NVIDIA is no longer content selling GPUs once and walking away. On July 1 the company introduced a revenue-sharing and credit-support model in a blog post co-authored by CFO Colette Kress. AI cloud providers procure NVIDIA infrastructure, sell services on it, and NVIDIA collects standard hardware revenue plus a slice of the cloud income that capacity generates.
So the chip becomes an annuity. NVIDIA gets paid at the point of sale, then keeps getting paid as the GPUs run. The company frames it as fixing a financing gap: small AI clouds can't easily bankroll tens of thousands of accelerators, and lenders stay wary of hardware whose resale value nobody's pinned down yet. Credit support pushes some of that risk onto NVIDIA in exchange for the upside.
Two partners are named. Sharon AI, an Australian operator listed on Nasdaq, is deploying up to 40,000 Grace Blackwell GB300 GPUs. Firmus is building a factory campus in Batam, Indonesia, scaling toward 360 megawatts and up to 170,000 GPUs. Combined, that's roughly 210,000 chips, per outside reporting, though the deployment ceilings are stated as maximums rather than firm counts.
Sharon AI CEO James Manning called it "a pivotal moment" for sovereign compute, which is the kind of line you'd expect from a company that just landed the deal. NVIDIA hasn't disclosed the terms of the backstop guarantees it's now carrying if utilization comes in soft. The structure echoes the roughly $6.3 billion commitment NVIDIA already made to CoreWeave for unsold capacity through 2032.
Bottom Line
NVIDIA now earns hardware revenue plus a recurring cut of cloud income on the same GPUs, starting with 210,000 chips committed by Sharon AI and Firmus.
Quick Facts
- Launched July 1, 2026
- Sharon AI: up to 40,000 GB300 GPUs
- Firmus: up to 170,000 GPUs, 360 MW, Batam, Indonesia
- Combined roughly 210,000 GPUs (outside estimate)
- Blog co-authored by CFO Colette Kress
