Nvidia paused its own chip-financing program after its employees flagged an antitrust risk -- weeks after trimming a separate $250 billion guarantee

Nvidia has paused new deals under its AI Compute Partnership, the financing program it launched in July 2026 to help smaller AI cloud providers buy its chips, the Wall Street Journal reported August 27, citing people familiar with the matter. Nvidia's own employees had warned current and potential customers that the program's terms -- which let Nvidia dictate which customers a provider could resell capacity to -- could draw antitrust scrutiny. It is the second Nvidia financing commitment scaled back inside of two weeks.

The mechanics, as reported: Nvidia and a cloud provider agree on a base hourly rate covering the provider's own costs, and Nvidia collects 50% of any revenue the provider earns above that threshold on Nvidia-powered capacity. Nvidia also promised to rent the capacity back itself if a provider couldn't otherwise resell it -- a backstop meant to let smaller, less-capitalized providers commit to chip purchases they couldn't otherwise finance. Nvidia disclosed $36 billion in total commitments under the program in its latest quarterly filing, spread across agreements that typically run six years. CFO Colette Kress told investors the arrangement had the potential to bring in "billions of dollars" for Nvidia over the medium to long term.

Two cloud providers were named as the program's first participants when it launched: Sharon AI, which committed to 40,000 Grace Blackwell GB300 GPUs, and Firmus Technologies, which committed to 170,000 GPUs for a 360-megawatt campus in Batam, Indonesia. (GB300 is the newest SKU in Nvidia's current Blackwell generation -- the chip tier data centers are racing to secure allocation of right now.) Neither company, nor Nvidia, has said whether those specific deals are among the ones now paused; reporting describes the pause as covering new agreements going forward, not a wind-down of what's already signed.

What tripped the pause, according to the Journal's sourcing, wasn't the size of the commitment -- it was a control clause. Nvidia required providers to rent chips only to customers Nvidia had approved, and signaled a preference for spreading capacity across many smaller AI companies rather than concentrating it with one large buyer. Antitrust law scrutinizes exactly this pattern when a dominant supplier starts deciding who its own customers are allowed to sell to. The commercial logic isn't hard to follow: a favored, well-capitalized customer guaranteed unlimited compute at a subsidized rate has less reason to build or buy from a Nvidia competitor, and a customer base kept deliberately fragmented stays financing-dependent on Nvidia rather than growing large enough to become one itself.

This is the second Nvidia financing commitment trimmed inside of two weeks. Nvidia had separately been negotiating what reporting in early August described as a $250 billion loan guarantee for OpenAI's proposed Ohio data center -- by mid-August, that figure had shrunk to $105 billion, backstopping only the project's first phase, after investors pushed back on how much of Nvidia's own balance sheet the company was putting behind a single customer's build-out.

Being the chip supplier, the bank, and the landlord in the same deal is the arrangement that drew Nvidia's own employees' antitrust concern.

Both retreats sit inside a busier pattern: in roughly the same three-week span, Nvidia also announced a financing alliance with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR meant to mobilize over $500 billion in third-party capital for AI infrastructure -- credit risk carried by outside banks and asset managers, not by Nvidia's own contracts or balance sheet. That structure sidesteps the exact question that appears to have triggered this pause: nothing in a bank's own lending terms hands Nvidia the power to decide which of its customers may resell capacity to whom.

None of this reads as Nvidia stepping back from financing the AI buildout -- the $500 billion bank alliance and the now-smaller OpenAI guarantee are both still live. It reads instead as a company correcting a specific, control-heavy deal structure before a regulator had to say something first, in favor of structures where Nvidia collects fees and interest rather than veto power over who its customers can sell to. A spokesperson says the broader "business model" introduced in July is unchanged; what's unresolved is whether the specific revenue-share deals -- the ones that made Nvidia supplier, financier and landlord in the same contract -- come back in a form that keeps the antitrust exposure, or one that doesn't.