Nebius Shifts from Leased to Owned Capacity as $50 Billion Backlog Drives $25 Billion Infrastructure Bet

Nebius pivots from leasing to owning AI infrastructure, securing $50B in contracts including $27B from Meta, as revenue surges 684% to $399M.

The cloud infrastructure race is entering a new phase, and Nebius is betting that owning the assets rather than renting them will be the winning formula. The Amsterdam-based AI cloud provider is no longer just selling computing cycles from third-party racks; it is now buying land, securing power connections, and building its own data centers from the ground up.

That strategic pivot was laid bare in the company’s latest quarterly update. More than 75% of Nebius’ contracted power portfolio now sits at facilities it controls directly, a sharp reversal from the predominantly leased model it ran just a year ago. The rationale is straightforward: vertical integration over access, cooling, chip integration, and the software layer gives Nebius more control over delivery timelines and margins.

The backbone of this transformation is a contract backlog that has grown to roughly $50 billion. At its core sit two marquee deals: a multi-year AI infrastructure agreement with Meta Platforms worth $27 billion, and a commitment from Microsoft valued at up to $19.4 billion. These agreements are not speculative orders—Nebius is building against committed capacity, which puts immense pressure on execution.

Financially, the company has the firepower to meet those commitments. Cash and equivalents stood at $9.3 billion at the end of March, up from $3.68 billion at year-end, after a $6.3 billion capital injection. That included a $2 billion equity investment from Nvidia and $4.3 billion in convertible senior notes. The Nvidia tie-up also secures early access to next-generation GPU hardware, a vital edge in a market where chip supply remains the bottleneck.

The scale of the buildout shows up in the investment budget. Nebius has raised its capital expenditure target for the current year to between $20 billion and $25 billion, with a large portion going toward capacity that will come online in 2026 to serve existing customers sooner. The medium-term goal for contracted power capacity was lifted to more than 4 gigawatts from 3 GW, with 3.5 GW already under contract today. Longer term, the company is aiming for 5 GW of active capacity by 2030.

Operationally, the numbers reflect the early fruits of that expansion. First-quarter revenue hit $399 million, a 684% jump from a year earlier. Almost all of that—$390 million—came from the AI cloud segment, which now represents 98% of group sales. Annualized recurring revenue in that business rose to $1.92 billion, a sequential increase of $670 million. On an adjusted basis, earnings before interest, taxes, depreciation, and amortization swung to a $130 million profit, compared with a $54 million loss in the prior year. The GAAP bottom line remains pressured by heavy depreciation, but the EBITDA inflection signals that the operating leverage is kicking in.

Site selection is moving in parallel. In Independence, Missouri, construction is underway on a 1-gigawatt campus spread across roughly 400 acres. Just a day after that announcement, Nebius revealed a second site in Pennsylvania, where it has secured land and up to 1.2 GW of power for its own AI factory, with a target of coming online in 2027.

On the capital markets side, one insider took advantage of the momentum to lock in gains. Board member Elena Bunina sold shares worth approximately $2.2 million through a prearranged trading plan, a move that did not rattle investors given the size relative to the company’s market cap.

For the full year, management is sticking to its revenue guidance of up to $3.4 billion, with a major capacity ramp expected in the third quarter. The path is clear: Nebius must now translate paper commitments into working power, cooling, and compute. If the Pennsylvania and Missouri builds stay on schedule, the company will take a significant step toward becoming the kind of infrastructure powerhouse that hyperscale AI customers will rely on for the rest of the decade.