Nebius Has a Demand Problem? No — It Has a Delivery Date.
Q3 revenue rose 571%, but a New Jersey power delay trimmed 2026 guidance and management retired the ARR metric it once called its north star.
NBIS · Earnings Call · 2026-11-12
The quarter the story changed
Nebius’s third quarter was, on the surface, another demand vindication. Revenue reached $980.8 million, up 571% year over year; adjusted EBITDA margin hit 47.5%; and management said Q3 agreements closed above $25 million per megawatt, up from $20 million to $25 million in Q2. Yet the stock fell more than 15% intraday, and the reason wasn’t demand. It was the calendar. The company trimmed 2026 revenue guidance to $3.0 billion-$3.2 billion from $3.0 billion-$3.4 billion because third-party colocation capacity — notably New Jersey — is late. The market had already been brutal: shares were 46% below their high, according to coverage around the report. The debate has shifted from “is AI demand real?” to “can Nebius physically deliver the megawatts it has sold?”
Arkady Volozh framed the tension sharply on the call:
The demand for AI is real and substantial. Everything else, like building the capacity and financing it is just a task. Tasks can be hard, but they all can be solved.
That is the right framing for the business. It is also the exact framing that gets tested when a task slips.
Prepayments crossed a threshold
The genuinely new development is not revenue growth. It is the financing structure. Management said prepayments on new contracts now cover more than 100% of the capital needed to bring contracted capacity online, up from 50%-60% last quarter. More than 90% of Q3 contract value carried a prepayment, and the weighted-average prepayment share rose about 20 percentage points from Q2. Payback shortened to 1 year 7 months from 1 year 10 months. That is a different business model from the one investors were underwriting a year ago. It means the build-out is increasingly funded by customers, not by the equity or debt markets.
“The most important change is that the prepayments on new contracts now cover more than 100% of the capital needed to bring this contracted capacity online. Just to remind you, last quarter, it was 50% to 60%.” — Arkady Volozh, CEO · 2026-11-12
This matters because the global tape is voting on exactly the opposite risk. In the 30-day window, the global AI data center theme is deeply negative, with 13 advancers against 57 decliners across the ticker set. The market is not paying up for AI infrastructure exposure; it is discounting the financing and delivery risk. Nebius’s answer is investment grade contracts and customer customer prepayments, which it says can support asset-backed financing at attractive terms. Dado Alonso framed the 2027 plan as 75%-85% of a $40 billion-$45 billion CapEx budget funded by prepayments and asset-backed financing, with $17 billion-$20 billion of prepayments expected next year. If that holds, Nebius is less dependent on the capital markets than the neocloud peer group. If it doesn’t, the prepayment story becomes a marketing line rather than a funding shield.
The delivery problem is real — and it was flagged before
The guide-down is not mysterious. Andrey Korolenko said the Vineland site’s power plan changed when DataOne decided to wind down temporary generation faster than expected, creating a gap until Bloom fuel cells ramp. The company says it is expediting Bloom generation and that the power is now on site, but some capacity has slipped into 2027. What makes this a story is the contrast with the prior call. In August, management was asked directly about Vineland and the New Jersey hearing. Andrey said then:
“overall, the switch to Bloom has been a valuable and a good pivot for the project, we believe, and with no significant impact expected on the project time line.” — Andrey Korolenko, Head of Construction or Operations (inferred from contract delivery and capacity discussions) · 2026-08-12
Three months later, the same site is central to a guidance cut. On the current call, Andrey was more candid:
“In late Q3, DataOne decided to wind down the temporary generation faster than it was expected. We are expediting the Bloom deployment as much as possible, but we still will have a gap due to this situation.” — Andrey Korolenko, Management Team Member · 2026-11-12
That is not a demand problem. It is an execution and disclosure problem. Management is now leaning harder into owned sites — Alabama, Missouri, two projects in Finland, Pennsylvania — where it controls land, power, shell, hardware, and operations. The next gigawatt is mostly own-site, which should reduce third-party dependency. But investors have just been reminded that “mostly” is not “entirely,” and that power generation is the hard part.
The metric retirement
Perhaps the most revealing sentence on the call was not about capacity. It was about ARR. Dado said Nebius will stop guiding to annualized run-rate revenue after the February 2027 update:
“After that update, we will no longer guide to ARR as our business is scaling into the next phase. This metric does not reflect the full scope of our base.” — Dado Alonso, CFO · 2026-11-12
That is a notable reversal. In February, Dado called ARR the “north star”:
“as we are in hypergrowth phase, ARR is the north star metric.” — Dado Alonso, Executive (likely CFO or similar finance leadership) · 2026-02-12
Now the company is retiring it. The stated reasons are reasonable: ARR is a one-month snapshot, and the asset light model can add choppiness. But investors should notice what is happening. Nebius is asking the market to value it on revenue and pretax earnings instead of a metric it spent several quarters teaching the Street to watch. That may be maturation. It may also be a signal that ARR is becoming less flattering as the business scales and as one-time partnership revenue enters the base. The company still expects ARR to exceed $7 billion for February, but after that, the training wheels come off.
What to watch
The bull case is now simple: demand is not the constraint; delivery is. Revenue is guided to roughly quadruple to $12 billion-$13 billion in 2027, with pretax breakeven in Q4 2027. CapEx is $40 billion-$45 billion, and contracted power is expected to reach 5 gigawatts by year-end, up from more than 4 gigawatts today. The bear case is equally simple: the company just missed on timing at a third-party site, and the next leg depends on owned-site execution at a scale it has never delivered. The stock’s drawdown suggests expectations have already reset. The next catalyst is not another multibillion-dollar contract. It is proof that the New Jersey tranches are ramping and that the owned sites hit their dates. Until then, the prepayment model is a strong answer to the financing question — but it cannot solve the calendar.