Source: NVIDIA Blog — 2026-07-01
Summary
NVIDIA introduced a new revenue-sharing and credit-support model for expanding AI compute access, partnering directly with AI cloud providers to stand up large-scale, multi-tenant "AI factories" rather than relying purely on one-time GPU sales. The move is aimed at unblocking compute bottlenecks for startups, model builders, and enterprises that can't front the capital for large upfront GPU purchases, as AI infrastructure demand shifts from one-time model training toward continuous, at-scale inference.
Key Takeaways
- Sharon AI is deploying up to 40,000 NVIDIA Grace Blackwell GB300 GPUs under the new arrangement.
- Firmus Technologies is building a DSX AI factory campus in Batam, Indonesia, targeting up to 360 MW of capacity and roughly 170,000 NVIDIA GPUs.
- The model shifts NVIDIA's economics toward revenue-sharing with cloud partners instead of one-time hardware sales.
- Independently picked up by CNBC and Reuters/Zawya, signaling this was a widely-syndicated, market-moving announcement rather than a routine partner press release.