Date: 2026-02-25 Quarter: Q4 FY25 (Dec-25) Market cap: ~$24.6B | EV/FY26 Revenue: ~7.5x | ARR: $1.25B | Revenue growth: +547% YoY
When I first wrote about Nebius in January of last year, I called it "a bit of a leap in the dark" and confessed that applying my usual valuation methodology — EV/S against a growth cohort, PEG, FCF multiples — was essentially malpractice given where the company was in its development. That was an honest framing. What I said at the time was that the investment case rested on three pillars: Volozh's credibility as a builder, the certainty of AI infrastructure demand, and Nebius's ability to ramp capacity ahead of that demand. Thirteen months later, Q4 FY25 is the first quarter where I can say with some conviction that those pillars are not just intact — they are demonstrably performing. Group Adj EBITDA turned positive for the first time in the company's history at $15.0m (+6.6% margin). ARR hit 1.25B, blowingpastthehighendofguidance(1.1B) by $150m, and this is not a coincidence — management has beaten ARR guidance on every single measurement since listing. Revenue re-accelerated QoQ to $227.7m (+56%), reversing the deceleration we saw in Q3. Active power jumped from 100MW to 170MW, 70% ahead of the stated target. And the most important new disclosure: FY26 CapEx guidance of 16B−20B. That number reframes the entire investment case. It confirms that Nebius has contracted demand sufficient to justify deploying roughly five times FY25 CapEx in a single year. It also makes clear that FCF will be deeply negative through FY26, and that the economic payoff is a FY27+ story. Neither of those facts should be a surprise to anyone who read my original piece. The thesis is strengthening.
Current position: 2,300 shares in the High Growth Portfolio. No position in the Premium Portfolio. I hold.
The growth numbers for Nebius are so large they almost resist comprehension. +547% YoY on $227.7m in a single quarter. FY25 revenue of $529.8m against FY24's $91.5m restated. The critical question is not whether those numbers are real — the Microsoft and Meta contracts are public knowledge and the deferred revenue balance is on the audited balance sheet — but whether they tell us anything about what is coming.
The re-acceleration from +39% QoQ (Q3) to +56% QoQ (Q4) is the key data point. I flagged Q3 QoQ deceleration as a watch item in my earlier portfolio notes, and Q4 resolves it cleanly. The driver is straightforward: the Microsoft contract first tranche was delivered in November 2025, pulling ARR from $551m to $1,250m in a single quarter. That is not organic acceleration — it is a known contract delivery. But knowing the mechanism does not diminish the signal. The fact that the first Microsoft tranche delivered as scheduled, on time, is management execution, not just demand.
The more interesting data point is the sequential dollar increment. Incremental revenue was $5m in Q4 FY24, $20m in Q1 FY25, $50m in Q2, $41m in Q3, and $82m in Q4. The dollar adds are accelerating after a one-quarter pause. That pattern is consistent with the infrastructure build — capacity additions create step-changes in revenue recognition.
The FY26 revenue guide of 3.0B−3.4B is the first full-year guide the company has issued for the forward period. Management has a consistent pattern of under-guiding and over-delivering. They guided 750M−1.0B ARR for end-2025 and delivered $1.25B. I take the $3.0-3.4B range as deliberately conservative. My working estimate for FY26 revenue is 3.4B−3.8B. What gives me confidence in that is not an extrapolation model — it is the $1.577B of deferred revenue sitting on the balance sheet combined with the known contract structure of Microsoft and Meta.
This is the part of the story I do not think gets enough attention, probably because the GAAP operating loss line dominates the headlines. But the gross margin trajectory here is one of the more remarkable things I have seen in a hardware-intensive business. Q4 FY24 gross margin was 40.1%. One quarter later (Q1 FY25) it was 46.7%. One quarter after that (Q2 FY25) it jumped to 71.4%. It has held at 70%+ for three consecutive quarters. The jump from 40% to 70%+ reflects the mix shift from early-stage data center ramp (high construction/labor costs) to stabilised operating scale (primarily GPU depreciation + network costs) as the Hopper-era capacity became fully operational. D&A is not in cost of revenue — it is reported separately — which means the 70% gross margin is genuinely representative of the service-layer economics once assets are deployed and billing.
At 70% gross margin, Nebius is operating in the range of pure software companies. That is a striking fact for what is ostensibly a hardware-intensive AI infrastructure provider. It validates the full-stack differentiation argument. If Nebius were merely renting GPU compute, margins would look like a commodity business. They do not.
The Core AI segment EBITDA margin is now 24.0%, up from 19.1% in Q3. That is the right underlying profitability indicator for the operating business, stripped of the Avride and TripleTen drag. At $214.2m of Q4 Core AI revenue generating $51.8m of EBITDA, this is a business that would be profitable on a standalone basis at current scale. That matters as a proof-of-concept for the economics of each additional dollar of capacity that comes online.
I need to be precise about what is driving the GAAP operating loss, because the headline -103% op margin in Q4 is not what most people think it is.
D&A reached $180.7m in Q4, up from $99.0m in Q3 — an 82.5% QoQ increase. That single line, which is a non-cash accounting charge against assets already purchased and deployed, accounts for the majority of the GAAP operating loss. At $4,066m of FY25 CapEx being depreciated on a 4-year schedule, the math is simple: annualized D&A run-rate would be approximately $1.0B just from the FY25 asset base, before the $16-20B of FY26 CapEx even starts depreciating. GAAP EBIT is a delayed reflection of prior CapEx — it will not be meaningful until revenue substantially outpaces the depreciation curve.
There is a modest positive adjustment coming: starting Q1 FY26, Nebius is extending server and network depreciation from 4 years to 5 years. This is an accounting policy change, not a business change. But on the existing asset base — roughly $5.6B of PP&E at Q4 FY25 — it reduces annualized D&A by approximately 20%, or roughly $200m at current run-rates. That is a real tailwind to near-term GAAP metrics, achieved with zero impact on business performance.
The one-time SG&A charge of $43.6m in Q4 was described by management as non-recurring. I take that at face value but note it was not explained in detail in the available documents. Adjusted for it, Q4 GAAP op margin would have been approximately -84%, still deep in investment phase. I am not going to over-analyse a one-time charge. If it recurs, it becomes a credibility issue. For now, it is noise.
The real operating leverage question is whether opex beyond D&A is leveraging. Product development was 23.3% of revenue in Q4 FY25 vs 91% a year ago. SG&A (ex the one-time charge) would be approximately 51% of revenue, down from 243% a year prior. The non-D&A cost structure is clearly deleveraging — but that is what you expect in a company growing revenue 547% while the sales and R&D team grows more modestly. The trajectory is right.
The leading indicator set for Nebius is stronger than any name I follow in the portfolio right now. Let me be specific.
ARR of 1.25Bis5.5 × Q4annualisedrevenue(227.7m × 4 = $911m). The gap between ARR and annualised revenue is the contracted but not yet recognised revenue. That $339m delta, combined with $1.577B of deferred revenue on the balance sheet, gives approximately $1.9B of pre-contracted but not yet recognised revenue. That is 3.6× full-year FY25 revenue. The pipeline feeding FY26 is essentially visible.
ARR QoQ re-accelerated to +127% from +28% in Q3. Every quarter since IPO: +177% → +73% → +28% → +127%. The Q3 deceleration was a one-quarter pause driven by the gap between Microsoft contract signing and first tranche delivery. Watching that ARR metric in FY26 will be the primary indicator of whether the $7-9B year-end target is tracking.
Data centers: 1 in Q4 FY24 → 7 at Q4 FY25 → 16 sites announced/in pipeline as of Feb 12, 2026. The site expansion is the clearest operational signal. Each new site is a physical commitment of capital and landlord relationships that cannot be undone. Contracted power of 2GW+ already secured, with 3GW+ target for year-end 2026.
Q1 FY26 is already fully sold out. That is stated management commentary from the February earnings call. It is the single most important sentence in the entire earnings release. Pipeline creation trajectory for Q1 2026 was described as "on track to exceed $4B." If Q1 revenue prints anywhere near $400-500m — which the Q1 FY25→Q4 FY25 trajectory makes entirely plausible — it will confirm the FY26 guide is on track.
The management guidance track record deserves to be called by its proper name: it is a pattern of deliberately undemanding estimates. ARR: guided $1.0B three quarters in a row, then guided $0.9-1.1B, delivered $1.25B. CapEx: guided $1.5B then $2.0B for FY25, spent $4.1B. When management consistently under-guides, I assign less conservatism than the guidance range implies. For the $3.0-3.4B FY26 revenue guide and $7-9B ARR guide, I expect the actual to exceed the high end of both.
Volozh has a clean record now. EU sanctions removed, company fully domiciled in Amsterdam, no Russia ties since the Yandex asset sale. He is visible — DLD Munich, Bloomberg interviews, the Accel Growth Summit. His quarterly shareholder letter is direct and quantified. When he committed to positive group EBITDA in H2 FY25, he delivered it in Q4 FY25. When he set an ARR target of $1.0B, the company delivered $1.25B. I have no credibility concerns about Volozh as an operator.
The one concern I flagged in my January 2025 article — and which Atlas's scuttlebutt findings confirm is real — is the Russian-speaking cultural monoculture. Glassdoor data shows a company where meetings switch spontaneously to Russian and non-Russian-speaking employees are structurally excluded. This is a real ceiling on Western talent acquisition and a potential problem for enterprise and government contract pursuits, particularly in NATO-aligned institutions. It has not visibly impacted revenue — the Microsoft and Meta relationships speak for themselves — but it is a risk that compounds over time if not actively managed.
The other risk that I want to be clear about: the $16-20B FY26 CapEx commitment is not ordinary capital deployment. It is approximately 5-6× FY25 actual CapEx and roughly 5-6× FY26 projected revenue. Nebius is entering the year with $3.678B cash and 4.128Bofdebt.FCFwillbeapproximately−10B or worse in FY26. The company must raise capital continuously — and I mean continuously — throughout the year. If capital markets close, or if the cost of debt rises materially, the buildout schedule is at risk. The counterparties on the receivable side (Microsoft, Meta) are the safest credits in corporate America, which provides comfort on demand. But financing supply is entirely separate from demand.
I also think the analyst community is undervaluing the ClickHouse stake. 25% of a company valued at 15BintheJanuary2026SeriesDimplies 3.75B of NAV, sitting on Nebius's balance sheet at far less than that figure. Any monetization event — secondary sale, IPO — reduces the financing burden materially. No timeline has been signalled. I am watching.
For a company at Nebius's stage, the traditional EV/S vs. cohort methodology that I use for SaaS names is not the right frame. There is no meaningful cohort of public neocloud infrastructure companies at this growth rate. CoreWeave is planning an IPO at ~$40B; Lambda is private; both are at earlier stages. Nebius is genuinely sui generis among public names.
The correct frame is EV relative to contracted forward economics:
| Metric | Value | Assessment |
|---|---|---|
| Market Cap | ~$24.6B | As of ~Feb 25, 2026 |
| EV/TTM Revenue ($529.8m) | ~46x | Wrong lens — investment year |
| EV/FY26 Revenue guide midpoint ($3.2B) | ~7.7x | Reasonable for 500%+ grower |
| EV/FY26 Adj EBITDA guide (40% × $3.2B = $1.28B) | ~19x | Reasonable |
| EV/Ending ARR ($8B midpoint of $7-9B guide) | ~3.1x | Attractive |
| ClickHouse NAV (25% × $15B) | $3.75B | ~15% of market cap |
At 7.7× FY26 forward revenue for a business still growing at extraordinary rates with 70% gross margins and a 24% Core AI EBITDA margin, this is not expensive. The EV/ARR of 3× is the most compelling single number — it means the market is valuing Nebius at 3× the run-rate revenue it will be generating at year-end 2026, which is just one more year from now.
I need to be honest about what remains unanswerable in this valuation. The 40% group EBITDA margin guide for FY26 is based on Adj EBITDA, which excludes D&A. The actual D&A for FY26 will be enormous — $16-20B of new CapEx on a 5-year depreciation schedule adds $3.2-4.0B per year of D&A on top of the existing run-rate. GAAP EBIT profitability is at least two years away, probably more. For anyone who prices stocks on GAAP P/E, this is not the investment. For anyone who understands that infrastructure assets depreciate on schedule while the contracted revenue streams from those assets compound for years, the economics are sound.
The asymmetry: if Nebius executes on FY26 guidance — $3.0-3.4B revenue, $7-9B ARR, 40% EBITDA — this is almost certainly a multi-year multi-bagger from current prices. If capital markets close or demand inflects sharply, the financing math fails and the stock could give up most of its gains. I think the former is substantially more probable than the latter given who the demand counterparties are. But I acknowledge both scenarios.
The thesis I stated in January 2025 has played out better than I expected in 13 months. The "leap in the dark" has landed on a surface that looks increasingly solid. The $1.25B ARR beat, the first positive EBITDA, the 70% gross margin stability, the 170MW active power vs. 100MW target, the $1.577B deferred revenue book — these are not small data points. They are the infrastructure of a very large business beginning to emerge from its buildout phase.
I hold my 2,300 shares in the High Growth Portfolio. I am not adding at current prices because the risk profile has shifted: at $24.6B market cap, the position sizing is already appropriate, and the $16-20B FY26 CapEx commitment introduces a financing variable that I want to track for one more quarter before committing additional capital. If Q1 FY26 revenue prints $400m+ and management reaffirms the FY26 guide, I will consider adding. Current weight in the High Growth Portfolio: approximately 7% — at my concentration limit for a name with this risk profile.
For new subscribers considering a position: this is a 3-4% allocation name at current prices. Not more. The reward potential is exceptional. The financing risk is real. Size accordingly.
Bert Hochfeld | TickerTarget | Q4 FY25 Earnings Review | 2026-02-25 Sources: Scout brief NBIS_earnings-review_2026-02-23; Atlas baseline (2026-02-23); corpus (Jan 2025 initial analysis, Feb/Apr 2025 portfolio notes, Jun 2025 portfolio review)