Nvidia's New Model: Backstopping GPU Capacity for Revenue Sharing, Evolving from Chip Seller to AI Compute 'Central Bank'

On July 1, Nvidia unveiled its "AI Compute Partnership" plan, offering financial backstops to emerging GPU cloud providers by committing to repurchase unsold GPU capacity at agreed prices, in exchange for a share of their cloud revenues. Early participants include Sharon AI and Firmus, with deployments of tens of thousands of GPUs. The model aims to unblock financing barriers for startups seeking large-scale AI computing, while reducing Nvidia's dependence on a handful of large cloud customers. Nvidia has already invested billions in equity and lease guarantees, constructing a multi-layered alignment of interests across equity, capacity repurchase, lease guarantees, and now revenue sharing.