Nvidia is trying something new. Instead of just selling GPUs, it’s letting AI startups use its chips now and pay later — with a cut of future revenue.
The program, detailed in a July 1 blog post co-authored with CFO Colette Kress, turns Nvidia into a financier. Cloud partners buy the hardware, then sell access to cash-strapped startups. Nvidia takes a recurring cut of whatever those chips earn.
It’s a smart hedge. Big buyers like OpenAI have started trimming orders. This opens the door to smaller players who can’t afford their own data centers but still need compute.
The lock-in is real. Startups that take these credits stay tied to Nvidia’s ecosystem for years. Sharon AI is already installing up to 40,000 Grace Blackwell GB300 chips under the deal. Firmus is building a 360-megawatt campus in Indonesia for 170,000 more GPUs, targeting Q1 2027.
Rivals are gaining ground. China recently trained a large model without Nvidia chips. Buyers keep testing cheaper alternatives. This program widens Nvidia’s moat at a time when competition is heating up.
Morgan Stanley expects Big Tech’s AI spending to hit $800 billion in 2026 and $1.1 trillion in 2027. That’s approaching US defense budget territory. Nvidia clearly wants a piece of that recurring stream, not just a one-time hardware sale.
The question is whether startups see Nvidia as a partner — or a landlord.
