Nvidia Offers Startups Compute in Exchange for Revenue Share

Nvidia has spent the AI boom selling shovels. Now it wants a cut of the gold.

The world's most valuable chipmaker unveiled a new business model on July 2 that lets fast-growing AI startups tap its GPUs without paying the full, punishing cost upfront. In exchange for token credits and access to computing infrastructure, participating companies will hand Nvidia a slice of their future product and cloud revenue, a structure that turns the company from a hardware vendor into something closer to a financial backer of the AI economy it powers.

The pitch is aimed squarely at the "AI-native" tier of the market: model builders, inference providers and agent platforms that need enormous compute but lack the capital, or the credit history, to lock in multiyear infrastructure commitments. Under the program, Nvidia is partnering with so-called AI clouds to build large-scale, multi-tenant "AI factories," then sharing in the revenue those factories generate.

"Emerging AI companies historically have had limited access to capital-intensive infrastructure, with even long-term commitments insufficient to unlock financing for compute," Nvidia executives Colette Kress and Raj Mirpuri wrote in a company blog post announcing the model. Nvidia, they said, will earn "both standard product revenue and a share of the cloud revenue on the supported capacity," giving it a "recurring, usage-linked earnings stream."

The First Partners

Two initial partners are anchoring the rollout, and together they represent potential access to more than 200,000 GPUs.

Sharon AI, an infrastructure startup, is deploying up to 40,000 of Nvidia's newest Grace Blackwell GB300 GPUs. "This strategic collaboration with NVIDIA marks a pivotal moment in Sharon AI's mission to deliver sovereign, large-scale AI compute infrastructure," said James Manning, the company's cofounder and CEO.

The larger bet is in Southeast Asia. Firmus Technologies, an Australian startup, is building a DSX-aligned AI factory campus in Batam, Indonesia, a fast-industrializing island just south of Singapore. The campus is expected to scale to 360 megawatts of power and as many as 170,000 Nvidia GPUs, and Firmus expects between $25 billion and $30 billion in committed offtake agreements over the partnership's first six years.

"AI-native companies need access to scalable, energy- and cost-efficient compute infrastructure to compete globally," said Tim Rosenfield, co-CEO of Firmus Technologies. "Firmus AI cloud is building a NVIDIA DSX-aligned AI factory, which will enable our cloud to help more customers access the compute they need to build and scale AI."

Nvidia pointed to AI natives such as Baseten, Fireworks AI and Together AI as illustrations of where demand is heading, companies that need immediate capacity for training, fine-tuning and high-volume agentic inference but also want commercial flexibility as products move from pilot to production.

Why It Matters

The revenue-sharing model is the clearest sign yet that Nvidia intends to embed itself into the AI stack not just as a supplier but as a financial partner, deepening a lock-in that already borders on total.

By absorbing some of the upfront risk of building AI infrastructure, Nvidia lowers the barrier for startups to standardize on its chips and software rather than shopping for cheaper alternatives from AMD, in-house silicon or cloud-native accelerators. Every GB300 that ships under a revenue-share deal is another customer whose economics are aligned with the Nvidia platform for years. The company effectively earns twice on the same deployed capacity: once on the hardware margin, and again on a usage-linked cut of the cloud revenue.

It also converts compute into a form of equity. Rather than a customer paying cash for chips, Nvidia is now betting on the commercial success of the companies it equips, a compute-as-capital arrangement that resembles the token-credit and vendor-financing deals it has struck with OpenAI, CoreWeave and others.

That is precisely what worries skeptics. The AI sector is already dense with circular financing, arrangements in which Nvidia invests in, or extends credit to, customers who then use that money to buy Nvidia chips. Revenue sharing threads the same needle from a different angle: Nvidia's growth increasingly depends on the fortunes of startups whose own survival depends on Nvidia's continued willingness to underwrite them. If AI demand cools, the interlocking commitments that look like momentum today could unwind quickly, with Nvidia holding both the hardware exposure and a claim on revenue that may never materialize.

For now, the demand signal is pointed sharply upward. Nvidia framed the model around a market shift from one-time model training toward "continuously operating AI factories that generate tokens at scale," the kind of always-on inference workload that keeps GPUs, and Nvidia's new revenue stream, humming.

What to Watch

The key question is how large a share Nvidia takes, and whether the terms hold up if AI startup revenue proves lumpier than projected. Neither Nvidia nor its partners disclosed the specific percentage of revenue changing hands, a detail investors and rivals will press for. Watch, too, whether marquee AI natives like Together AI or Fireworks formally sign on, how quickly the Batam campus clears site, power and construction hurdles, and whether regulators start scrutinizing the increasingly circular flow of capital binding Nvidia to its own customers. If the program scales, expect competitors to answer with financing offers of their own, turning the fight for AI compute into a fight over who is willing to share the risk.