Nebius reported $582 million in revenue for the second quarter, up 454% from a year ago and ahead of the roughly $567–570 million Wall Street had modeled, while the per-share loss narrowed to about $0.12 against a consensus around $0.56. In an AI-infrastructure name trading off a growth rating, that is the kind of print that makes the story easy to tell. The harder question is the one the headline skips: is this revenue one-off momentum, or is it sitting on a signed base that can keep feeding it? That question, not the print, is what decides how much weight the upside deserves.
Nebius tracked below consensus in 2025-Q3 and Q4-2025, then beat by a widening margin in Q1-2026 and Q2-2026, an accelerating contract-backed revenue beat.
| Period | Reported revenue | Consensus |
|---|---|---|
| 2025-Q3 | 146.1 | 157.9 |
| 2025-Q4 | 227.7 | 247.3 |
| 2026-Q1 | 399 | 375.1 |
| 2026-Q2 | 582.3 | 567.2 |
The clue that this is more than momentum is how fast the streak flipped. Through late 2025, NebiusNBIS-- kept missing expectations — $146 million against a $158 million consensus in Q3, $228 million against $247 million in Q4. Then the beats turned and widened: $399 million against $375 million in the first quarter of 2026, and now $582 million against $567 million. Same growth story, different reception. What changed is that the revenue started arriving on schedule, which is what a contracted base, rather than a hope, tends to produce.
Management says the booked support is enormous. It reaffirmed a full-year 2026 target for exit annual recurring revenue (ARR — the run-rate at which booked recurring revenue is flowing, annualized) of $7–9 billion, and points to a contracted backlog exceeding $40 billion, with some secondary coverage citing roughly $46 billion. For an ordinary reader the scale is hard to hold, so anchor it in the signed names: Microsoft's relationship is estimated at $17–19 billion, Meta's agreement is worth up to $27 billion, and Nvidia committed to more than 5 gigawatts of systems through 2030 alongside a roughly $2 billion strategic stake. That is the list that turns "revenue upside" from a theme into a set of obligations with counterparties.
But a backlog is not revenue, and the mechanism is where the discipline lives. The chain runs from contracted power to connected power to utilization and pricing, and only then to billed revenue, clearing at the year-end ARR number. The gap is visible in the capacity numbers. Nebius had more than 3.5 gigawatts of power contracted in the first quarter and guided above 4 gigawatts by year-end, yet it targets just 800 megawatts to 1 gigawatt of actually connected power by then. Most of what is signed is not yet billed; it is future capacity that has to be built, permitted, interconnected, and supplied with GPUs before it shows up in revenue. At the high end of guidance, that implies roughly $9.9 million of revenue per megawatt of connected capacity — a useful measure of whether pricing and utilization hold as the machine scales, or erode as fresh, lower-use capacity comes online.
That is why the single number that settles the contract-backed thesis is not the $582 million quarter. It is the exit ARR at year-end. If Nebius lands inside its own $7–9 billion range, the booked-to-billed conversion is happening, and the recurring-revenue story gains the signed basis the current multiple keeps asking for. An exit below $7 billion would signal that bookings are converting into revenue more slowly than modeled, and the growth premium would be resting on intent rather than delivery.
None of this is guaranteed, and the evidence carries its own hedges. The backlog and power figures are cumulative, forward commitments drawn largely from first-quarter and guidance material, not additions from the second quarter — a consistent range, not a fresh haul. The roughly 50% EBITDA margin on the AI-cloud segment is a disclosed segment figure with no GAAP reconciliation supplied, and the full company still lost money (operating margin around minus 50%) on a build guided to roughly $20–25 billion of capex this year. Permitting, interconnection delays, and GPU supply are the execution risks that could turn signed capacity into an idle asset. That is the honest shape of the trade: the upside is contract-backed, not narrative, but it is contract-backed promise, and the promise only pays out if the year-end exit ARR lands inside the range management set for it.
Comments
No comments yet