- Financing Model: Nvidia has introduced a revenue-sharing and credit-support model for AI cloud providers, which rent computing capacity to customers.
- Revenue Share: AI clouds sell Nvidia-powered services while Nvidia receives product revenue and future cloud-revenue participation.
- Partner Scale: Sharon AI plans up to 40,000 Nvidia Grace Blackwell GB300 GPUs, while Firmus is expected to reach 170,000.
- Unclear Terms: Financing mechanics, direct-lending roles and revenue-share percentages remain undisclosed for customers.
Nvidia has introduced a revenue-sharing and credit-support financing model for AI cloud providers, which rent AI computing capacity to customers, that moves deals beyond one-time graphics processing unit hardware sales. AI clouds can procure Nvidia infrastructure for AI-native companies, enterprises and independent software vendors while Nvidia receives product revenue and future cloud-revenue participation from supported capacity.
Revenue-linked financing gives AI cloud providers lower upfront buildout pressure because the model ties future payment to utilization instead of only an initial GPU invoice. Unresolved terms include whether Nvidia could supply financing directly or broker third-party lending, how usage is measured, and what revenue-share percentage clouds owe if workloads succeed.
How the Financing Model Changes AI Cloud Costs
For smaller cloud providers, the new structure creates a different cash-flow path because emerging AI companies have often struggled to finance compute without carrying the full cost of a GPU-heavy buildout before customer demand arrives. Participating AI clouds sell NVIDIA-powered cloud services, and Nvidia keeps standard hardware revenue plus a recurring usage-linked percentage of cloud income.
As utilization becomes part of supplier economics, usage data, billing telemetry and cost-per-inference tracking become payment variables instead of only internal operating metrics. AI companies have been moving from training runs toward production inference, where deployed systems generate tokens continuously and idle capacity can become expensive quickly. Funding pressure also reaches inference chips outside Nvidia’s ecosystem, where private-market bets depend on whether new hardware can lower the cost of serving AI output.
Access pressure also affects startups, enterprises, research groups and regional AI players that would otherwise wait through site selection, power procurement, construction and hardware deployment. Financial guarantees to emerging GPU cloud providers exchange supplier support for future cloud revenue, while compute access depends on whether supported capacity turns into billable workloads rather than idle hardware.
Partner Scale and Market Landscape
Named partners put scale behind Nividia’s new financing model. Sharon AI and Firmus are among the first companies to work with Nvidia under the structure, giving it measurable deployments rather than leaving it as a generic supplier program.
Sharon AI adds an early AI cloud partner with a large hardware target, deploying up to 40,000 Nvidia Grace Blackwell GB300 GPUs as part of an Australia data-center buildout. Its GPU scale ties future revenue sharing to large blocks of hosted Nvidia capacity that still need enough paying users.
Firmus Technologies extends the partner examples with a DSX-aligned AI factory campus in Batam, Indonesia, expected to scale to 360 megawatts and up to 170,000 Nvidia GPUs. Together, the planned and expected partner capacity could reach roughly 210,000 GPUs if both buildouts reach those ranges.
These deployments also make Nvidia’s infrastructure vocabulary relevant to financing. AI cloud providers may use Nvidia’s DSX AI factory platform to deliver rented AI computing services. An AI factory is full-stack infrastructure for running scalable AI workloads.
Nvidia’s wider AI factory work already reaches memory and system partners through AI factory infrastructure partnerships. CoreWeave, a rent-a-GPU cloud operator, and Lambda, another GPU cloud provider, have borrowed billions of dollars to bankroll data-center buildouts, which shows why lower-upfront supplier support could appeal to smaller clouds.
What Remains Unclear
Unresolved terms center on who carries financing risk and how the revenue share is calculated. Nvidia’s model could give a company product revenue at deployment and, if a supported AI cloud succeeds, revenue participation from customer income, but public details remain thin.
Those terms matter because revenue participation could help Nvidia if future demand for new GPUs weakens while cloud utilization remains high. With up to 210,000 partner GPUs planned across Sharon AI and Firmus, Nvidia’s 2026 financing model still lacks the revenue-share percentage and direct-financing role that would price the deal for customers. Customer pricing will depend on whether the lower upfront path becomes cheaper AI cloud capacity or a long-term charge on every supported workload.
