Imagine reporting the best quarter in your company’s history — revenue up more than fivefold, profits swinging from a loss to nearly half a billion dollars — and watching your shares fall 16% before lunch.
That is exactly what happened to Nebius on Thursday. The Amsterdam-based AI cloud provider, backed by a $2 billion investment from Nvidia, delivered third-quarter numbers that beat every forecast on Wall Street. The market’s answer was a sell-off.
The paradox says something important about where the AI boom stands in late 2026. Growth is no longer enough. Investors now want proof that the growth can be delivered on time — and paid for.
The Numbers Behind the Nebius Q3 Earnings Beat
Nebius reported third-quarter revenue of $980.8 million, according to CNBC. That is a 571% jump from the $146.1 million it made in the same quarter a year ago, and comfortably ahead of the $907.5 million analysts had expected. The company’s full third-quarter results were published on Thursday morning.
The surge was not a one-quarter fluke. Revenue for the first nine months of 2026 reached $1,962.1 million, up 549% year on year. Nebius is now a billion-dollar-a-year business in all but name, and it got there in a fraction of the time most companies take.
Profitability improved just as sharply. Adjusted EBITDA — earnings before interest, tax, depreciation and amortisation — came in at $465.8 million, swinging from a loss of $5.2 million in the same period last year.
The net loss narrowed to $37.5 million, or 13 cents a share, from $119.6 million a year earlier. On an adjusted basis, the company swung to a net profit of $90.9 million, far better than the $271.4 million loss the market had pencilled in.
Even the spending tells a story of a company scaling at full tilt. Capital expenditure in the quarter totalled about $7.3 billion, a 664% jump from a year ago, as Nebius raced to build the data-centre capacity its customers are demanding.
Why the Shares Fell Anyway
If the numbers were this strong, why did the stock drop? The answer was in the guidance — the part of the earnings release that looks forward, not back.
Nebius said full-year 2026 revenue would now land at the low end of its previously announced $3 billion to $3.4 billion range. It also pushed back its annualised revenue run-rate target: instead of reaching $7 billion to $9 billion by December 2026, it now expects to pass $7 billion by February 2027.
Chief executive Arkady Volozh went further, telling investors that annual recurring revenue is no longer the right metric for measuring the business. For a company whose entire pitch rests on explosive recurring revenue, that was a difficult message to hear.
The shares fell about 10% in premarket trading, then dropped more than 16% by Thursday’s close, according to CNBC. It was not an isolated move: CoreWeave, the rival AI cloud provider whose own revenue jumped 168% this quarter, fell about 10% in premarket trading the same day.
The two sell-offs landed in a week when the whole AI trade was under pressure. The Federal Reserve had warned days earlier that AI-related debt could trigger a large stock market correction, and Nvidia’s chief executive Jensen Huang had just resigned after a board investigation. Nebius walked into a market already looking for reasons to worry.
Data Centres Running Behind Schedule
The immediate cause of the guidance cut was capacity. Nebius said some third-party data-centre capacity it had expected to switch on this year is now scheduled for the first half of 2027.
That delay matters because Nebius sells computing power, and it can only sell what it has built. Revenue it had expected to book in December will now arrive in February instead. The company stressed that the contracts behind that capacity remain in place — the demand has not gone away, only the timing has slipped.
It is a familiar problem in the AI infrastructure race. Everyone is building at once, and power, land and hardware are all in short supply. A data centre that was supposed to open in November can easily become a February opening when a single permit or transformer delivery runs late.
Volozh tried to reassure investors on the quality of the business behind the delay. New contracts signed in the third quarter, he said, came with longer terms, higher prices and bigger upfront payments from customers. The company also expanded its customer base, adding names such as Databricks, Datadog, Palantir and Snowflake.
The $27 Billion Spending Plan
The second shock in the release was the scale of Nebius’s spending plans. The company now expects capital expenditure of about $27 billion in 2026, up from its earlier forecast of $20 billion to $25 billion.
And the bill keeps growing. Nebius gave an initial 2027 capital expenditure outlook of $40 billion to $45 billion, as it builds capacity for next year — in particular at sites it owns outright rather than rents.
That is an enormous sum for a company that made less than $2 billion of revenue in the first nine months of the year. It is also the logic of the neocloud business model: spend billions on Nvidia chips and data centres now, then rent them out for years at a premium.
The bet only works if the customers keep coming — and, crucially, if they keep paying upfront. Volozh’s point about longer contracts and larger advance payments was aimed squarely at that worry. If customers are willing to pay in advance, the spending is funded by demand rather than debt.
Nebius also gave a first 2027 revenue outlook of $12 billion to $13 billion — roughly four times its 2026 total — and said it expects to reach pre-tax breakeven by the end of 2027. Those are bold numbers, but they depend on the same capacity arriving on schedule.
What It Means for the AI Infrastructure Boom
Nebius is not a household name, but it has become one of the purest ways to measure the AI build-out. Nvidia invested $2 billion in the company in March, and its customers are the same AI labs and software firms driving the boom.
Its results therefore read like a health check on the whole sector. The demand side is not the problem: revenue up 571%, customers signing longer and more expensive contracts, a pipeline that stretches into 2027.
The problem is execution and financing. Building data centres is slow, expensive and getting more so. Every neocloud provider — Nebius, CoreWeave and their peers — is spending tens of billions to keep up, and each quarter’s results are judged as much on the spending as on the sales.
That is why a 571% revenue jump and a 16% share-price fall can happen on the same day. The market already believed Nebius could grow. What it wanted to know was whether it could grow on time and on budget — and Thursday’s answer was not quite good enough.
What Happens Next
Nebius executives were due to discuss the results on an analyst call on Thursday morning, where the capacity timeline and the 2027 spending plan were expected to face the toughest questions.
The next real test comes in February, when the company now expects to cross the $7 billion annualised revenue run rate it had originally promised for December. If the delayed capacity switches on as planned, the December shortfall will look like a timing issue. If it slips again, investors will start asking harder questions about execution.
The fourth quarter will also show whether the spending plan is sustainable. A $27 billion capital bill in 2026, followed by $40 billion to $45 billion in 2027, needs either very strong operating cash flow or very willing lenders. The upfront payments Volozh highlighted will be closely watched.
For now, Nebius remains one of the fastest-growing companies in the AI economy — a business that multiplied its revenue nearly sevenfold in a year and turned profitable on an adjusted basis. Thursday’s sell-off does not change that record. It does show, though, that in the AI infrastructure race, growing fast is no longer the hard part. Building fast enough, and paying for it, is.
