Nvidia has paused some of the deals through which it financially backed large GPU deployments by cloud providers in exchange for a share of their revenue, according to a report by Data Center Dynamics. The move is reportedly linked to the risk of increased antitrust scrutiny, though the company maintains that the model unveiled in July remains in place and continues to evolve.
Nvidia’s pivot in 30 seconds
- Nvidia has reportedly paused some revenue-sharing deals with AI infrastructure providers.
- The program finances large clusters and allows Nvidia to rent out GPU capacity that goes unused.
- Firmus and Sharon AI were among the first participants.
- Nvidia says the model remains active and has not been canceled.
- The company recently disclosed $108.5 billion in guarantees and other commitments tied to AI cloud infrastructure.
The report is especially significant because it shows just how far Nvidia has moved beyond simply being a chipmaker. The company is now also involved in financing, utilization, and building the infrastructure needed to put those chips to work.
The model officially announced by Nvidia on July 1 targets so-called AI clouds: specialized providers that build large accelerated-computing platforms and then sell that capacity to model developers, startups, and enterprises.
The problem is that building these data centers requires enormous amounts of capital before there are enough customers to use all that capacity.
Nvidia designed a formula to reduce part of that risk.
Nvidia Sells the GPUs — and Also Backs Their Utilization
The mechanism unveiled in July introduces a different kind of relationship between Nvidia and certain cloud providers.
The company can provide credit support and act as a backstop for deployed capacity. Under some agreements, it agrees to rent, at a set rate, any GPUs the provider fails to place with its own customers.
In exchange, Nvidia doesn’t just collect revenue from selling its infrastructure. It also shares in the revenue generated by the backed capacity.
When it unveiled the program, the company explained that it wanted to align its financial interests with those of the new AI clouds. Nvidia sells the infrastructure and also gains a revenue stream tied to its utilization.
For the cloud provider, there’s a clear upside: it lowers the risk of investing billions in accelerators without knowing whether it will manage to keep them busy.
Utilization matters enormously in this market.
A GPU built for artificial intelligence can carry a hefty price tag, but the real economic asset is keeping it working as many billable hours as possible. A large cluster with low utilization can quickly turn into a financial problem.
The first participants announced were Firmus and Sharon AI.
Firmus was developing a 170,000-GPU deployment in Batam, Indonesia, while Sharon AI had announced infrastructure with 40,000 GB300 accelerators.
The approach turns Nvidia into something more than just a technology supplier.
If the company guarantees part of the capacity, rents out any leftover GPUs, and receives a cut of the revenue, it ends up in a much closer financial relationship with the companies that go on to offer services built on its hardware.
That’s where the regulatory question comes in.
Antitrust Risk Enters AI Financing
According to the report published by Data Center Dynamics, which cites people familiar with the matter, Nvidia reportedly decided to pause some deals over the possibility that they could draw greater antitrust scrutiny.
This wouldn’t amount to a full cancellation of the program.
Nvidia told DCD that the new business model introduced in July to expand access to compute capacity remains in operation and keeps evolving due to strong demand.
The distinction matters.
For now, it can’t be said that Nvidia has abandoned this financing system or that all of its deals have been suspended. The available reporting points to a pause in some initiatives.
The possible regulatory concerns are easier to understand by looking at the position Nvidia is building around the market.
The company manufactures the GPUs, provides much of the software used to run them, and develops networking, interconnects, and complete systems. At the same time, it’s partnering with cloud providers, committing to large data center projects, and taking part in new financing structures.
On August 10, it also announced deals with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to create independent platforms aimed at mobilizing more than $500 billion in third-party capital for AI infrastructure over time.
That doesn’t mean Nvidia will directly provide that $500 billion. The initiative is aimed at attracting institutional capital toward AI compute projects.
Infrastructure is thus starting to be treated as a financial asset capable of generating revenue through the sale of compute capacity.
$108.5 Billion Shows the Scale of the Commitments
Nvidia’s financial figures show just how large this strategy has grown.
The company closed its fiscal 2027 second quarter, ended July 26, 2026, with $96.2 billion in revenue, up 18% from the prior quarter and 106% above the same period a year earlier. The data center business contributed $89 billion, up 117% year over year.
Around those results, an even bigger figure shows up.
DCD notes that Nvidia’s financial filings record $108.5 billion in guarantees and other agreements tied to the development of AI compute capacity.
Of that amount, $3.5 billion is reportedly tied to guarantees related to land, power, and buildings for AI clouds, while $105 billion corresponds to SB Energy for OpenAI’s campus in Ohio.
Nvidia officially announced its agreement with SB Energy for the PORTS-Pike technology campus in Ohio on August 17. The company will provide credit support to initially secure 4.25 GW of IT capacity tied to land, power, and buildings, with an option on another 3.75 GW. OpenAI is expected to be the customer for the full 8 GW. Nvidia also announced a $1.5 billion investment in SB Energy.
These deals show how the AI bottleneck is shifting.
During the first stage of the generative AI boom, the main problem was getting hold of enough GPUs. Now it’s also necessary to secure power, land, buildings, financing, and customers capable of keeping the infrastructure busy.
Nvidia is trying to have a hand in every one of those pieces.
The company sums it up with a phrase Jensen Huang has repeated over the past few months: compute is revenue. His thesis is that AI infrastructure can become a productive asset because it turns compute capacity into tokens that are then sold.
The model makes economic sense, but it also changes the relationship between manufacturer and customer.
Traditionally, a chipmaker sold its products, and the risk of putting them to use fell mainly on the buyer.
In some of these new deals, Nvidia can sell the hardware, back its financing, partially guarantee its utilization, and then share in the revenue it generates.
The pause reported in some of these deals suggests that this level of integration has limits the company will need to weigh carefully.
It’s also a reminder that the race to build AI data centers can’t be measured just by counting announced GPUs. Much of these projects depend on financial contracts, power commitments, future customers, and utilization levels that will have to play out over years.
Nvidia keeps growing at an exceptional pace. Its latest results confirm that demand for infrastructure keeps rising. But the more the company gets involved in financing and economically operating the clouds that buy its own chips, the blurrier the line becomes between being a supplier to the market and directly taking part in building it.
Frequently Asked Questions
Has Nvidia canceled its revenue-sharing deals with cloud providers?
No. The available reporting indicates Nvidia has paused some deals, while the company maintains that the program unveiled in July remains in place.
How do these Nvidia deals with AI clouds work?
The model combines financial backing with revenue sharing. Under certain agreements, Nvidia can back capacity and rent out any unused GPUs, while collecting a share of the revenue generated by the backed infrastructure.
How much revenue did Nvidia report last quarter?
Nvidia posted $96.2 billion in revenue for the second quarter of its fiscal 2027, up 106% year over year. The data center business reached $89 billion.
Why might this model concern regulators?
The possible concern stems from Nvidia’s growing involvement across different layers of the AI infrastructure market. Reporting on a potential antitrust review comes from sources cited by the press and doesn’t imply any regulatory finding against the company.
Source: datacenterdynamics
