Nvidia offers to take a cut of AI cloud revenue on top of hardware sales in new optional financing vehicle — trades tokens for revenue cut

Nvidia has announced a new business model under which it’ll be able to double-dip for revenue on the same silicon: once when its partner AI clouds use its hardware, and again as an ongoing percentage of the revenue that hardware generates. In a blog post co-authored by CFO Colette Kress, the company pitched the “revenue-sharing and credit-support model” as a way to open compute access to startups that can’t finance it themselves. In practice, cash-poor AI companies trade a slice of whatever they eventually earn for tokens today, while a supplier already running roughly 75% gross margins reaches into its customers’ income statements for a second helping of cash. Australia’s Sharon AI and Singapore-based Firmus Technologies are the first named partners.

Under the structure, participating AI clouds procure Nvidia infrastructure and sell Nvidia-powered cloud services to end customers. Nvidia collects its usual product revenue on the hardware plus a percentage of the cloud income earned on that capacity, which the blog post describes as a recurring, usage-linked earnings stream. Per Bloomberg, developers receive token credits in exchange for a slice of their future sales, but neither Nvidia nor its partners has disclosed the revenue-split percentages.

It’s no secret that the credit-support side of this model will help to address a financing gap that Nvidia itself has identified. Even signed, long-term customer commitments have failed to convince lenders to fund large-scale deployments, leaving smaller clouds unable to borrow against the demand they had already generated.

In an 8-K filing dated June 12th, Sharon AI disclosed that the agreement runs for six years and covers 72 MW of new Australian data center capacity built to Nvidia’s DSX AI factory design, scaling to as many as 40,000 Grace Blackwell GB300 GPUs. The Nasdaq-listed neocloud separately holds a revenue-share facility of up to $200 million with investor Digital Alpha, disclosed in its CY25 results, meaning portions of its income are now pledged in two directions. Meanwhile, Firmus is building a DSX-aligned campus in Batam, Indonesia, that’s expected to scale to 360 MW and house up to 170,000 Nvidia GPUs.

Nvidia has spent much of the last year funnelling cash directly to its customers, including a $30 billion participation in OpenAI’s $110 billion funding round and backing for xAI’s $20 billion Colossus 2 financing, arrangements that drew repeated circular financing criticism. The new model inverts that: rather than investing capital that returns as GPU orders, Nvidia extends credit support and collects a royalty on its partners’ sales for years afterward.

That royalty also ties a slice of Nvidia’s income to utilization instead of hardware sales. If partner clouds can’t keep racks rented, the usage-linked stream shrinks, a live concern given the depreciation pressure already building on operators paying off hardware that Nvidia refreshes pretty much every year.