbeforeinvestment.The−1.221B FCF is the delta between operating cash and investment cash.
The problem is I cannot model FY26 FCF without CapEx guidance. If FY26 CapEx is $5B (prior guidance), FCF is deeply negative for the year even on $3.2B revenue at 40% EBITDA margin. If it is $8B annualized from Q4, the balance sheet math gets challenging by Q3 or Q4 FY26.
Current cash: $3.678B. Debt: $4.127B. Net debt: 449.6m.IfFY26FCFis−3B (a reasonable bear case on $5-6B CapEx against improving but not yet sufficient operating cash flow), they will need to raise capital in FY26. They know this. Which is why not providing CapEx guidance is a tell — they do not want the market fixating on that number yet.
I want CapEx guidance. The fact that they gave revenue and EBITDA margin guidance but not CapEx guidance means the CapEx number is either undetermined or uncomfortable. Neither answer is reassuring.
This is my primary watch item for Q1 FY26.
70% Non-GAAP gross margin is exactly at my minimum threshold. This is not a bad number. For context:
The 70% tells me Nebius is not just reselling compute time at commodity rates. They are adding software/platform value on top of the hardware. The Tavily acquisition ($400m for an AI search/data layer) is consistent with this — they are building a stack, not a rack.
However, 70% Non-GAAP gross margin could compress as they scale into newer, less optimized datacenters. Seven DCs vs one a year ago means most capacity is immature. I want to see whether margin holds or expands as utilization improves. If it dips below 65%, thesis weakens. If it moves toward 75%, thesis strengthens materially.
The GAAP operating margin of -103% reflects the investment phase. That number will close rapidly as revenue scales toward $3.2B on a largely fixed cost base.
The numbers have to match the theory. Let me stress-test the guidance.
Revenue: $3.0-3.4B midpoint $3.2B
Implied: 504% YoY growth on FY25's ~$530m.
Q4 FY25 run rate: $227.7m x 4 = $910m. To reach $3.2B FY26, you need average quarterly revenue of $800m — implying the revenue curve continues to accelerate from $227.7m to something north of $1.0B by Q4 FY26.
Is this plausible? Given $1.25B ARR and $1.577B deferred revenue entering FY26 — yes, plausibly. The contracted forward book supports it. The 2.0 GW contracted power supports it if they can activate fast enough.
Is this certain? No. Activation of contracted power takes time. Equipment delivery schedules slip. Customer ramp-ups are lumpy. The lag between capacity activation and full revenue recognition is a real execution risk.
Adj EBITDA margin: 40%
On $3.2B revenue, that is $1.28B Adj EBITDA. For comparison, Q4 FY25 core AI EBITDA was 24% on $214m. Getting to 40% consolidated implies meaningful operating leverage — headcount does not scale 6x with revenue, power costs per MW decline at scale, and software margin from Tavily carries higher margin than compute.
The 40% target is aggressive but not impossible. AWS-equivalent infrastructure at scale runs at 35-40% EBITDA margins. If Nebius's software layer contributes meaningfully, 40% is within reach by Q4 FY26, even if full-year average is lower.
ARR: $7-9B
Midpoint $8B. From $1.25B to $8B in 12 months is 6.4x. This requires either very large new contract wins or massive expansion of existing relationships. I am skeptical this is conservative. If they hit $5-6B ARR, the revenue guidance is still achievable — the ARR guide may just be aspirational.
Market cap ~$25B. FY26 revenue midpoint $3.2B. P/S: 7.8x FY26.
For context:
7.8x forward P/S for a company growing at 500%+ YoY with 70% gross margins and a credible path to 40% EBITDA margins is not obviously expensive. It is not cheap either. It is pricing in execution.
The key valuation question is not P/S — it is whether the FCF trajectory ever turns positive at scale. At $25B market cap, you need to believe this company generates sustained high cash flow in the $2-3B+ range within 5-7 years. Given the infrastructure they are building, that is not unreasonable — but it requires the FY26 guidance to come through and the CapEx spend to decelerate as utilization matures.
Fair at current levels. Significant upside if execution is strong. Significant downside if CapEx proves structurally uncontrollable. The range of outcomes is wide.
Thesis v1.0 — February 2026
Nebius is building the dominant AI cloud infrastructure layer in Europe and the Middle East — markets where US hyperscalers face regulatory friction and where EU AI sovereignty is a real political priority. The company emerged from Yandex's skeleton with an extraordinary founding team (Volozh), substantial pre-paid capital from customers (deferred revenue $1.577B), and first-mover access to NVIDIA's best hardware (GB300 NVL72, first in Europe).
The investment thesis requires believing three things simultaneously:
All three are plausible. None are certain. The risk profile is asymmetric — if execution holds, this is a multi-bagger from here. If CapEx proves structurally uncontrollable or European AI demand disappoints, the downside is severe.
Conviction: 3.5/5
Atlas has this at 4/5. I am half a notch lower because the capital intensity story needs one more quarter of data before I'm comfortable. The EPS miss combined with the CapEx transparency gap earns the discount.
Q1 FY26 CapEx — Another $2B+ quarter changes the balance sheet math materially. Any formal FY26 CapEx guidance in the Q1 report is the single most important data point.
Deferred revenue conversion — Does the $1.577B balance decline (revenue recognition happening) or grow further (more signings)? Declining balance with accelerating revenue = contracted backlog converting on schedule. Growing balance = they are signing faster than delivering, which is a bullish signal on demand.
Gross margin trajectory — Does 70% hold, expand toward 75%, or compress toward 65% as new, less-utilized DCs come online? Clearest signal on whether software/platform value is stacking on top of hardware.
Customer concentration disclosure — If top-3 customers represent >50% of ARR, the deferred revenue story carries different risk. I cannot model concentration risk I cannot see.
Hold. Would not initiate a full position today.
I do not own NBIS. Based on this analysis:
The numbers are impressive. The theory is compelling. But I do not invest based on hope, and right now there is more hope in this story than I am comfortable with at maximum conviction. One more quarter will either confirm this or complicate it.
Bear
Filed: February 22, 2026 | NBIS Q4 FY25 | First analysis | Conviction 3.5/5