Nvidia has paused parts of the revenue-sharing programme it launched in July

Nvidia has halted some of the transactions under the financing programme it introduced eight weeks ago, in which it extends credit to AI cloud companies buying its chips and takes a cut of what those chips later earn, the Wall Street Journal reported on Thursday.

The scheme was announced on 1 July, and it does an unusual amount of work for one arrangement. Nvidia sells the chips, guarantees to rent back capacity that a customer cannot resell, and then collects a share of the cloud revenue those chips generate, meaning the company is paid at the point of sale and again along the way.

Two objections surfaced internally, according to The Wall Street Journal. Employees flagged that the structure could attract antitrust scrutiny, and they raised the related question of how far Nvidia could reasonably go in dictating how its customers run their businesses.

That second concern was not hypothetical. The company had been restricting which parties could lease the chips in question, and it preferred that capacity be spread across several smaller AI firms rather than concentrated with a single large customer, conditions that some partners found more constraining than they had expected.

Nvidia did not confirm the pause but did not dispute the programme’s existence either. “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 said, per the Wall Street Journal.

The wording leaves the company room to revise the terms or fold the arrangement into an existing programme, both of which the Journal reported are under consideration. TNW has not independently verified the report, which rests on unnamed sources.

The two deals announced at launch give a sense of the scale involved. Sharon AI, an Australian operator, signed for up to 40,000 Grace Blackwell GB300 chips, and Firmus took commitments for up to 170,000, roughly 210,000 accelerators between them.

Firmus said at the time that it expected between $25bn and $30bn in committed customer offtake during the first six years of its deployment. Sharon AI had raised $1.6bn in a private placement in June to fund its side.

The revenue-share percentages themselves have never been disclosed, which is one reason outside analysis of the programme has been thin. What is known is the shape rather than the economics.

Neither Sharon AI nor Firmus has said publicly whether its own agreement is among those affected, and the reporting did not name the paused deals. That leaves the two largest publicly known commitments in an unclear position.

Nvidia has been building this apparatus for a while, as it backstopped $6.3bn of CoreWeave capacity in September 2025, and it has since offered startups compute on deferred payment terms and assembled a $500bn financing platform with six large financial institutions behind it.

Taken together, those moves make the chipmaker something closer to a lender and landlord of the AI buildout than a component supplier. That is precisely the position that makes a competition lawyer sit up, particularly when the same company also holds more than $40bn in AI equity stakes taken this year.

No regulator has opened an inquiry into the programme, and none of the reporting suggests one is imminent. The pause, on the account given, came from inside the building.

The timing is awkward in one respect. Nvidia designed the programme to unlock purchases by cloud operators who could not raise the capital to buy chips outright, and a pause removes that route for the companies least able to find another one.

The hyperscalers are unaffected, because they were never the target. Anyone with a balance sheet deep enough to buy accelerators outright has been doing exactly that, on ordinary commercial terms, all year.

What happens next is a commercial question rather than a legal one for now. Demand for the chips has not softened, the financing gap the programme was built to close has not gone away, and Nvidia has said only that the model continues to evolve.