Nvidia has denied a Wall Street Journal report claiming it put some transactions under its recently introduced 'take or pay' AI Compute Partnership on hold. The report came less than two months after Nvidia unveiled the initiative in early July and just days after the company detailed the effort during its earnings call. A Wccftech report citing the same Journal story said Nvidia decided to pause the initiative last week, though the program has not been scrapped and could resume.
According to the Journal's reporting, Nvidia had told some cloud providers participating in the program that they could lease its GPUs only to Nvidia-approved customers, and the company reportedly preferred to spread capacity across multiple smaller AI companies rather than let one large customer take most of it. Some cloud operators pushed back, arguing they should control which customers they serve, and this reported friction, combined with internal concerns about potential antitrust scrutiny, was cited as a reason transactions were paused.
Nvidia disputed the characterization of a pause. A company spokesperson told Tom's Hardware that '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.' Nvidia's statement indicates the program continues to exist but is changing, without confirming that any deals were frozen.
Under the AI Compute Partnership model, Nvidia commits to a take-or-pay arrangement covering a portion of a facility's capacity along with a minimum revenue guarantee, which is meant to give lenders confidence to finance AI data center buildouts. In exchange, Nvidia shares in a portion of the facility operator's revenue earned above that guaranteed floor. Nvidia CFO Colette Kress described this as a model in which the company 'gets paid twice: once on the hardware sale, and again through the share of rental revenue.' If demand is weak, Nvidia's take-or-pay obligation requires it to cover the gap between actual revenue and the guaranteed minimum, or alternatively to rent back unused capacity for its own needs.
Nvidia disclosed $36 billion in these six-year agreements as of July 26, 2026, according to an SEC filing cited by Tom's Hardware. Separately, Wccftech reported that Nvidia's latest earnings disclosed $108 billion in commitments, guarantees, and other agreements—including financial guarantees, credit support, financing arrangements, and data center leases—tied to its broader efforts to enable AI cloud computing, with partners including Sharon AI and Firmus. Nvidia's SEC filings note it may earn a share of revenue from supported capacity, but that lower-than-expected AI compute demand or pricing could reduce that revenue.
The two reports differ slightly in framing: Tom's Hardware emphasizes that the Journal's report does not establish Nvidia abandoned the program, only that some deals were paused, while Wccftech frames the move as Nvidia 'quietly freezing' its credit support and revenue-sharing initiative following internal talks about antitrust exposure. Both outlets note Nvidia has not disclosed the specific economic terms or capacity percentages involved in these agreements.