NVIDIA launches GPU backstop financing program and takes revenue share from cloud partners' AI compute sales.
On July 1, Nvidia unveiled the "AI Compute Partnership," a financing vehicle under which the chipmaker acts as a financial backstop for neocloud customers' GPU deployments while taking a recurring share of the cloud revenue those GPUs generate. The program's first two adopters are Sharon AI (NASDAQ: SHAZ), an Australian sovereign-cloud provider planning up to 40,000 Grace Blackwell GB300 GPUs, and Firmus Technologies, which is building a 360MW campus in Batam, Indonesia, targeting up to 170,000 Nvidia GPUs.
The arrangement creates a dual revenue stream for Nvidia: standard product revenue from the GPU sale, plus a usage-linked cut of the cloud services those chips power. Nvidia described the model as providing "economic alignment with a revenue-sharing and credit-support model" in a blog post co-authored by CFO Colette Kress. The specific revenue-share percentage was not disclosed.
The backstop mechanism works by Nvidia guaranteeing a floor utilization rate on deployed GPUs. If customer demand falls short, Nvidia commits to renting idle capacity at a predetermined rate, absorbing the downside risk that would otherwise sit with lenders or equity investors. In exchange, Nvidia earns both its standard product revenue and a recurring, usage-linked share of cloud revenue generated on the supported capacity.
This structure addresses a persistent bottleneck in the neocloud sector: smaller operators often have customer demand but cannot secure financing quickly enough to build infrastructure, since lenders view GPU residual values as uncertain. By backstopping idle capacity, Nvidia effectively underwrites the asset, making GPU clusters more bankable. The company framed the program as accelerating "adoption of Nvidia platforms among the high-growth, high-conviction AI native sector."
Sharon AI, which listed on Nasdaq in February 2026 via a $125 million IPO, is deploying up to 40,000 Grace Blackwell GB300 GPUs under a six-year agreement with Nvidia. The company raised an additional $1.6 billion in private placement in June 2026 to fund its Nvidia-based AI factory and has partnered with VAST Data for 600PB of storage infrastructure supporting roughly 100,000 GPUs across its sovereign cloud operations in Australia and the Asia-Pacific region.
Firmus Technologies, a private company, is constructing a campus in Batam, Indonesia, built to Nvidia's DGX SuperPOD reference architecture specifications. The facility is expected to scale to 360MW and house up to 170,000 Nvidia GPUs. Both operators plan to serve AI-native companies including Baseten, Fireworks AI, and Together AI, as well as enterprise inference workloads.
The AI Compute Partnership represents a structural evolution in Nvidia's business model. Previously, Nvidia had provided demand guarantees and customer referrals to help neoclouds raise debt — CoreWeave secured $6.3 billion and Lambda $1.5 billion through arrangements in which Nvidia played a supporting role. The new model makes Nvidia's financial participation explicit, with the company now directly sharing in its customers' cloud economics.
This shift creates a recurring revenue layer on top of Nvidia's hardware sales, potentially smoothing earnings volatility tied to GPU refresh cycles. Analysts have drawn comparisons to GE Capital's equipment financing model, noting both the upside of recurring income and the counterparty risk if neocloud operators face financial stress. Nvidia's blog post positioned the arrangement as providing "a recurring, usage-linked earnings stream" that diversifies revenue beyond one-time chip sales.
Nvidia shares traded at $194.83 on July 3, down 1.4% on the session, on a market capitalization of approximately $4.7 trillion. The stock is up 4.5% year-to-date and 22.3% over the past twelve months. Sharon AI shares fell sharply, dropping 14.2% to $67.91 on volume of 2.6 million shares. The stock, which IPO'd at roughly $16.55 in February, had rallied to a high of $97.48 in June before the pullback. Firmus Technologies is privately held and does not have a public ticker.
The financing model keeps a broader ecosystem of independent GPU cloud providers alive rather than concentrating AI compute inside the hyperscalers — a strategic imperative for Nvidia, which benefits from multiple distribution channels for its hardware. By making GPU clusters more financeable, Nvidia may accelerate buildout timelines for operators that would otherwise face 12- to 18-month fundraising cycles.
The structure also gives Nvidia optionality: in a demand downturn, the company gains access to GPU capacity it can deploy for its own inference services or sublease, effectively building a buffer inventory at customer expense. Critics have flagged this as a form of "double-dipping" — Nvidia profits from the sale and then again from utilization. Supporters counter that it aligns incentives, since Nvidia now has a direct financial stake in its customers' commercial success.