What happened
Tom’s Hardware reports that Nvidia denied pausing its recently introduced AI Compute Partnership, a “take or pay” program designed to support new AI data-center construction. The denial followed a Wall Street Journal report claiming that Nvidia had put some transactions on hold after cloud partners objected to the company’s reported efforts to influence their customer decisions and amid possible antitrust concerns. Nvidia said the program remains in place and is evolving.
Tom’s Hardware reports that Nvidia on Friday denied a Wall Street Journal report saying the company had placed some transactions under its recently introduced AI Compute Partnership on hold. Nvidia introduced the initiative in early July, according to Tom’s Hardware, and the dispute emerged less than two months later. The report attributed the alleged pause to partner frustration with Nvidia’s attempts to influence how cloud operators ran their businesses, as well as internal concerns about possible antitrust scrutiny. The source does not independently establish that Nvidia suspended the program as a whole or identify every transaction allegedly affected.
An Nvidia spokesperson told Tom’s Hardware that the business model introduced in July “is still in place and continues to evolve due to high demand.” That statement confirms the initiative continues, but it does not resolve whether specific deals were delayed, changed, or paused. Tom’s Hardware says the Wall Street Journal report claimed Nvidia told some participating cloud providers they could lease Nvidia GPUs only to customers approved by Nvidia. The same report said Nvidia preferred to distribute capacity among multiple smaller AI companies instead of allowing one large customer to take most or all of the capacity. Those reported conditions, and the reactions attributed to cloud operators, have not been independently confirmed in the supplied source.
The partnership is structured around demand commitments rather than direct lending, according to Tom’s Hardware’s account of Nvidia’s earnings-call explanation. Nvidia can commit to rent a portion of a newly built facility’s capacity and guarantee a minimum level of revenue, helping lenders assess whether the project can support financing. In return, Nvidia can receive a share of revenue earned above the guaranteed floor. Nvidia Chief Financial Officer Colette Kress described the arrangement as creating revenue from both the hardware sale and a recurring share of rental income. The precise contract terms remain undisclosed.
Source details: tomshardware.com ↗
Why it matters
The program could affect how new AI data centers are financed and how scarce GPU capacity is allocated. Nvidia’s commitments can give lenders confidence that a facility will have a baseline revenue stream, while Nvidia may receive a share of revenue above an agreed minimum. But reported restrictions on which customers can access that capacity could raise questions about cloud-provider independence, competition, and Nvidia’s expanding role beyond selling hardware.
The immediate significance is control over access to AI compute. GPUs are a critical input for training and running AI systems, and cloud providers normally decide which customers receive their capacity. If the Wall Street Journal’s account is accurate, Nvidia’s role in guaranteeing demand may have been paired with influence over tenant selection. That could give Nvidia a say in the commercial decisions of companies that operate the infrastructure around its chips. Tom’s Hardware reports that at least some operators objected because they believed they should retain control over which customers they serve.
The financing model also illustrates how AI infrastructure expansion can depend on commitments made before a data center has secured enough customers. Modern facilities require substantial spending on buildings, power, cooling, networking, and compute hardware before revenue is established. Tom’s Hardware says Nvidia’s take-or-pay commitments can provide lenders with confidence that part of a facility’s future income is supported by Nvidia. If demand is strong, Nvidia may share in revenue above a minimum; if demand is weak, the agreement may require Nvidia to cover some shortfall or take unused capacity, depending on the contract.
The possible antitrust dimension matters because Nvidia is both a dominant supplier of AI accelerators and, through this initiative, a potential participant in the market for renting those accelerators. The source does not report a regulatory finding, lawsuit, formal investigation, or conclusion that the arrangement violates competition law. It reports only that potential scrutiny was an internal concern cited in connection with the alleged pauses. The practical question is whether the program expands compute availability while preserving independent cloud competition, or whether its conditions could narrow customer choice or favor selected AI companies.
What to watch next
The key unresolved issue is whether any individual transactions were paused or renegotiated, and whether Nvidia’s reported customer-approval requirements were part of formal contracts or proposed operating terms. Nvidia has not publicly disclosed the agreements’ detailed economic terms, the share of facility capacity covered, or the identities of participating cloud providers. Further filings, partner statements, or regulatory scrutiny could clarify how the program operates in practice.
The next useful evidence would be clarification about the difference between Nvidia’s statement that the program is “still in place” and the Wall Street Journal’s claim that some transactions were put on hold. Nvidia could disclose whether any agreements were delayed, revised, or terminated, and whether changes reflected commercial negotiations, compliance reviews, or antitrust concerns. Cloud providers participating in the program could also describe whether Nvidia imposed customer-approval requirements and how those requirements were documented.
Nvidia has disclosed that its commitments totaled $36 billion as of July 26, 2026, with agreements typically running for six years, according to Tom’s Hardware’s account of a Securities and Exchange Commission filing. However, the company has not disclosed what portion of a facility’s capacity it typically commits, the hardware value associated with the commitments, the precise minimum-revenue guarantees, or the percentage of excess revenue Nvidia may receive. Those details are necessary to judge the program’s financial exposure and competitive effects.
The story should not be read as evidence that Nvidia has abandoned its AI infrastructure strategy. The supplied report instead describes a program that Nvidia says continues while potentially changing its terms. Watch for new regulatory filings, partner disclosures, or reporting that establishes whether customer restrictions were actually implemented, how many deals were affected, and whether the initiative changes access to GPU capacity for large versus smaller AI companies. Until then, the reported backlash and alleged pause remain attributed claims rather than independently confirmed facts.
