Nvidia Hits Pause on AI Cloud Revenue-Sharing Deals After Internal Antitrust Warnings

Less than two months after unveiling it, Nvidia has quietly pressed pause on parts of its revenue-sharing financing program for AI cloud providers. The Wall Street Journal reported on August 27, citing people familiar with the matter, that the chipmaker stepped back from some deals in the initiative last week — a striking reversal for a program that was pitched as a cornerstone of the next phase of AI infrastructure buildout. The company says the broader business model “is still in place and continues to evolve due to high demand,” but the pause lands at a delicate moment: investors are already scrutinizing Nvidia’s deepening financial entanglement with its own customers, and regulators on multiple fronts are watching the AI supply chain.

What the program actually did

Nvidia announced the initiative on July 1, 2026, under the banner of “unlocking AI compute at scale” with capital partners. The mechanics were novel. Nvidia would extend credit support to smaller AI cloud companies — the so-called neoclouds — to help them finance GPU-heavy data center buildouts. In exchange, Nvidia would take a share of the cloud revenue generated by the chips it helped finance. Crucially, Nvidia also sought to rent compute capacity back from cloud customers if they were unable to sell it, giving those firms a guaranteed buyer of last resort — which in turn made it far easier for them to borrow the capital needed to buy Nvidia’s hardware in the first place.

Under the model, Nvidia would earn revenue twice: once on the sale of the hardware itself, and again as a cut of its customers’ cloud revenues derived from Nvidia-powered capacity. According to the Journal’s reporting, the proposed deals would have given Nvidia 50 percent of any revenue cloud providers earned through its chips beyond a certain threshold. On its earnings call this week, Nvidia said the model had the potential to drive billions in revenue over the medium- to long-term.

Why the deals got paused

The trigger, per the Journal, was internal. Some Nvidia employees expressed concerns to current and potential customers that the initiative could draw antitrust scrutiny. The sensitivities center on the extent to which Nvidia — already the dominant supplier of AI accelerators — can dictate how its customers do business once it is also their creditor and revenue-share partner.

Those sensitivities were not hypothetical. In the early weeks of the program, Nvidia irked some potential partners with the extent of control it sought. The company told some cloud providers they could rent its chips out only to approved customers, and signaled it preferred that supported capacity be distributed among multiple smaller firms rather than concentrated with one large customer. Combine that with a 50 percent revenue share above a threshold, and the arrangement starts to look less like vendor finance and more like vertical control of a distribution channel — exactly the kind of structure that attracts antitrust attention when the vendor in question controls the overwhelming majority of supply.

It is worth noting what has not changed. Nvidia says the program overall remains active, and the pause applies to some deals rather than the whole initiative. But the fact that Nvidia pulled back less than two months in — after its own staff flagged the risk to customers — tells you the concerns were taken seriously at a senior level.

The circular financing debate

The pause arrives amid mounting investor scrutiny of what critics call “circular” deals in the AI economy. Nvidia pumps money into AI labs, cloud providers and infrastructure projects; those entities use the money to buy Nvidia chips; the purchases show up as Nvidia revenue. The company’s most recent earnings disclosures put hard numbers on how big this has become: $366 billion in total future commitments across supply, capacity, cloud agreements, leases, equity investments and capex, alongside maximum gross guarantee exposure of $108.5 billion tied to large AI infrastructure buildouts. Nvidia itself estimates the AI labs it helps finance will represent roughly a quarter of its business next fiscal year.

Defenders of the model — Jensen Huang among them — argue the risk is low and the financing simply matches capital to genuine demand that would otherwise go unserved. Skeptics counter that vendor financing of this shape has a poor historical track record, most infamously the telecom equipment makers of the late 1990s who lent money to customers who then bought their gear right up until they couldn’t pay it back. The difference, Nvidia’s defenders respond, is that GPU compute currently has real, observable, multi-year demand from paying end customers. The truth is probably unresolved — which is precisely why a pause to rethink deal structure is the prudent reading of this news rather than a sign the program has failed.

What it means for neoclouds and the market

For smaller AI cloud providers, the pause removes, at least temporarily, one of the cheapest paths to capital in the market. Neoclouds that had been negotiating revenue-share terms now face more expensive or slower alternatives — traditional debt, equity dilution, or scaling back buildout plans. The better-capitalized players, such as CoreWeave and Nebius, which have their own investment relationships with Nvidia (Nvidia disclosed a 9.3 percent stake in Nebius, and made a $2 billion investment in March), are far less exposed to the pause than the smaller firms the program was explicitly designed to help.

For Nvidia watchers, the episode is a reminder that the company’s most delicate challenge is no longer manufacturing supply — it is managing the web of financial relationships that supply sits inside. Every financing structure Nvidia designs must thread between two failure modes: too little support, and the ecosystem’s growth stalls for lack of capital; too much control, and Nvidia finds itself regulatorily responsible for how an entire industry prices and distributes compute. A 50 percent revenue share plus customer-approval rights leaned hard toward the second failure mode. Pulling back from that, quickly, suggests someone in Santa Clara understood the arithmetic.

The program is not dead — Nvidia insists it “continues to evolve.” But the next version of these deals will almost certainly carry lighter revenue-share terms, fewer restrictions on who neoclouds can serve, and more distance between Nvidia the financier and Nvidia the chip supplier. For an industry that has spent two years debating whether Nvidia’s financing push is genius or risk, this week delivered the clearest signal yet that even Nvidia sees the line — and has decided not to test where it sits.