Why Nvidia Just Paused Its AI Cloud Revenue-Share Deals

Nvidia spent July telling smaller AI cloud companies it would help them buy chips they couldn't otherwise afford. By late August, some of its own employees were warning that the same program might be illegal.

The Wall Street Journal reported Thursday that Nvidia has paused parts of a financing initiative it launched barely two months earlier — one that offered AI cloud providers credit support in exchange for a cut of their revenue, according to a Reuters report citing people familiar with the matter. Nvidia stepped back from the program last week, according to the Journal's sourcing, though the company says it could still revamp the deal structure or fold it into something else.

What was Nvidia actually offering?

The pitch, laid out in a July 1 blog post from chief financial officer Colette Kress and AI-cloud ecosystem chief Raj Mirpuri, solved a real problem: fast-growing AI cloud operators wanted Nvidia's most expensive chips but couldn't get banks to lend against them. A cluster of Grace Blackwell GPUs worth hundreds of millions of dollars today can lose much of its value once Nvidia's next architecture ships — nobody but Nvidia knows exactly when or how much.

So Nvidia offered to absorb that risk itself. Under the program, AI clouds would sell Nvidia-powered computing capacity to customers, and Nvidia would collect twice: once from selling the chips, and again as a share of whatever cloud revenue that capacity generated. If a partner couldn't fill its GPU capacity with paying customers, Nvidia agreed to rent or buy back the unsold capacity at a set price — a guarantee that made the hardware bankable in the first place.

The first two named partners show the scale involved. Sharon AI signed on to deploy up to 40,000 Grace Blackwell GB300 GPUs. Firmus Technologies committed to a 360-megawatt AI factory campus on Batam Island, Indonesia, built for up to 170,000 GPUs — a deal its executives expect to generate $25 billion to $30 billion in committed customer business over six years. Combined, the two deals covered roughly 210,000 chips, and the framework extended a similar arrangement Nvidia struck with CoreWeave back in September 2025, a $6.3 billion commitment running through April 2032.

"AI-native companies need access to scalable, energy- and cost-efficient compute infrastructure to compete globally."

Tim Rosenfield, co-CEO of Firmus Technologies

Why did Nvidia pull back?

Two problems surfaced fast, according to the Journal's reporting. Inside Nvidia, some employees grew concerned the arrangement could draw antitrust scrutiny — a chipmaker that both supplies the hardware and takes a cut of what its customers earn from it sits in legally sensitive territory. Outside the company, some prospective partners bristled at how much control Nvidia wanted over their businesses: the company told certain customers they could only rent chips out to Nvidia-approved renters, and signaled it preferred spreading capacity across several smaller AI firms rather than handing it to one large customer.

Nvidia's public position is that nothing has really changed. The new business model we introduced in July that opens up compute access to the fast-growing AI ecosystem is still in place and continues to evolve due to high demand, a spokesperson told Reuters. That's a company saying the framework survives even if specific deals under it don't, for now.

Does this change anything for Nvidia's numbers?

Not yet, and the timing makes that clear. The pause surfaced one day after Nvidia posted fiscal second-quarter results showing revenue of $96.2 billion, more than double a year earlier, with data center sales climbing on both fronts: hyperscaler revenue more than doubled to $48.7 billion, while the AI Clouds, industrial and enterprise category — the smaller, faster-growing customers this program was built for — hit $40.3 billion, up 138% annually. Kress told analysts on the earnings call that demand from AI labs Nvidia backs directly will make up roughly a quarter of next year's business.

That's the part worth watching for anyone holding Nvidia stock through an index fund rather than following deal-by-deal headlines: the pause hits a financing mechanism, not the underlying chip sales it was designed to accelerate. What it does complicate is Nvidia's push to turn lumpy, one-time hardware sales into a recurring, usage-based revenue stream — the kind Wall Street tends to reward with a richer valuation multiple. Nvidia has leaned on a compute crunch already straining the rest of the AI industry to justify that shift; whether it can keep collecting a cut of its customers' revenue without inviting a formal antitrust look is now a separate, unresolved question sitting alongside its next earnings report.