NVIDIA has unveiled a new business model that grants AI cloud providers direct access to its DSX AI factories — purpose-built facilities housing thousands of Grace Blackwell GPUs. Instead of requiring startups to secure massive capital for infrastructure, NVIDIA will share revenue and provide credit support, letting partners like Sharon AI and Firmus deploy capacity rapidly. Sharon AI alone plans to install up to 40,000 GB300 GPUs under the arrangement.
The shift reflects a broader transition in AI economics. As workloads move from training to continuous inference, demand for "token factories" that generate output at scale has outpaced traditional cloud procurement cycles. Site selection, power contracts, and hardware bring-up can take years; NVIDIA's model compresses that timeline to months.
For model builders, agent platforms, and enterprises, this means faster onboarding to full-stack accelerated computing without multi-year commitments. NVIDIA gains a recurring, usage-linked revenue stream tied directly to token production rather than one-time hardware sales. Regional AI players also benefit, as the multi-tenant design supports sovereign compute requirements across jurisdictions.
The first DSX factories are already under construction. Whether this partnership structure becomes the default financing layer for AI infrastructure remains to be seen.
