Nebius's $40B AI Buildout Relies on Prepayments and Loans

TechCrunch reported on 12 November 2026 that Nebius intends to fund a $40 billion AI infrastructure expansion through customer prepayments and loans. The same report notes that delays at a Microsoft data center are raising questions about the company's financing model. The report does not specify the terms of the prepayments, the lenders involved, or the size of the delay.

For anyone whose product or marketing pipeline depends on rented compute, the story is less about one company's balance sheet than about how AI capacity gets paid for — and what that means for cost, availability and planning.

Prepayments as a financing instrument

Cloud capacity has usually been financed by the provider and paid for by the customer month to month, with commitments layered on top. Building against customer prepayments inverts part of that: the buyer funds the construction before consuming the service. The arrangement can work when demand is contracted and the buyer needs guaranteed access, but it ties the provider's cash flow to a small number of large customers.

The report does not say which customers are prepaying, what they receive in return, or whether the money is tied to specific sites. Those gaps matter for anyone evaluating a multi-year compute commitment, because the counterparty risk sits with the party holding the deposit.

What a data center delay actually signals

A single delayed facility is not proof of a broken model. Large data centers routinely slip against schedules for reasons that have little to do with demand: power connections, construction, equipment lead times. The relevant question is whether the delay is an isolated execution problem or a symptom of financing that depends on milestones being hit on time.

The report does not say how far the Microsoft data center has slipped, nor whether Nebius's own spending plan is affected. Treat the delay as a signal to check, not a conclusion.

What a team can verify this week

Most teams will not buy $40 billion of infrastructure, but many now hold AI compute commitments that behave like financing agreements. A few practical checks:

• List every AI vendor contract that involves a prepayment, a minimum spend or a reserved-capacity commitment, and note the amount at risk if delivery slips.
• Ask each vendor what happens to prepaid amounts in case of delay or non-delivery, and get the answer in writing rather than in a sales call.
• Model your unit economics twice: once with current AI costs, once with costs 30% higher for two quarters. If the second scenario breaks the business, the exposure is a financing problem, not a pricing one.
• Keep a fallback provider or an open-weight model path for the workloads that matter most, and test it before you need it.

The broader context is that AI infrastructure is being financed with structures closer to project finance than to ordinary software subscriptions. That shift eventually reaches buyers through contract terms: longer commitments, prepayment clauses and capacity guarantees. The report leaves open how Nebius's model will hold if delivery slips, but the question it raises is one every team renting AI capacity should ask its own suppliers.