Nvidia pauses part of its data center revenue-share program

Nvidia has put some contracts under its revenue-sharing program for AI data centers on hold, according to the Wall Street Journal, whose reporting the German outlet Golem.de summarized. Internally, employees reportedly worried that the arrangement could draw scrutiny from antitrust regulators. Nvidia disputes that account: a company spokesperson told Tom's Hardware that the business model unveiled in July "is still in place and continues to evolve due to high demand."

How the revenue share was supposed to work

The program is known internally as DSX AI Factory, or the AI Compute Partnership. It targets operators of new AI data centers, who must spend billions before they have locked in enough paying customers. According to Golem.de, Nvidia sought to negotiate a floor rental price for its GPUs with these partners, covering the operator's expenses and depreciation; above that floor, Nvidia would keep 50 percent of any additional revenue. In exchange, the company promised to help with financing, planning and construction, and to buy back compute capacity the operator could not sell itself. Nvidia's chief financial officer, Colette Kress, described the model on the company's earnings call, as quoted by Tom's Hardware: Nvidia gets paid twice, once on the hardware sale, and again through its share of rental revenue. Unlike circular financing, Nvidia does not lend money directly; it provides demand commitments and guaranteed revenue levels instead, which Tom's Hardware says likely fed the internal worry about antitrust exposure. Per a filing with the U.S. Securities and Exchange Commission, Nvidia's commitments, typically six years in duration, totaled 36 billion dollars as of July 26, 2026.

Internal concerns

Golem.de, citing the Wall Street Journal, reports that employees raised the alarm for two reasons. First, fear of antitrust review, since Nvidia inserts itself deeply into how its customers operate. Second, the company reportedly planned to require partners to lease its GPUs only to Nvidia-approved customers. Tom's Hardware adds that Nvidia preferred to spread available capacity across multiple smaller AI companies rather than let one large customer take most of it, out of concern that a dominant tenant could end up competing with Nvidia's own projects. Several cloud operators reportedly pushed back against this interference, arguing they should retain control over which customers they serve. In response, Nvidia put some of the deals on hold, according to the report.

• Program launch: early July 2026, under the names AI Compute Partnership and DSX AI Factory.
• Mechanics: a floor rental price, plus 50 percent of revenue above it, backed by an Nvidia demand commitment.
• Scale: 36 billion dollars in commitments as of July 26, 2026, per an SEC filing.
• Allegation per the WSJ: Nvidia wanted to dictate which customers its partners could lease GPUs to, favoring many small tenants over one large one.
• Nvidia's official position, given to Tom's Hardware: the program remains in place and continues to evolve amid high demand.

A pause, or just a denial?

Tom's Hardware points out that the WSJ report itself does not establish that Nvidia abandoned the program, only that some deals were paused. Nvidia's denial, in turn, indirectly confirms that something is changing: a spokesperson told the outlet the model "continues to evolve due to high demand," which is itself a form of change, just not a full suspension. Golem.de notes it remains to be seen whether Nvidia sticks with the concept at all, since the company has already pitched the idea to a number of investment firms.

What it means for U.S. cloud and AI companies

For U.S. cloud and AI companies, the stakes are direct. Neoclouds such as CoreWeave, Lambda and Crusoe have relied on Nvidia-backed financing commitments to build out new data center capacity, and any change to who gets approved as a tenant could reshape which AI startups can actually rent that GPU capacity. The episode also lands as U.S. antitrust regulators have already been asking questions about Nvidia's dominance in AI chips; a program that lets Nvidia sell hardware, help finance the data center, and effectively vet its partners' customers is exactly the kind of vertical arrangement that invites that kind of scrutiny.