The announcement, made on July 1, describes a partnership model that Nvidia has never used at scale: instead of simply selling chips, the company will help AI cloud providers build and fill data centers, sharing in the revenue those centers generate and supporting their financing. The new model, built around revenue-sharing and credit-support mechanisms, is aimed at accelerating the construction of large, multi-tenant AI factories — the facilities that rent computing power to startups, model developers, enterprises, and research institutions that cannot build their own.
The logic is straightforward. Nvidia’s chips are the foundation of the AI computing boom, and its revenue has grown explosively as cloud providers and technology giants have bought them in quantity. But the company’s leaders have concluded that the next phase of growth depends on more than selling chips: the industry needs more places to run them, and the companies building those places need help financing the multibillion-dollar construction. By supporting cloud providers with revenue-sharing agreements and credit support, Nvidia converts its role from supplier to partner — and creates a stream of recurring income tied to how much computing its partners actually sell.
The structure of the deals matters as much as the principle. In a typical arrangement, according to people familiar with the model, Nvidia would provide its chips to a cloud provider on terms that defer some of the cost, with the provider paying Nvidia from the revenue generated by the computing it rents out. Nvidia’s credit support would help the provider secure the debt financing needed for land, power, and facilities. In exchange, Nvidia shares in the upside: if the AI factory is heavily used, Nvidia earns more; if it sits idle, Nvidia earns less.
The model is an attempt to solve a coordination problem that has defined the AI economy. The companies that need computing — startups training models, enterprises deploying applications — often cannot commit to the capital expenditure required to own it. The companies that build data centers need demand certainty before they invest billions. Nvidia, sitting at the center of both, can see the demand and the supply, and the new model lets it connect them: build capacity, fill it with customers, and get paid on the basis of actual usage.
The approach also carries risks. Revenue sharing means Nvidia’s income from these deals is less predictable than chip sales, and the company is effectively taking on the risk that AI demand does not grow as fast as expected. If the AI factories built under the program fail to fill, Nvidia’s credit support could turn into losses. The company’s executives have said the risk is manageable, pointing to the persistent shortage of AI computing and to the backlog of customers waiting for capacity, but the model is new, and its economics will only be tested over time.
The competitive implications are significant. Nvidia’s dominant position in AI chips has been built on hardware and software, and the new model extends that dominance into the financing and operation of the industry’s infrastructure. Rivals that sell chips without offering similar support — including the cloud providers’ own custom silicon programs — may find themselves at a disadvantage when data center builders compare the total cost of getting capacity online. The model also strengthens Nvidia’s relationship with the cloud providers that are both its customers and, in some cases, its competitors.
For the cloud providers, the offer is a trade-off. Nvidia’s support makes it cheaper and faster to build an AI factory, but it also ties the provider more closely to Nvidia’s technology and terms. Providers that have invested in alternative chips, or that design their own accelerators, will have to weigh the convenience of Nvidia’s financing against the strategic independence of going their own way. The industry’s history suggests most will take the money.
The announcement comes at a moment when the AI infrastructure market is consolidating. The biggest cloud providers are spending tens of billions of dollars a year on data centers, and a wave of newer entrants — funded by investors who believe computing demand will outstrip supply — is building capacity to rent. Nvidia’s model gives the newcomers a path to construction that does not depend entirely on their own balance sheets, which could accelerate the build-out and, eventually, bring prices down.
The broader significance is what the model says about the AI industry’s economics. For the first three years of the AI boom, the value chain was simple: chip makers sold chips, cloud providers rented them, and developers built on top. Nvidia’s move blurs those lines, with the chip maker taking a role in financing, building, and operating the infrastructure that uses its products. It is a bet that the industry’s growth is constrained by capital and capacity, not by demand — and that the company best positioned to remove those constraints will own the cycle.
Whether the bet pays off will be visible in Nvidia’s revenue mix in the coming quarters. The company has said the recurring income from revenue-sharing deals will grow over time, complementing its chip sales. If the model works, it could become the template for how AI infrastructure gets built in the years ahead — not through individual transactions, but through partnerships in which the supplier shares the risks and the rewards of the boom.
