NVIDIA has introduced a new business model to accelerate the deployment of AI infrastructure by combining revenue sharing with credit support for AI cloud providers. Announced July 1 in a blog post co-authored by CFO Colette Kress, the initiative is designed to expand access to accelerated computing for AI startups, model developers, enterprises, research organizations, and regional AI providers as production inference workloads continue to grow.
The announcement reflects an industry-wide shift from primarily training large language models to operating production AI services. As AI adoption grows, infrastructure requirements are increasingly centered on AI factories that continuously generate inference tokens at scale. These environments demand rapidly deployable GPU infrastructure, high utilization rates, and multi-tenant architectures that can support sustainable inference economics.
Emerging AI companies have struggled to secure financing for GPU infrastructure. The capital requirements for large-scale AI deployments often outpace available funding, making it difficult for startups to build the compute capacity required for production workloads.
NVIDIA’s new model addresses that challenge by enabling AI cloud providers to acquire NVIDIA infrastructure under a framework that combines revenue sharing with credit support. Participating providers will deliver NVIDIA-based cloud services to AI-native companies, enterprise customers, and independent software vendors (ISVs). NVIDIA will generate revenue through both infrastructure sales and a share of cloud revenue associated with the supported capacity.
The company expects this approach to accelerate adoption of its AI platform while creating a recurring, usage-based revenue stream tied directly to infrastructure consumption.
For organizations building foundation models, operating inference services, developing AI agents, or deploying enterprise AI applications, the model is intended to shorten the time required to obtain production-scale compute resources. Customers can tap into existing AI cloud capacity instead of waiting for new data center construction, power availability, and hardware deployment.
The initiative centers on AI cloud providers building NVIDIA DSX AI factories to serve regional and enterprise AI workloads. NVIDIA identified Sharon AI and Firmus as among the first participants.
Sharon AI plans to deploy up to 40,000 NVIDIA Grace Blackwell GB300 GPUs as part of its infrastructure expansion. Firmus is developing a DSX AI factory campus in Batam, Indonesia, with plans to scale the facility to 360MW and support up to 170,000 NVIDIA GPUs.
NVIDIA also highlighted AI-native cloud platforms including Baseten, Fireworks AI, and Together AI as examples of organizations driving demand for on-demand accelerated computing. These providers support workloads ranging from model training and fine-tuning to post-training optimization and large-scale inference for enterprise and developer customers.
Overall, the initiative underscores NVIDIA’s continued focus on expanding AI infrastructure beyond hardware sales by enabling cloud providers to deliver production-scale GPU capacity to a broader range of AI developers and enterprises.
Amazon