Nebius plans to spend $40 billion on AI infrastructure, funding it through customer prepayments and loans rather than equity markets. It is the clearest test yet of whether the “neocloud” model — building data centres for a single anchor tenant — can scale into something resembling a national infrastructure strategy.
The approach has an elegant logic. If someone like OpenAI or Microsoft needs compute capacity, they prepay, and Nebius uses that cash to secure construction loans. The customer gets guaranteed access; Nebius avoids diluting shareholders. It aligns incentives in a way that pure venture capital often does not.
But the model is only as strong as the anchor customers. Nebius faces delays at a Microsoft data centre partner, and any slowdown in customer demand would leave it holding expensive construction debt without the revenue to service it.
When private profit meets public purpose
The Guardian editorial today argues that returns for venture backers do not automatically translate into national prosperity. Nebius is a case in point: even if its model works for shareholders, it concentrates AI infrastructure around the needs of the world’s richest tech firms. Smaller economies and public-sector users get priced out.
The deeper question is whether any private financing model can build the kind of widely accessible AI infrastructure that resembles a public utility. Nebius may prove that customer prepayments can fund construction. Proving they can fund equitable access is a much harder case to make.