Date: 2026-02-23 (updated from 2026-02-21; incorporates transcript digest, quant-prep, and scuttlebutt stage outputs) Quarter: Q4 FY25 (Dec-25) Market cap: ~$24.6B | EV/TTM Rev: ~47x | EV/FY26 Rev (midpoint): ~7.5x | Revenue growth: +547% YoY
Q4 FY25 was the quarter Nebius turned the corner: first positive group EBITDA (15m), ARRbeatguidanceby253.0-3.4B revenue, 40% Adj EBITDA margin, $7-9B ARR) is credible given the 1.58Bdeferredrevenuebookandknownhyperscalercontracts(Microsoft 18B TCV, Meta ~$3B TCV). The critical new disclosure: FY26 CapEx of $16-20B, a 3.2x raise from the prior $5B guide. This is the most important number in the report — it defines the capital model for the next two years and confirms FCF will remain deeply negative through FY26. Economic returns are a FY27+ story. Conviction: 4/5.
| Criterion | Threshold | Actual | Pass? |
|---|---|---|---|
| Revenue YoY growth | >30% | +547% | PASS |
| Gross margin | >60% | 70.0% [GAAP] | PASS |
| Revenue per quarter | >$50M | $227.7M | PASS |
| Data availability | 4+ quarters | 10 quarters | PASS |
| Share dilution | <10% annual | ~7% (238→253m shares over FY25) | PASS |
| GAAP profitability trajectory | Improving | EBITDA positive Q4; GAAP op margin -103% but +327pp YoY improvement | MARGINAL PASS |
Gate: PASS
| Factor | Rating | Detail |
|---|---|---|
| Growth | Strong | +547% YoY Q4; FY25 $529.8m vs FY24 91.5mrestated.CoreAIgrew 83049.8m → +41.0m → +81.6m. |
| Trajectory | Accelerating | QoQ re-accelerated: +55.9% vs +39.0% prior quarter. ARR QoQ re-accelerated sharply: +127% vs +28% in Q3. Both metrics confirm inflection. |
| Margins | Mid-High | Gross 70.0% [GAAP] — stable for 3 quarters post the 40→70% step-change in Q2 FY25. Group Adj EBITDA turned positive (+6.6%). Core AI segment at 24% Adj EBITDA margin. GAAP op margin -103% driven by D&A ramp; underlying unit economics expanding. |
| Dominance | Strong | #2 neocloud. First in Europe with NVIDIA GB300 NVL72 in production. Full-stack (managed Kubernetes, Postgres, MLOps) vs bare-metal peers. Vera Rubin NVL72 planned H2 2026. Sold out of capacity Q2-Q4 FY25 and Q1 FY26. |
| Valuation | Fair | 47x TTM revenue is not the right lens. At 3.2BFY26midpoint: 7.5xforwardrevenue, 19xforwardAdjEBITDA.Reasonablefora50010B+ in FY26. |
| Special | Present | 1.58Bdeferredrevenue(6.9xQ4revenue).Microsoft 18B + Meta ~3BincontractedTCV.ClickHousestake 3.75B implied (25% × $15B Series D). Q1 FY26 sold out. ARR has beaten guidance on every measurement since IPO. |
| Q323 | Q423 | Q124 | Q224 | Q324 | Q424 | Q125 | Q225 | Q325 | Q425 | |
|---|---|---|---|---|---|---|---|---|---|---|
| Cal date | Sep-23 | Dec-23 | Mar-24 | Jun-24 | Sep-24 | Dec-24 | Mar-25 | Jun-25 | Sep-25 | Dec-25 |
| Revenue ($m) | 5.0 | 6.7 | 11.4 | 14.5 | 32.1 | 35.2 | 55.3 | 105.1 | 146.1 | 227.7 |
| YoY % | — | — | — | — | +542% | +425% | +385% | +625% | +355% | +547% |
| QoQ % | — | +34% | +70% | +27% | +121% | +10% | +57% | +90% | +39% | +56% |
| Gross Margin % [GAAP] | — | — | — | — | — | 40.1% | 46.7% | 71.4% | 70.6% | 70.0% |
| Op Margin % [GAAP] | -1552% | -1396% | -727% | -774% | -251% | -430% | -234% | -106% | -89% | -103% |
| Adj EBITDA ($m) | -67.6 | -81.3 | -70.9 | -58.1 | -45.9 | -75.5 | -62.6 | -21.0 | -5.2 | +15.0 |
| EBITDA Margin % | — | — | — | — | — | -214% | -113% | -20% | -3.6% | +6.6% |
| Core AI Rev ($m) | — | — | — | — | 27.1 | — | 38.2 | 93.6 | 131.1 | 214.2 |
| Core AI EBITDA Mgn | — | — | — | — | — | — | — | — | 19.1% | 24.0% |
| CapEx ($m) | — | — | — | — | — | 417.6 | 544.0 | 510.6 | 955.5 | 2,056.0 |
| D&A ($m) | — | — | — | — | — | 33.3 | 49.2 | 75.2 | 99.0 | 180.7 |
| ARR ($m) | — | — | 31.8 | — | — | 90.0 | 249.0 | 430.0 | 551.0 | 1,250.0 |
FY25 full-year: Revenue $529.8m | CapEx 4, 066m|OCF+401.9m | FCF -$3,664m
Q1 FY25 revenue corrected to $55.3m (prior analysis used $50.9m from SA summary; 6-K is authoritative). All figures exclude Toloka (reclassified to discontinued ops Q2 FY25).
| Metric | Expected (from prior quarter guidance) | Actual Q4 FY25 | Verdict |
|---|---|---|---|
| Group Adj EBITDA | Positive in Q4 2025 | +$15.0m / +6.6% | MET — first in company history |
| ARR at Dec-2025 | 900m–1.1B | $1.25B | BEAT — +14% vs high end; systematic pattern of under-guidance |
| Active power | ~100MW (internal target) | ~170MW | BEAT — +70% above target |
| Revenue vs consensus | $242.8M | $227.7M | MISS -$15.1M (-6.2%); immaterial given ARR over-delivery |
| EPS (GAAP) | -$0.53 consensus | -$0.68 | MISS -$0.15; explained by D&A ramp ($180.7m, +443% YoY) + one-time SG&A ($43.6m non-recurring) |
| FY26 CapEx (prior analysis: "unknown — the critical missing piece") | Watched for disclosure | 16B–20B | DISCLOSED and materially higher than implied by prior run-rate guidance of $5B |
| Customer concentration | Watched for color | Microsoft ~18BTCV + Meta 3B TCV | CONFIRMED — two hyperscalers dominate contracted book |
Delta assessment:
The ARR beat (+25% vs guide high end) is the single most important positive. ARR has beaten guidance on every measurement since listing — this is systematic conservative guidance, not noise. The pattern: guide $1.0B, guide $0.9-1.1B, deliver $1.25B. For FY26, management guides $7-9B ARR. I would expect actual to land $8-10B.
The $16-20B FY26 CapEx is the most important new information. Prior analysis flagged the absence of this figure as the critical gap. Now it's disclosed and the number is extraordinary — 3.2x the prior 5Bguide, implying 4-5B/quarter in FY26. This has two simultaneous implications: (1) management has contracted demand sufficient to justify committing 16 − 20Bincapital, and(2)FCFwillbe−10B+ in FY26, requiring external financing at scale. Both are true. The question is whether the contracted demand is irrevocable. Given Microsoft and Meta are counterparties, credit risk is near-zero.
The revenue miss vs consensus (-6.2%) is noise. Quarterly reported revenue trails ARR by design (recognition over contract delivery schedules). The forward picture — $1.58B deferred + $7-9B ARR guide — matters far more.
D&A ramp explains the EPS miss structurally. $180.7m D&A in Q4 (+82.5% QoQ) as prior GPU cohorts begin depreciating. Starting Q1 FY26, depreciation extends from 4 to 5 years — this will reduce D&A approximately 20% on the existing asset base and is quietly bullish for near-term EBIT trajectory. It's a non-cash accounting change that directly improves reported operating metrics without any change in business performance.
One-time SG&A of 43.6minQ4isunexplainedinavailabledocuments; onthecallmanagementdescribeditasnon − recurring.Atfacevalue, adjustingforthis, Q4oplosswouldbe 191m and op margin ~-84% — materially better, still in investment phase.
Bullish divergence — strongest signal in the portfolio.
| Metric | Q4 FY24 | Q1 FY25 | Q2 FY25 | Q3 FY25 | Q4 FY25 | Trend |
|---|---|---|---|---|---|---|
| ARR ($m) | 90 | 249 | 430 | 551 | 1,250 | Re-accelerating after Q3 dip |
| ARR QoQ % | — | +177% | +73% | +28% | +127% | Sharp re-acceleration |
| Deferred Revenue ($m) | — | — | — | — | 1,577 | 6.9x quarterly revenue |
| Active Power (MW) | 100 | 100 | 100 | 100 | 170 | First ramp after 4 quarters flat |
| Data Centers | 1 | 2 | 4 | 4 | 7 | +3 in one quarter |
| Core AI EBITDA Margin | — | — | — | 19.1% | 24.0% | Expanding |
| Q1 FY26 pipeline ($B) | — | — | — | — | >$4B (mgmt) | On track |
The ARR-to-revenue divergence is the critical leading indicator. ARR 1.25Bannualisesto 1.25B/yr. Q4 reported revenue annualises to ~$911m. The $339m gap represents contracted but not yet recognised revenue. Add $1.577B deferred revenue on the balance sheet — approximately $1.9B of pre-contracted but unrecognised revenue, or 3.6x FY25 actual revenue.
Revenue will catch ARR with a 2-3 quarter lag. This makes the 3.0 − 3.4BFY26guidedirectlyderivable : itrequires 750-850m/quarter by Q4 FY26, which follows from $1.25B current ARR and $7-9B targeted ARR for Dec-2026. The math works.
ARR showed one deceleration in Q3 (+28% QoQ) that I flagged as a concern in the prior analysis — now resolved. The Q4 re-acceleration to +127% aligns with the Microsoft contract first tranche delivery in November 2025. Future tranches (H2 2026 predominantly) will drive additional ARR step-changes.
Customer breadth: 25+ named customers including Stanford (CRISPR-GPT research), Prime Intellect, Recraft (V4 model, "seamless transition from Hopper"), Cursor, Cloudflare, Shopify, HeyGen, World Labs, Black Forest Labs. SGLang team achieved 2x throughput on DeepSeek R1 on Nebius GPUs — published benchmark result that creates organic credibility in the developer community. However, the Microsoft (~18B)andMeta( 3B) contracts dwarf the long-tail entirely in dollar terms. The diversified customer list is real but economically marginal against hyperscaler concentration.
Technical credibility — high: First cloud globally running production GB300 NVL72 on 800 Gbps NVIDIA Quantum-X800 InfiniBand. MLPerf v5.1 submission for Blackwell configs. 442 GB/s write / 987 GB/s read on production storage. 10-25% performance improvement from topology-aware job scheduling. These are independently verifiable benchmarks. Vera Rubin NVL72 planned for H2 2026 — Nebius will be among the first deployers again.
Employee culture — persistent structural risk: Glassdoor 4.4/5 overall (44 reviews, 84% recommend) is misleading. Beneath the headline: meetings switch spontaneously to Russian, ~95% of employees are Russian-speaking, non-Russian speakers structurally excluded. Multiple reviews describe a "company says it wants to be international but does nothing to achieve it" dynamic. One titled "Toxic culture, bad managers" describes verbal conflicts. This creates a real ceiling on Western talent acquisition and is a credibility concern with certain enterprise and government customers (particularly NATO-aligned institutions that remain sensitive to Russian-origin firms).
Management credibility — strong operationally: Volozh is a proven builder. EU sanctions removed (formal public condemnation of Ukraine war). Microsoft relationship has a Yandex-connection layer (Mikhail Parakhin, former Yandex colleague, now at Microsoft) that creates key-person dependency risk. Volozh is highly visible and consistent in message across DLD Munich, Accel Growth Summit, Bloomberg appearances in 2025-2026. Guidance track record: ARR beaten every quarter since listing.
Analyst consensus: 6 Buy / 1 Hold / 0 Sell. Average PT 151( + 50101). Only bear: Morgan Stanley at Hold, $126 PT — execution uncertainty. No sells despite $16-20B CapEx guide and deeply negative FCF. Compass Point Strong Buy $150, Citizens Buy $175. The absence of sells is itself a signal.
Feb 12, 2026 pre-announcement: 9 new data center sites disclosed (US: Missouri, Alabama, Oklahoma, Minnesota; EMEA: France, Israel, UK). Total portfolio = 16 sites globally. $1.15B equity raise. Volozh preemptively building the site pipeline ahead of Q1 FY26 reporting. This is management signalling conviction, not desperation.
Competitor context: CoreWeave is 4-5x larger in revenue (~$1B/quarter vs 228m), has 33datacentersvs7, 420MWactivepowervs170MW.CoreWeaveplannedIPOin2026at 40B valuation. SemiAnalysis places CoreWeave in Platinum performance tier, Nebius in Gold. The scale gap is real and not closing in 2026. Nebius's response: differentiate on full-stack managed services, European positioning, and self-service developer experience (32 GPU self-service, no quota approval).
| Metric | Current | FY26 Guide (Midpoint) | CoreWeave (est.) | Assessment |
|---|---|---|---|---|
| EV/TTM Revenue | ~47x | — | ~12-15x (post-IPO) | Rich on TTM; wrong lens |
| EV/Forward Revenue | — | ~7.5x ($3.2B) | ~8-12x | Fair |
| EV/Forward Adj EBITDA | — | ~19x ($1.28B at 40% mgn) | ~20-25x | Fair |
| EV/Forward ARR | — | ~3x ($8B midpoint) | — | Attractive |
| ClickHouse NAV | $3.75B implied | — | — | ~15% of market cap in single stake |
| Market cap | ~$24.6B | — | ~$40B (CRWV) | — |
The valuation paradox: 47x TTM looks extreme but FY25 is the infrastructure ramp-up year, not steady state. The right frame is forward. At 3.2BFY26revenueand401.28B): ~7.5x forward revenue, ~19x forward EBITDA. For a 500%+ grower, that's not expensive.
The critical caveat: The 40% Adj EBITDA margin excludes D&A. D&A will be enormous in FY26 — $16-20B of new CapEx depreciating on a 5-year schedule generates $3.2-4.0B/year of D&A, growing as prior cohorts also depreciate. GAAP EBIT will remain deeply negative through FY26. FCF will be -$10B+. This is an infrastructure build, priced on the FY27-2030 steady-state, not on near-term economics. That's appropriate — but requires sustained execution and financing access.
ClickHouse as hidden NAV: 25% stake in ClickHouse at 15BSeriesDvaluation= 3.75B implied. That's ~15% of Nebius's market cap sitting in a minority stake not consolidated on the P&L. Monetisation (secondary sale or IPO) would materially reduce the financing burden.
Extreme capital intensity with negative FCF through FY26. $16-20B FY26 CapEx vs 3.2Brevenue = 5 − 6xCapEx/revenueratio.Netdebtalreadynegative(449.6m net debt at Q4). FY26 will deepen this substantially. Financing access is existential — any capital markets disruption threatens the buildout schedule.
Hyperscaler customer concentration is both the thesis and the tail risk. Microsoft (~18BTCV)andMeta( 3B TCV) are responsible for the vast majority of contracted revenue. These same companies are Nebius's direct competitors in cloud infrastructure. Contract non-renewal, pricing renegotiation, or substitution with in-house capacity post-initial term would be catastrophic. Diversification is happening but the long-tail is economically immaterial relative to hyperscaler concentration.
D&A will overwhelm GAAP EBIT for years. D&A reached $180.7m in Q4 (+82.5% QoQ). At $16-20B FY26 CapEx on a 5-year depreciation schedule, D&A could reach $4-5B annually within 2-3 years — far exceeding operating income. GAAP EBIT profitability is a 2027-2028 event at the earliest. Management's 20-30% medium-term EBIT target requires revenue to outrun the D&A curve.
Russian-centric culture caps Western talent and enterprise credibility. Glassdoor pattern is clear and consistent: structural language exclusion, monoculture, poor work-life balance. Enterprise sales and government contract pursuits require Western-facing credibility that the current culture actively undermines. Fixes require years and management priority — neither is demonstrated.
CoreWeave scale gap and planned IPO. CoreWeave is 4-5x larger, better-capitalised, and planning a public market listing that will provide additional capital. A fully-capitalized CoreWeave competing for the same enterprise and hyperscaler contracts narrows Nebius's market share opportunity. Nebius's full-stack differentiation is real but untested at enterprise scale against a Platinum-tier competitor.
Q1 FY26 revenue print. Needs to land ~$400-500m to confirm FY26 guide trajectory. Q1 is already fully sold out. A print at or above $400m eliminates execution doubt on the guide.
ARR milestones toward $7-9B. Microsoft majority delivery in H2 2026 will drive ARR step-changes. Each delivery tranche announcement is a catalyst. A Q2 or Q3 ARR print above $4B confirms the guide is on track.
ClickHouse monetisation event. Secondary sale or IPO at any valuation above the 15BSeriesDreferencewouldcrystallise 3.75B of NAV and reduce financing pressure. No timeline signalled — watching for any disclosure.
NVIDIA Vera Rubin NVL72 deployment (H2 2026). Being among the earliest deployers of next-gen silicon reinforces the technical differentiation narrative and supports premium pricing vs commodity GPU providers.
EU AI policy / government contracts. EU AI Act enforcement, data sovereignty mandates, and national AI funding (Israel pipeline active) create structural demand for non-US cloud providers. Any major government contract win would validate the European moat as a commercial revenue driver, not just a positioning claim.
Atlas note: Prior analysis (2026-02-21) flagged "no FY26 CapEx guidance" as the critical missing piece. That piece is now disclosed: $16-20B. It confirms both the scale of ambition and the scale of the financing commitment. Q1 FY25 revenue also corrected from $50.9m to $55.3m per the 6-K (SA summary error). wsm007 holds NBIS at approximately 10.5% portfolio weight.
Generated by Atlas | 2026-02-23 | Source: Scout brief NBIS_earnings-review_2026-02-23 + scuttlebutt/quant-prep/transcript-digest stage outputs