NVIDIA introduced a new AI cloud business model on July 1 that gives emerging AI clouds a way to finance NVIDIA-powered capacity while giving NVIDIA a share of the cloud revenue those systems generate.
The company says the model combines revenue sharing with credit support. AI clouds sell NVIDIA-powered services to AI-native companies, model builders, enterprises, software vendors, research organizations, and regional AI players. NVIDIA earns standard product revenue and a share of cloud revenue on the supported capacity.
That makes the announcement more than another GPU-capacity post. NVIDIA is trying to turn AI factory buildouts into a broader economic relationship with cloud operators, not just a hardware sale.
The business model is the news
NVIDIA’s post describes the problem plainly: emerging AI companies have historically had limited access to capital-intensive infrastructure, and even long-term customer commitments may not be enough to unlock compute financing.
The new structure is meant to change that. The cloud provider can procure NVIDIA infrastructure for downstream customers. NVIDIA supports the economics through credit support and takes a usage-linked share of the cloud revenue on that supported capacity.
This matters because AI infrastructure is becoming a utilization business. A data center full of accelerators is valuable only if customers can keep it busy with training, fine-tuning, post-training, high-volume inference, or agent workloads. NVIDIA’s model ties part of its upside to that downstream demand.
It also changes how readers should understand “AI factory” announcements. Capacity is no longer just a count of GPUs or megawatts. The financing model, customer mix, and utilization path are part of the product.
Sharon AI and Firmus are the first test cases
NVIDIA named Sharon AI and Firmus among the first companies working with it under the model.
Sharon AI is deploying up to 40,000 NVIDIA Grace Blackwell GB300 GPUs. Firmus is building a DSX AI factory campus in Batam, Indonesia, that NVIDIA says is expected to scale to 360 megawatts and up to 170,000 NVIDIA GPUs.
Those are large numbers, but the useful point is not only scale. NVIDIA is trying to make regional and emerging AI cloud capacity easier to fund, then align the supplier with the revenue that capacity creates.
The customer examples in the post point to a specific demand profile: Baseten, Fireworks AI, and Together AI need cloud capacity for model training, post-training, fine-tuning, and agentic inference. Their customers need flexibility as products move from pilots into production.
The risk is circular dependence
NVIDIA’s post presents the model as a way to accelerate compute access. That is credible. It can help cloud providers secure financing and help AI companies avoid waiting through site selection, power procurement, construction, and hardware bring-up.
The harder question is how much downstream demand is independent of NVIDIA’s support. If NVIDIA sells hardware, helps support the buyer’s financing, and then takes a share of the buyer’s cloud revenue, investors and customers will want to know whether reported capacity reflects durable end-user demand or a supplier-supported buildout cycle.
That does not make the model weak. It makes the model important to track. AI infrastructure economics are moving from simple capex headlines toward contract structure, utilization, and who carries risk when capacity is slow to fill.
