Nebius Plans $40 Billion in 2027 Spending as Its Vineland Site Slips

Nebius has laid out $40 billion in planned spending for 2027, according to its third-quarter earnings call. The company plans to fund the buildout by relying on customer pre-payments and loans backed by customer contracts.

The plan comes as delivery problems at one of its sites are in public view. Microsoft is one of Nebius's largest customers, paying for AI computing capacity at a data center in Vineland, New Jersey, that is months behind schedule. The contract with Microsoft is worth up to $19.4 billion. Its stages, which are discrete blocks of capacity, were originally meant to be deployed in 2025 and 2026, and the contract includes financial penalties for late delivery.

Only just over half of the Microsoft capacity is now planned for delivery by the end of 2026. The remainder is planned for the second quarter of 2027.

Nebius attributed the Vineland delay to its operator, DataOne, which decided to wind down temporary generation faster than expected as the site switches to Bloom Energy fuel cells. The change of power source is the reason given for the slip, and the schedule now depends on that transition.

Under the summary of the Microsoft contract, Microsoft can terminate the affected service after a grace period if the dates are missed and no alternative capacity is offered. That clause turns a schedule slip into a commercial question with a defined window attached to it.

Nebius chief revenue officer Marc Boroditsky said the company's deals with Microsoft and Meta both came with significant prepayment that arrived long before delivery. Chief executive Arkady Volozh said building at this scale is hard, that delays are part of the industry, and that Nebius is not immune.

On financing, Nebius's newsroom reports a $3.5 billion financing facility with Citi. The same newsroom reports that the company priced an upsized private offering of $5.0 billion of convertible senior notes, and that it closed a private offering of convertible senior notes with aggregate gross proceeds of approximately $5.75 billion.

The broader context here is the financing structure. Neocloud operators, which rent GPU (graphics processing unit) capacity to hyperscalers, the large cloud providers, and to model developers, depend on customers paying ahead of delivery and on lenders accepting contract cash flows as collateral. Boroditsky's description of prepayments explains how the model can fund construction at this scale. It also means that when capacity arrives late, the cash has often already been received, and the terms of the remaining obligation become the operative risk.

In my view, the most instructive detail is the power change rather than the spending figure. A site moving from temporary generation to fuel cells, which produce electricity through an electrochemical reaction rather than by burning fuel, is a reminder that the binding constraint on AI capacity is increasingly electrical. Schedule risk sits with power equipment and site operators as much as with accelerators and networking. Nebius's own statement that delays are part of the industry points the same way.

If the Vineland transition goes as planned, the Microsoft capacity will come online in two stages, and the prepayment-plus-contract financing model will have been tested under a real delay rather than a hypothetical one. Whether that test ends with capacity delivered on the revised schedule is what the next several quarters will show. The $40 billion plan, if executed, would give Nebius a larger footprint to build on. How the company handles this first slip will indicate how much of that footprint can be financed on the same terms.