Nebius’s $40 Billion AI Buildout Faces a Delivery Test

Introduction

The race to build AI computing capacity is encouraging a new class of cloud providers to finance infrastructure around long-term customer commitments. Nebius is taking that strategy to an unusually large scale. On its third-quarter earnings call, the company outlined between $40 billion and $45 billion in planned 2027 capital spending, with roughly $40 billion highlighted in its broader expansion plan. The ambition is significant, but so is the execution risk: the financing depends on data centers being delivered on time and customers continuing to stand behind their commitments.

Key points

• Future customer cash is central to the plan. Nebius expects customer prepayments and asset-backed financing tied to investment-grade contracts to cover up to 85% of its 2027 capital spending. The company identified Microsoft and Meta as customers whose contracts can support this type of financing, while also saying it could use additional prepayments if more cash were needed.
• A major Microsoft project is running late. Capacity at a Vineland, New Jersey, data center was originally scheduled for deployment in 2025 and 2026. Nebius now expects only slightly more than half of the Microsoft capacity to be delivered by the end of the year, with the remainder moving to the second quarter of 2027. The contract is worth up to $19.4 billion and includes financial penalties and termination rights if deadlines are missed without alternative capacity.
• Power and site issues are contributing to the delay. DataOne, the third-party operator, accelerated the shutdown of temporary generation as the site transitions to Bloom Energy fuel cells. The facility had also been fined $1.1 million by New Jersey regulators over unpermitted natural-gas generators, while nearby residents have filed a noise-related lawsuit.
• Rapid growth has not removed near-term pressure. Nebius reported third-quarter revenue of $981 million, up 571% year over year. About one-third came from a one-time payment connected to a new partnership model. The company cut its 2026 revenue forecast by $200 million to $3 billion, citing delays at third-party sites, and its shares fell 17% after the announcement.

Why it matters

Nebius is effectively using project-finance logic for AI infrastructure: secure a customer commitment, collect money or borrow against the contract, and build the capacity that will generate future revenue. The approach can reduce the amount of capital the company must fund upfront and help large customers reserve scarce computing capacity. It also creates a tight connection between construction schedules and financial stability.

If a facility is delayed, costs continue while revenue may not arrive. A customer may demand substitute capacity, seek penalties, or terminate the affected service after a contractual grace period. That risk is particularly relevant as Microsoft reportedly increases oversight of third-party data-center projects and reassesses agreements that fail to deliver on time. A prepayment is therefore not the same as unconditional revenue; it remains connected to the provider’s ability to perform.

Nebius says the sites it controls directly are on schedule or ahead of schedule, and that prepayments from newer contracts more than cover the cost of the capacity being built. It ended the quarter with $7.4 billion in cash and forecasts $12 billion to $13 billion of 2027 revenue, with pretax break-even expected in the fourth quarter.

The central question is not whether demand for AI capacity exists. It is whether Nebius can align financing, energy supply, construction, and customer delivery at unprecedented scale. Its $40 billion plan could validate the neocloud playbook—or show how quickly that model becomes vulnerable when one major project slips.

Source: TechCrunch AI