NBIS — Earnings Review Q4 FY25 (muji)

Date: 2026-02-22 Quarter: Q4 FY25 (Dec-25) Long NBIS ~10.5% (wsm007 portfolio) Atlas baseline: Conviction 4/5, Fair valuation, Exceptional growth with execution risk



First Write-Up Note

No prior muji corpus on NBIS. This is the founding take. Grounding everything in the Q4 FY25 data.


The Setup: What Is Nebius, Actually?

Here's the thing most people get wrong about NBIS. They look at "GPU cloud" and think commodity. That's the wrong lens.

Nebius is building a full-stack AI cloud — compute (H100/H200/Blackwell), storage, managed Kubernetes, managed PostgreSQL, MLOps tooling. When an AI startup lands on Nebius, they're not just renting GPUs. They're embedding into a platform. That's a building block, not a process tool. The customer's AI product is built on top of Nebius primitives. That's the architecture that scales.

Compare to CoreWeave — which is excellent but fundamentally bare-metal GPU leasing with a thin layer on top. CoreWeave is bigger (~10x GPU fleet) but Nebius is building something stickier. The Tavily acquisition (agentic search, ~$400m, Feb 2026) signals they want to own the application layer too. That's optionality I'm watching carefully.

"Scale in platform" check: YES. GPU infra expands naturally into managed services → MLOps → AI application primitives. The TAM expands with every layer added.


The Numbers

Revenue

Q125 Q225 Q325 Q425
Mar-25 Jun-25 Sep-25 Dec-25
Revenue ($m) 50.9 105.1 146.1 227.7
QoQ % +44.6% +106.5% +39.0% +55.8% ^^
YoY % +347% +625% +355% +547% !!
Core AI Rev ($m) 38.2 93.6 131.1 214.2
Core AI % of Total 75% 90% 94% ^^

The QoQ re-acceleration from +39% in Q3 to +55.8% in Q4 — that's the ^^ notation. Dollar adds: +15.7 → +54.2 → +41.0 → +81.6. Largest sequential dollar add in company history.

FY25 total: ~530mvsFY24 93m = +468% full-year growth !!

Margins & Profitability

Metric Q325 Q425
Gross Margin [GAAP] 70.0% !!
Op Margin [GAAP] -89% -103%
EBITDA ($m) -5.2 +15.0 !!
Core AI Adj EBITDA ($m) 25.1 51.8
Core AI Adj EBITDA Margin 19.1% 24.0% ^^
EPS [GAAP] -$0.40 -0.68(miss0.14)
OCF ($m) +$834 !!
FCF ($m) -$1,221.7

The GAAP op margin tick from -89% to -103% is noise — the revenue scaling through the P&L is just behind the cost recognition from the Q4 CapEx surge. The real signal is the EBITDA flip to positive (+$15m) and Core AI Adj EBITDA at $51.8m (24% margin, up from 19.1%).

70% gross margin on GPU cloud is extraordinary. This is not commodity economics. This is a managed platform with margin characteristics closer to software than hardware.

Leading Indicators

Metric Q424 Q125 Q225 Q325 Q425 Trend
ARR Run-Rate ($m) 90 249 430 551 1,250 ^^ !!
Deferred Revenue ($m) 1,577
Data Centers 1 2 4 4 7 !!
Active Power (MW) 100 100 100 100 170 stepped
Core AI EBITDA Margin 19.1% 24.0% ^^

The ARR trajectory is the most important data series in this earnings report. $90m → $249m → $430m → $551m → $1,250m !!

That's not linear. The ARR added in Q4 alone ($699m) exceeds total ARR just 6 months ago. There is something very different happening in Q4 — almost certainly a major contract or wave of enterprise wins. Without the transcript I can't confirm what drove it, but the magnitude is extraordinary.

NUANCE: ARR run-rate of 1, 250mvsannualisedQ4reportedrevenueof 910m is a $340m gap. Add in $1,577m of deferred revenue (pre-paid contracts awaiting recognition). This is the most bullish data point in the quarter: contracted demand is materially ahead of reported revenue. Revenue is a lagging indicator here. ARR + deferred revenue is the leading truth.

CapEx & Balance Sheet

Metric Q424 Q125 Q225 Q325 Q425
CapEx ($m) 418 544 511 956 2,056
Cash ($m) 3,678
Debt ($m) 4,127

The CapEx doubling Q-over-Q from $956m to $2,056m is the elephant in the room. **Net debt position: 450m. * *Q4FCF : −1.2B. This is aggressive capital deployment.

Here's the thing though — this CapEx is demand-driven, not speculative. The deferred revenue ($1.577B) represents customer money already received. You don't deploy $2B of GPUs speculatively — you deploy to fulfill signed contracts. That's a crucial distinction.

The open question (transcript pending): Does CapEx moderate in 2026? FY26 guide of 40% Adj EBITDA at 3.2Brevenuemidpointimplies 1.28B Adj EBITDA. That only works if CapEx steps down materially from Q4 run-rate. Management needs to clarify.

FY26 Guidance

Metric FY26 Guide
Revenue $3.0B - $3.4B
ARR (year-end) $7.0B - $9.0B
Adj EBITDA Margin 40%

The ARR guide ($7-9B) is the remarkable number here. From $1.25B exit ARR to $7-9B exit ARR in one year. That implies signing contracts worth 5-7x current ARR in the next 12 months. If real, this is a company about to undergo a second gear shift.

My stance on the guide: Conservative on revenue relative to what the ARR ramp implies. $3.2B midpoint is roughly 5x Q4 annualised run-rate — achievable if the Q4 power and data center capacity ramps come online and deferred revenue recognises as planned. The ARR guide is the bolder bet.


Platform Assessment

Architecture: Cloud-native. GPU-first but with managed services layered above compute. Full infrastructure stack — not legacy, not on-prem, not Windows. Clean.

GTM: Self-serve (AI-native startups) + enterprise direct + national AI projects. No SI dependency. That's the right architecture for scaling unit economics.

Competitive moat:

  1. NVIDIA Cloud Partner (1 of 5) — exclusive access to latest hardware (GB300 NVL72 live in Finland first in Europe !!). This is a supply chain moat most competitors cannot replicate.
  2. Full-stack differentiation — managed Kubernetes, PostgreSQL, MLOps vs CoreWeave's GPU leasing. Stickier customer relationships.
  3. European geographic moat — EU AI sovereignty tailwind. Non-US hyperscaler with European data centers. This is an underappreciated angle.
  4. Ex-Yandex engineering depth — world-class infrastructure engineers who built systems at massive scale. Glassdoor 4.4/5, 84% recommend.

Tavily acquisition (~$400m, Feb 2026): Agentic search. This is muji's radar. If Nebius embeds agentic search primitives into its platform, AI developers get a "building block" they can call from their ML pipelines on Nebius infrastructure. That's platform expansion into the application layer. Worth monitoring — if this turns into platform revenue rather than standalone product revenue, it's a meaningful signal.

"Building block" or "process tool"? Building block. GPU compute + managed ML infra is embedded in customer applications. Customers don't switch cloud infrastructure mid-training run.

Tier Classification: Tier 1. 547% growth, 70% GM, accelerating ARR, EBITDA positive at core level, strong leading indicators. Full conviction threshold on growth.


NUANCE

A few things the market may be missing:

1. The ARR-to-revenue gap is structured, not accidental. 1.25BARRvs 910m annualised revenue isn't a reporting quirk. It's evidence of a ramp structure in customer contracts — capacity is being signed before it's live. As the 170MW→1GW power build-out comes online, contracted revenue will recognise. The deferred revenue ($1.58B) is the cash evidence that this is real.

2. Q4 CapEx surge = fulfilling contracts, not speculation. Infrastructure companies that deploy capital ahead of demand go bankrupt. Infrastructure companies that deploy capital to fulfill signed contracts go to $100B+ market caps. The question is which is happening here. The deferred revenue being 7-8x quarterly revenue strongly suggests the latter.

3. European AI sovereignty is a genuine structural tailwind. The EU AI Act, data localisation requirements, and government digital sovereignty programs favour non-US AI infrastructure. Nebius is the only full-stack GPU cloud company headquartered outside the US with major European data centers. That's not a coincidence — that's a market positioning decision with multi-year tailwinds.

4. The EPS miss isn't the story. -0.68vs0.54 consensus. Without the transcript, reasons are unclear. But NBIS isn't a company where EPS matters right now. Revenue growth, ARR, gross margin, and EBITDA inflection are the operating metrics. EPS miss could reflect CapEx depreciation timing, financing costs on the debt, or one-time items. I want to hear management's explanation but I'm not adjusting my view on it.

5. No NRR data. This is a critical gap in muji's standard toolkit. Without net revenue retention, I can't assess expansion within existing customers vs. new customer acquisition. Given the ARR trajectory, my working assumption is NRR is strong — you don't grow ARR 14x year-over-year purely from new logos. But I need this data. Adding to watch list for 10-Q.


Thesis Status

Thesis: Strengthening.

Nebius entered Q4 as a promising but early-stage AI infrastructure play. Q4 results — 547% growth, EBITDA positive, 70% GM, $1.25B ARR, $1.58B deferred revenue, 7 data centers, GB300 NVL72 live — confirm the thesis is executing.

What would change my view:

My stance: Hold full position. If Q1 FY26 revenue prints $400m+ and CapEx guidance moderates from Q4 levels, this is a top-conviction add. The ARR ramp and deferred revenue are real. The platform architecture is differentiated. The European moat is underappreciated.

Tier 1 growth at what is increasingly reasonable forward valuation (~7.8x FY26 guided revenue). IMHO the risk here is mostly upside risk — if ARR converts as guided ($7-9B by year-end), we're looking at a potential $50-100B company.

-muji

Long NBIS ~10.5%


Promises to Track

Promise Made Status
FY26 revenue $3.0-3.4B Q4 FY25 PR Watching
FY26 ARR $7-9B Q4 FY25 PR Watching
FY26 40% Adj EBITDA margin Q4 FY25 PR Watching — CapEx key variable
Group EBITDA positive H2 FY25 Q4 FY24 CC Achieved (Q4 +$15m) !!
Core EBITDA positive Q3 FY25 Q4 FY24 CC Beat by 1 quarter (Q2)
FY25 ARR 750m1B Q4 FY24 CC Beat ($1.25B actual)

Management has a strong promise-keeping track record. Every FY25 commitment was met or beaten. That earns credibility for the FY26 guide — but the ARR target ($7-9B) is 5-6x bolder than anything they've committed to previously.


What I'm Watching Next

  1. Transcript (expected ~2 weeks): CapEx guide for FY26. Customer concentration. ARR ramp structure. EPS miss explanation.
  2. Q1 FY26 revenue: If $400m+, this is a monster. If $250-300m, the deferred revenue is taking longer to recognise than guided.
  3. NRR disclosure: Need to know expansion dynamics within existing customers.
  4. Tavily integration: Is this a platform building block or a standalone bolt-on?
  5. NVIDIA Rubin timeline: GB300 NVL72 already live. Rubin (H2 2026) is the next hardware gen. First-mover access = competitive moat.

Generated: 2026-02-22 | Quarter: Q4 FY25 | Type: earnings-review