Nvidia Demands a Cut of Cloud Revenue After Selling Its AI Chips
In an announcement detailing partner financing, Nvidia revealed a revenue-sharing model: in exchange for underwriting debt and buying back idle capacity, the vendor takes an ongoing cut of cloud hosting profits.
Under a financing framework detailed by Nvidia and co-authored by chief financial officer Colette Kress, cloud hosting providers building infrastructure with Nvidia hardware must share their ongoing leasing income with the chipmaker. Rather than relying solely on upfront hardware sales, Nvidia will capture recurring revenue from data centers deployed for artificial intelligence workloads.
The setup revolves around what the company calls DSX AI factories—large-scale multi-tenant facilities designed for model training and inference. According to reporting by The Information, Nvidia takes a slice of the host's cloud revenue that gradually tapers over the duration of the contract. Neither Nvidia nor its partners have disclosed the specific percentage splits.
In return, Nvidia operates as a financial backstop. Traditional lenders often hesitate to extend credit to newer cloud providers because predicting the long-term depreciation curve of specialized AI accelerators remains difficult. Nvidia fills the gap by offering credit support: if a provider fails to secure enough tenants, Nvidia pledges to buy back the unsold capacity at a pre-agreed price.
The first two announced partners are Sharon AI and Firmus. Sharon AI entered a six-year commitment covering 72 megawatts of data center capacity in Australia, deploying up to 40,000 Grace Blackwell GB300 units. Firmus is developing a 360-megawatt campus in Batam, Indonesia, designed to scale up to 170,000 accelerators with an estimated $25 billion to $30 billion in contracted demand over six years.
As noted by Tom's Hardware, a vendor already maintaining gross margins around 75% is now taking a cut directly from client balance sheets. The arrangement broadens Nvidia's existing circular commitments across the sector, which include a $6.3 billion compute buyback agreement with CoreWeave alongside investments in OpenAI and xAI infrastructure. If tenant demand for compute decelerates, Nvidia risks absorbing that downturn twice—through reduced hardware orders and shrinking cuts of cloud leasing revenue.
Source: habr
Primary announcement: Nvidia
Comments
Help shape the next version: Add context or suggest a correction. AI review can add points toward a rewrite. Reviews and updates may take time; a full meter does not guarantee a new version.