Nvidia trades GPU financing for revenue cut
Nvidia is financing GPU buildouts for smaller cloud operators in exchange for a share of their cloud revenue; first partners are in Australia and Indonesia.
Nvidia has launched a formal program to backstop GPU infrastructure buildouts for smaller cloud operators in exchange for a share of their cloud revenue. Instead of a one-time hardware sale, Nvidia absorbs some of the financial risk of a GPU deployment and collects ongoing revenue from the cloud business running on those chips.
Two named partners have emerged. Sharon AI is deploying 40,000 Nvidia GB300 Grace Blackwell Superchips in Australia. Firmus Technologies is building a facility with roughly 170,000 GPUs in Batam, Indonesia. The revenue-share percentages have not been disclosed.
For a smaller cloud operator that cannot comfortably fund tens of thousands of high-end GPUs upfront, this lowers the entry barrier. For Nvidia, it means recurring revenue from deployments that might otherwise have been delayed or downsized, and a commercial stake in how much revenue its partners generate.
What it means for you
If your cloud provider is a neocloud rather than a hyperscaler, it may now be operating under a revenue-sharing agreement with its GPU supplier. That adds a layer worth understanding: the provider’s pricing and capacity decisions may be partly shaped by what it owes Nvidia per dollar of revenue, not just by the capital cost of hardware it bought outright.
For context on the broader competition for AI infrastructure, what the AI chip race means for you covers the stakes. This program is Nvidia moving to capture recurring cloud economics on top of hardware margins, which gives smaller cloud providers a path to scale that depends on Nvidia staying their financial partner, not just their supplier. Claude’s availability on GB300 Blackwell Ultra in Azure is a separate illustration of the same hardware layer mattering for AI inference economics.