Nvidia's new financing program pairs chip sales with a cut of the cloud revenue those chips generate, making the company's datacenter earnings partly a function of whether its customers can fill the servers they borrow to build.
Nvidia is no longer content to be paid once for the chips it sells. Under a new financing program, the company helps AI cloud startups secure datacenter loans, then collects a cut of the cloud revenue those chips generate. The result is a recurring earnings stream that rides on Nvidia's customers' actual usage.
The structure was first reported by The Register, which called it a "double-dipping" arrangement. Nvidia's own framing, in a company blog post announcing the program, is softer: a "recurring, usage-linked earnings stream" layered on top of standard product sales. Both descriptions point to the same underlying shift. Nvidia is converting chip customers into revenue-share partners.
The mechanism is not, strictly speaking, a standard vendor-financing deal. Nvidia may only be brokering third-party capital rather than putting its own balance sheet behind the loans, according to The Register. What is novel is the revenue split: even when Nvidia is not the lender of record, it still expects a cut of the cloud revenue generated by the chips it helped finance. That makes the company's datacenter earnings partially a derivative of its own customers' utilization rates, rather than a clean sale of hardware.
Two named deployments illustrate the scale. Sharon AI has committed to deploying up to 40,000 Nvidia Grace Blackwell GB300 GPUs under the program, and Firmus is building a 360-megawatt DSX AI factory campus in Batam, Indonesia, scaling toward up to 170,000 Nvidia GPUs. Neither figure implies Nvidia is on the hook for the loans themselves; both imply Nvidia has a stake in whether the resulting compute capacity actually finds paying tenants.
The structure extends a playbook Nvidia has been building for at least a year. The Nvidia-IREN strategic partnership, announced to accelerate deployment of up to five gigawatts of AI infrastructure, used equity-like terms and revenue exposure to lock in long-dated GPU demand. The new program generalizes that pattern to a broader set of emerging AI clouds, with capital partners supplying the loan principal. Nvidia collects on both ends: hardware margins today, recurring cloud revenue tomorrow.
That is the part worth watching. Peer AI clouds such as CoreWeave and Lambda have already borrowed billions from venture capital and hedge funds to bankroll datacenter buildouts, the same market gap the Nvidia program targets. If a meaningful slice of the new program's loans flow into operators whose only path to repayment is selling tokens, inference, or training cycles into a softening enterprise AI budget, Nvidia's "usage-linked" line item becomes a leading indicator of demand stress rather than a lagging one. The clean test is the next quarter's earnings call. If Nvidia reports the usage-linked stream under the same datacenter umbrella as standard chip sales, the question to ask is not how big the chips number got, but how much of it is Nvidia's own money coming back through a customer.