Nvidia launches revenue-sharing model for AI startups - qz.com

NVIDIA has launched a revenue-sharing program aimed at providing AI startups, model builders, and enterprises with access to its computing infrastructure without requiring upfront capital investment i... Read original →

NVIDIA’s revenue-sharing model is less a financing gimmick and more a surgical strike at the capital barrier throttling AI democratization. Technically, it locks the Grace Blackwell GB300 into an inescapable ecosystem, forcing software stacks, compilers, and interconnect protocols deeper into CUDA dependency—squeezing out RISC-V or open-source AI accelerators. Compliance-wise, deployments in Indonesia slash power costs but expose partners to EUV export controls and local data-center sovereignty laws, especially as the U.S. and EU tighten scrutiny on AI infrastructure. Against Google’s TPU v5e and Amazon’s Trainium2 vertical integration, NVIDIA counters with 'compute-as-equity,' using ecosystem stickiness to neutralize hardware substitution. Within 18 months, this will spawn asset-light AI cloud operators—but also misalign GPU supply: surging inference demand amid slowing training cycles may bottleneck CoWoS advanced packaging capacity at TSMC, paradoxically elevating Taiwan, China’s strategic role in the supply chain.