AI Cloud Buyers Now Prepay Full Capital Costs

Nebius's customers covered more than 100% of related spending on third-quarter deals, a shift that moves financing risk onto renters while capacity timing slips into 2027.

Updated on Nov. 12, 2026 in Data Centers

AI Cloud Buyers Now Prepay Full Capital Costs

Nebius reported third-quarter 2026 revenue of $980.8 million, up 571% year over year, and lowered its 2026 revenue outlook to the low end of $3.0 billion to $3.4 billion after third-party capacity slipped into 2027. The results describe reported financials from a company now scaling AI cloud infrastructure, not a new product launch.

Why it matters

The quarter shows AI cloud providers increasingly funding data-center buildouts through customer prepayments and asset-backed debt rather than equity alone. Capacity delays at a New Jersey site contracted to Microsoft show how timing risk in GPU-era infrastructure can move revenue targets across fiscal years.

Nebius reported Q3 revenue of $980.8 million and adjusted EBITDA of $465.8 million, with a net loss from continuing operations narrowing to $37.5 million from $119.6 million. Prices per GPU-hour rose more than 20% on average, and about $325 million of the quarter's revenue was non-recurring, which the company has not characterized as repeatable.

The players

Nebius

An Amsterdam-headquartered AI cloud provider that rents GPU capacity and operates owned data-center sites in the United States and Finland.

Microsoft

A cloud and software company that contracted colocation capacity at Nebius's New Jersey site, whose delivery has been delayed.

Citi

A global bank that lined up a $3.5 billion senior secured facility with Nebius in October 2026.

Bloom Energy

A fuel-cell manufacturer that is supplying the fuel cells the New Jersey site is shifting to for permanent power.

The details

Nebius's customers committed prepayments tied to the capital spending on their contracts, and prepayments on third-quarter deals covered more than 100% of related capital spending, up from 50% to 60% on second-quarter deals. The company expects prepayments plus asset-backed financing tied to investment-grade customer contracts to fund 75% to 85% of 2027 spending. Graphics processing units (GPUs), the chips that run AI workloads, are rented by the hour, and the company's pricing per GPU-hour rose more than 20% on average.

Timeline

  1. December 2025: Nebius's original target of $7 billion to $9 billion run-rate revenue was set.

  2. August 2026: Nebius raised $5.75 billion in convertible notes.

  3. October 2026: Nebius lined up a $3.5 billion senior secured facility with Citi.

  4. Q4 2026: Nebius expects more than half of Microsoft-contracted New Jersey capacity to be delivered.

  5. February 2027: Nebius expects to top $7 billion annualized run-rate revenue.

The Tech Race

Nebius's prepayment coverage rising above 100% of capital spending on third-quarter deals extends a financing pattern first visible in its second-quarter contracts, where coverage was 50% to 60%. The company's capacity delays also show how dependent these providers are on timely power and construction, a constraint that separates them from competitors relying on existing facilities.

Customers of AI cloud services face higher GPU-hour pricing, which rose more than 20% on average at Nebius. Delayed capacity at the New Jersey site means contracted deliveries may arrive later than originally scheduled, and the company has not provided a firm date beyond its fourth-quarter 2026 expectation.

The takeaway

Nebius's 2027 targets depend on prepayments and asset-backed financing covering 75% to 85% of spending, so watch whether committed customer prepayments reach the projected $17 billion to $20 billion. Its next checkpoint is the February 2027 announcement on whether annualized run-rate revenue tops $7 billion.

What happens next

Nebius expects to top $7 billion in annualized run-rate revenue in February 2027, and expects more than half of the Microsoft-contracted New Jersey capacity to be delivered in the fourth quarter of 2026.

Further reading

For more on how AI cloud capacity gets built and financed, explore Data Centers.