Nvidia unveils revenue-sharing financing that takes a recurring cut of AI cloud income

Nvidia has introduced an optional revenue-sharing financing model that lets AI cloud partners draw token credits against future compute capacity now. In return, the chipmaker collects standard hardware revenue plus a recurring cut of the cloud income that capacity generates. Chief Financial Officer Colette Kress and Raj Mirpuri laid out the financing model in a company blog post on July 1, 2026.

The move changes what Nvidia is to its customers. Instead of a one-time hardware sale, it becomes a recurring stakeholder in whether AI data centers fill with paying tenants. Nvidia also reportedly agreed to buy back unsold GPU capacity at an agreed price if a partner cannot find customers, a backstop that shifts some buildout risk onto the chipmaker itself.

Two named partners anchor the launch. Sharon AI is deploying up to 40,000 Nvidia Grace Blackwell GB300 GPUs built to Nvidia’s DSX AI factory design. Firmus Technologies is building a DSX campus in Batam, Indonesia, expected to scale to 360 megawatts and up to 170,000 GPUs. Combined, the two could reach roughly 210,000 GPUs.

The structure ties Nvidia’s results more tightly to demand for the AI compute its own chips create, and the buyback pledge is a bet that capacity will sell. It also helps keep the buildout financeable for smaller clouds that could not otherwise front the hardware bill.

Nvidia did not disclose the size of its recurring cut, the buyback pricing, or how many partners have signed beyond the two named. Whether the model spreads past a handful of greenfield builds is the open question.