Nvidia has announced a new revenue-sharing and credit-support model that allows AI cloud providers to access large volumes of GPUs without paying the full cost upfront. Instead, Nvidia collects standard product revenue on chips plus a share of the cloud revenue generated from renting them out. Two early partners — Sharon AI and Firmus — are already operating under this model, with Firmus building a 360-megawatt AI factory in Indonesia housing up to 170,000 GPUs. The arrangement targets AI-native companies like Baseten, Fireworks AI, and Together AI that need elastic compute access without the balance-sheet risk of building data centers. For Nvidia, it creates a recurring usage-linked income stream on top of hardware sales, though it also exposes the company to AI demand slowdowns through both chip sales and shared cloud revenue.
