A Cloud Builder Sells 2027 Before It Finishes 2026

Nebius published a number on Nov. 12 that most companies of its age would consider impolite: revenue of $980.8 million for the third quarter, up 571% from a year earlier. Almost all of it, 99%, came from renting out artificial-intelligence computing. Adjusted earnings before interest, taxes, depreciation and amortization ran at a 53% margin. The company still lost $37.5 million.

Both facts are accurate, and the distance between them is the story. A business can report a 53% adjusted margin on roughly a billion dollars of quarterly revenue and a net loss in the same table when the buildout it is financing carries depreciation and interest that the adjusted figure excludes. Nebius is not hiding that. It is asking investors to look at the operating line and accept the rest as the cost of getting bigger.

The growth rate deserves a note on arithmetic. A 571% increase means the year-earlier quarter was around $146 million, so the company added more than $830 million of quarterly revenue in twelve months. Growth at that speed is less a demand signal than a construction schedule: revenue arrived when GPUs arrived, in the order the buildings came online.

Then came the part that separates Nebius from the other neoclouds. Management gave preliminary guidance for 2027, two years out, of $17 billion to $20 billion in revenue, and set capital spending for that year at $12 billion to $13 billion. It said 75% to 85% of that spending is already covered by customer prepayments and investment-grade financing, a range that works out to roughly $9 billion to $11 billion.

Guidance two years ahead is unusual for a company at this stage, and it is being offered for a reason. Buyers of large compute contracts, the labs and platform companies that sign for capacity years before they can use it, want to know their supplier will still be building. Lenders want to know someone else has committed first. Publishing a 2027 number is a way of telling both groups that the order book exists and is not a rumor.

The prepayment structure is the more interesting disclosure. A customer that pays in advance is financing its supplier’s construction, and it does so because it cannot get the capacity any other way. That inverts the normal direction of working capital in cloud computing, where providers build and customers pay monthly. It also transfers a portion of the construction risk to the buyer, who accepts it in exchange for a slot.

“Prepayments are how this generation of cloud companies solved its financing problem without solving its customer-concentration problem,” said an analyst at a European research firm who covers data-center economics and asked not to be named discussing individual companies. “You get the money up front and you give up the ability to sell that capacity to anyone else.”

Concentration is the obvious exposure. When a large share of contracted revenue comes from a handful of AI developers, their spending decisions become the company’s demand forecast, and their funding rounds become its credit event. Nebius does not disclose the mix in its headline results, and the guidance range is wide enough, $3 billion wide, that it can absorb a great deal of movement before a shortfall shows.

There is also the matter of what the 53% margin means once the depreciation arrives. Adjusted EBITDA is measured before the cost of the equipment is spread across its useful life. A GPU fleet bought in 2025 and 2026 will be depreciating through 2027 and 2028, the same years the guidance covers, and the useful-life assumption behind that depreciation is being tested across the industry as successive chip generations shorten the economic life of the previous one.

The guidance also sets a test the company has chosen for itself. Management could have published 2027 revenue without the capital-spending figure, and the two numbers together are what make the promise checkable. A reader can now ask whether $12 billion to $13 billion of equipment produces $17 billion to $20 billion of revenue, and can compare that ratio with what the hyperscalers report on their own construction.

None of this makes the quarter weak. It makes it early. Nebius has done what a capital-intensive challenger has to do: convert demand into contracts, contracts into prepayments, prepayments into buildings, and buildings into revenue fast enough that the interest bill stays smaller than the growth. The model works as long as each link holds in that order.

The comparison that matters is not with the hyperscalers, who fund their own construction from cash flow and can wait out a soft year. It is with the other companies pursuing the same trade, several of which have signed similar advance-payment deals with the same small set of customers. If those customers slow down at the same time, the sector’s financing structure turns from an advantage into a shared exposure.

For the next two quarters, watch two figures: the share of revenue from the largest customers, and whether capital spending tracks the low or high end of the 2026 plan. The first tells you how much of the 2027 promise rests on a few signatures. The second tells you whether the buildings are going up on schedule, which is the only thing that turns a prepayment into revenue.

  • Related Posts

    • November 14, 2026
    • 51 views
    Anthropic Packages Claude for the Factory Floor

    In a machine shop that supplies aerospace brackets, the quoting process has looked the same for two decades. An engineer opens a spreadsheet, pulls costs for material and machine time,…

    • November 14, 2026
    • 50 views
    Chipmakers and Cloud Providers Form an Alliance Around SGLang

    At the SGLang summit on Nov. 12, the slide that mattered was the one with eight logos on it. LMSYS, the research group behind the Chatbot Arena model leaderboard, used…