Nebius Q4 2025 Earnings: AI Cloud Profit, ARR & Capex
Nebius Q4 2025 results show positive adjusted EBITDA, profitable AI cloud margins, power capacity buildout, $7–$9B ARR guide, and $2.1B capex risks.
## 1) The “miss” masks a much more important inflection: *real operating leverage is showing up*
Nebius’ latest disclosure shows Q4 2025 revenue of **$227.7 million** and **group Adjusted EBITDA turning positive at $15.0 million**—its “first quarter of positive adjusted EBITDA at the Group level.” ["Revenues"](0001104659-26-013946|1) ["15.0"](0001104659-26-013946|1) ["first quarter of positive adjusted EBITDA"](0001104659-26-013946|2)
The underappreciated signal is that, even while ramping infrastructure, **cost of revenues fell to 30% of revenue (from 60% a year ago)**—a clean read-through that fixed costs are being absorbed as utilization scales. ["as a percentage of revenues"](0001104659-26-013946|2) ["down from 60%"](0001104659-26-013946|2)
## 2) The “alpha between the lines”: AI Cloud is already profitable enough to subsidize the buildout
Management discloses that the **core AI cloud business generated $51.8 million of adjusted EBITDA at a 24% margin** in Q4—strong enough to offset losses elsewhere and still put the consolidated group in the black on an adjusted basis. ["adjusted EBITDA of $51.8 million"](0001104659-26-013946|2) ["adjusted EBITDA margin of 24%"](0001104659-26-013946|2) ["despite Avride and TripleTen"](0001104659-26-013946|2)
That mix shift is the hidden driver: the market tends to trade “consolidated revenue vs. consensus,” but the filings point to **unit economics in the core segment that can compound as capacity comes online** (and as non-core losses become proportionally smaller). ["core AI cloud business"](0001104659-26-013946|2)
## 3) Capacity is the real product—and Nebius is quietly positioning for “power secured” as a moat
In its CEO shareholder communication, Nebius lays out a capacity roadmap that reads like a supply-chain advantage: **nine new sites** across the U.S. and EMEA, **>2 GW of contracted power already secured**, an expectation of **>3 GW contracted power by year-end**, and **800 MW to 1 GW of connected power by end-2026**. ["nine new sites"](0001104659-26-013946|2) ["secured more than 2 GW"](0001104659-26-013946|2) ["more than 3 GW"](0001104659-26-013946|2) ["800 MW to 1 GW"](0001104659-26-013946|2)
This matters because in AI infrastructure, *the constraint is rarely demand*—it’s power + deployment execution. Nebius is effectively telling you its bottleneck is being addressed ahead of time. ["Capacity remains the enabler"](0001104659-26-013946|2)
## 4) The guidance “tell”: ARR is enormous, but the definition is aggressive (and that’s the point)
Management says it is **“on track to end the year with ARR of $7 billion to $9 billion”**—but also defines ARR as **last month of the quarter’s revenue × 12**. ["ARR of $7 billion to $9 billion"](0001104659-26-013946|2) ["revenue from the last month"](0001104659-26-013946|2)
Between the lines: they’re optimizing around *run-rate exit velocity*, not the trailing twelve-month story. If monthly revenue is accelerating as new capacity connects, the ARR framing can re-rate sentiment quickly—well before GAAP profitability catches up. ["calculated by taking revenue"](0001104659-26-013946|2)
## 5) The risk investors may be underweighting: capital intensity + controls/auditor transition
Nebius discloses **~$2.1 billion of Q4 capex**, largely GPUs and data center expansion—i.e., the model is scaling fast, but it is also **capital-hungry**. ["capital expenditures were approximately $2.1 billion"](0001104659-26-013946|2)
Separately, the company filed a same-day 6‑K on **changes in certifying accountant**, stating it will recommend **appointing Deloitte** for FY2026 and **dismissing Reanda** after completion of the FY2025 audit and Q1 2026 review (subject to shareholder approval). ["appointment of Deloitte"](0001104659-26-013947|0) ["dismissal and termination of Reanda"](0001104659-26-013947|0)
The key nuance: the filing also notes that Reanda’s report for FY2024 included an **adverse opinion on internal control over financial reporting**—a “fine print” item that can matter to institutional position sizing until it’s remediated. ["adverse opinion"](0001104659-26-013947|0) ["internal control over financial reporting"](0001104659-26-013947|0)
