NVIDIA's Revenue-Sharing Plan Opens AI Compute to Startups
Pay-as-you-grow GPU access with Sharon AI and Firmus deploying 210,000+ GPUs.
NVIDIA is transforming GPU access with a new revenue-sharing and credit-support business model. Under this plan, AI cloud providers can acquire NVIDIA infrastructure — including Grace Blackwell GB300 GPUs — without the traditional capital-intensive upfront costs. Instead, NVIDIA receives standard product revenue plus a recurring share of the cloud revenue generated on the supported capacity. This aligns NVIDIA's earnings with actual usage and accelerates platform adoption among high-growth AI-native customers.
Initial partners include Sharon AI, planning to deploy up to 40,000 GB300 GPUs, and Firmus, developing a 360-megawatt DSX AI factory campus in Batam, Indonesia with up to 170,000 GPUs. The model targets inference providers, agent platforms, and enterprises that need immediate, scalable compute for training, fine-tuning, and high-volume agentic inference. Companies like Baseten, Fireworks AI, and Together AI exemplify the demand for flexible, usage-based GPU access as AI shifts from model development to large-scale inference.
- Revenue-sharing model lets AI clouds acquire NVIDIA infrastructure with credit support, paying standard product revenue plus a share of cloud revenue.
- Sharon AI will deploy up to 40,000 NVIDIA Grace Blackwell GB300 GPUs; Firmus will build a 360MW AI factory with up to 170,000 GPUs in Indonesia.
- Fast-tracks access for AI-native platforms (Baseten, Fireworks AI, Together AI) needing immediate compute for training and high-volume inference.
Why It Matters
Democratizes GPU access for startups and enterprises, reducing time-to-market for AI deployment by removing capital barriers.