Nvidia pulls back AI cloud revenue-sharing program amid antitrust and control concerns

Nvidia has paused parts of a two-month-old financing program that gave the chipmaker a direct share of cloud providers' rental revenue on top of its hardware sales, after employees warned internally that the arrangement's reach into how customers run their businesses could invite antitrust scrutiny, according to reporting by the Wall Street Journal confirmed by multiple outlets. The retreat touches a financing structure that has become central to how smaller "neocloud" operators are able to afford the billions of dollars in Nvidia chips needed to build AI data centers.

The company has not abandoned the initiative outright. A spokesperson said the broader business model unveiled in July "is still in place and continues to evolve due to high demand," and people familiar with the matter told reporters Nvidia could still restructure the program or fold it into a different arrangement. But the pause, which reportedly took effect last week, applies to new deals under the specific revenue-sharing structure while Nvidia reconsiders how much control it can exercise over customers without crossing legal lines.

The numbers

Nvidia introduced the program, sometimes referred to as the AI Compute Partnership, on July 1. Under the structure, Nvidia would guarantee a base hourly rate covering a cloud provider's costs and agree to rent back any capacity the provider could not sell, while collecting roughly half of any revenue earned above that floor. The company's first two named partners were Sharon AI, an Australian cloud provider planning to deploy up to 40,000 Grace Blackwell GB300 GPUs, and Firmus Technologies, which is building a 360-megawatt AI factory campus in Batam, Indonesia, designed to house up to 170,000 GPUs — roughly 210,000 accelerators combined.

Nvidia's own quarterly report filed with the Securities and Exchange Commission for the period ended July 26 disclosed that commitments under the AI cloud agreements totaled $36 billion, structured to run through 2032 and beyond, on top of $279 billion in supply and capacity commitments and $366 billion in total future obligations. Those figures sit alongside a separate, larger initiative Nvidia announced in August: a set of independent financing platforms built with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR intended to mobilize more than $500 billion in third-party capital for AI data centers — a program Nvidia has kept distinct from the revenue-sharing deals now on hold.

How the program came together

The pause follows a run of extraordinary results that have made Nvidia's balance sheet central to the entire AI buildout. The company's fiscal second-quarter earnings, reported on August 27, showed revenue of $96.2 billion, up 106 percent from a year earlier, with data center revenue climbing 117 percent to $89 billion. For the first time in the company's history, Nvidia gave investors a full year of forward guidance, projecting roughly 70 percent revenue growth into fiscal 2028 — a signal of confidence that also raised the stakes on how it finances the buildout underneath that growth.

The revenue-sharing model was Nvidia's attempt to solve a financing bottleneck facing so-called neoclouds: smaller cloud operators that need to spend billions on Nvidia chips and data-center infrastructure but lack the balance sheets of hyperscalers like Microsoft, Google or Amazon to do so on their own. By offering a revenue guarantee, Nvidia made it easier for these firms to borrow against future cloud income — while positioning itself to earn twice on every GPU: once on the equipment sale, and again on a cut of the rental income the chips generated.

That structure quickly drew internal unease. According to the Journal's reporting, Nvidia employees raised concerns with both current and prospective partners that the deal's terms — including a requirement that customers lease capacity only to Nvidia-approved buyers, and a preference for spreading GPU capacity across several smaller firms rather than concentrating it with one large customer — gave Nvidia a degree of say over how independent companies run their businesses that could attract regulatory attention.

Who is affected

The immediate effect falls on neocloud operators that had been counting on the financing model to underwrite expansion. Shares of publicly traded cloud providers with exposure to Nvidia-linked GPU financing, including CoreWeave, Nebius and IREN, fell in the sessions after the pause was reported, with IREN among the hardest hit. The move also lands amid broader investor anxiety about "circular" financing arrangements in the AI sector, in which a chip supplier's capital effectively flows back to itself through equity stakes, revenue guarantees and offtake agreements — a pattern some analysts have flagged even as they remain bullish on Nvidia's underlying demand.

The pause also arrives while Nvidia is already navigating scrutiny of its market position more broadly. The Department of Justice has been investigating whether Nvidia's dominance in AI chips and its deal structures with customers violate antitrust law, an inquiry that predates the revenue-sharing program but forms the backdrop against which Nvidia's own employees appear to have judged the new initiative too risky to expand without changes.

"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."

That statement, attributed to an Nvidia spokesperson, is the company's only public comment so far. Nvidia has not detailed which specific deals were paused, how many customers were affected beyond Sharon AI and Firmus, or what changes, if any, it plans to make to the program's terms.

What people are saying

Coverage of the pause has largely framed it as a self-correction rather than a regulatory setback: Nvidia is reported to have stepped back on its own initiative, before any formal government action targeted the program specifically, after weighing legal exposure against the commercial upside of guaranteeing revenue for smaller cloud partners. Industry observers have noted that the underlying problem the program was designed to solve — undercapitalized cloud operators struggling to finance GPU purchases at the scale Nvidia's roadmap demands — has not gone away, meaning some version of vendor-assisted financing is likely to persist in one form or another.

• Program launched: July 1, 2026, with Sharon AI and Firmus Technologies as first partners
• Scale: roughly 210,000 GPUs across the two initial deployments
• Disclosed commitments: $36 billion in AI cloud agreements per Nvidia's SEC filing
• Separate $500 billion financing platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR remains unaffected

What happens next

Nvidia has left open the possibility of restructuring the revenue-sharing arrangement or absorbing it into a different financing product, rather than shutting it down entirely. For now, the company's statement that the broader compute-access model "continues to evolve" leaves ambiguous whether new deals under the original terms will resume, be renegotiated with lighter customer restrictions, or be replaced by a structure closer to the asset-manager-led platforms it unveiled in August. Investors and neocloud operators are likely to look for clarity when Nvidia next reports results or discloses additional financing agreements in subsequent SEC filings, while the pause is expected to feed into the wider debate in Washington and on Wall Street over how much control a dominant chip supplier should have over the companies that depend on it for compute.