What Nvidia announced
On July 1, 2026 Nvidia unveiled a new business model for what it calls AI factories: AI cloud companies procure Nvidia infrastructure and sell computing services on top of it, and Nvidia earns its usual product revenue plus a share of the cloud revenue generated by the capacity it supports, alongside revenue-sharing and credit-support mechanisms that, in Nvidia's words, align the economics of both sides. The first named partners are Sharon AI, which is deploying up to 40,000 Grace Blackwell GB300 GPUs and whose chief executive James Manning frames the deal as sovereign, large-scale AI compute, and Firmus, which is building a 360-megawatt campus in Batam, Indonesia with up to 170,000 GPUs. Baseten, Fireworks AI and Together AI are named as prospective customers of that capacity. Independent coverage, from Forbes to analyst notes, described the same mechanics in plainer words: the neocloud gold rush is now vendor-financed, and the vendor takes a cut of the rent.
From selling shovels to collecting rent
The old story about Nvidia was the shovel seller in a gold rush: it sold the hardware and let others carry the business risk of renting it out. This program ends that separation. When the supplier provides credit support to its buyers and takes a recurring, usage-linked share of their rental income, it is no longer only a supplier; it is a financier and a silent landlord of the GPU cloud market. This is also a different animal from the 2025 equity loops, in which Nvidia invested in customers such as OpenAI who then bought its chips. Equity is a one-off bet that can be sold. A contractual share of every rented GPU hour is a permanent seat inside the market's cash flows, and it gives the supplier a direct interest in keeping utilisation, and therefore prices, high.
The floor under your compute price
For anyone who buys GPU capacity rather than Nvidia shares, two consequences matter. First, pricing: a market in which the dominant supplier earns a percentage of the rental revenue has a structural floor, because the party that controls the supply of new capacity now loses money twice when rents fall. Second, counterparty risk: credit support sounds reassuring, and in the near term it is, but it concentrates the sector's creditworthiness on a single balance sheet. If Nvidia one day tightens that support, capacity growth, provider solvency and prices would all move at the same moment, in the same direction. A European business renting compute in euros should therefore read its provider's financing the way it reads a landlord's mortgage: it defines how the rent behaves under stress.
Sovereign compute, financed by the dependency
The sharpest irony sits in the marketing. The first flagship partner sells its capacity as sovereign AI compute, yet the program that makes it possible is financed, credit-backstopped and revenue-shared by the very US supplier that sovereignty programs are meant to reduce dependence on. That is sovereignty of the sticker, not of the stack. The practical move for an owner is not outrage but diligence: when you contract AI compute, ask who finances your provider, what happens to your contract if that support changes, and whether your workloads could move elsewhere in weeks rather than years. Multi-sourcing models has become normal practice; multi-sourcing the compute underneath them is the next discipline, and it is cheaper to build before you need it.
Read next: AI Tokens Now Have a Rush Hour · OpenAI Offers Washington a Stake
