The Labs, the Neoclouds and Nvidia's Own MoneyThin moat

The fastest-growing part of Nvidia's customer base is the one that did not exist five years ago: the model labs and the specialist AI clouds built to serve them. In August 2026 Nvidia agreed to guarantee up to $105 billion of the land, power and shell for a 4.25-gigawatt campus in Ohio that will host its infrastructure exclusively under twenty-year leases to OpenAI, and said each generation deployed there could be worth $150 billion to $200 billion of its revenue¹. Alongside them sit the neoclouds — CoreWeave and its peers — which exist principally to buy Nvidia hardware and rent it out.

What complicates this group is that Nvidia is frequently on both sides of the transaction. It holds an equity stake of around 11% in CoreWeave, whose business is purchasing Nvidia GPUs, and it has invested in or committed to several of the labs whose compute commitments it then books as demand. Supporters call this ecosystem development, which it partly is; critics call it circular financing, which it also partly is. Both descriptions can be true of the same dollar.

The exposure is qualitatively different from the hyperscalers'. Microsoft and Amazon will pay their bills in any macro environment; a model lab burning capital against future revenue, or a neocloud servicing GPU-collateralised debt, is a credit as much as a customer. Watch Nvidia's receivables and the share of revenue coming from customers it has also financed, and watch whether the announced gigawatts convert into deployed capacity on schedule. Demand that depends on your own balance sheet is the kind that disappears exactly when you would most like it to hold.