NVIDIA has paused part of its newly launched program for financial support and revenue sharing with AI cloud providers, according to a report from The Wall Street Journal. The move comes less than two months after the model was publicly unveiled, and follows internal discussions about the antitrust scrutiny it could draw. The decision is especially notable because NVIDIA recently disclosed $36 billion in commitments tied to AI cloud deals, part of a much larger financial structure aimed at accelerating the build-out of AI infrastructure.
NVIDIA’s about-face in 20 seconds
- NVIDIA has paused part of its credit-support and revenue-sharing program with AI cloud providers.
- The company reportedly debated internally the risk of drawing antitrust scrutiny.
- NVIDIA disclosed $36 billion in commitments tied to AI cloud deals as of 07/26/2026.
- The program hasn’t been canceled and could resurface in modified form.
- The company still holds other large commitments and guarantees tied to AI data centers.
This is about far more than a financial deal. NVIDIA was trying to build a model in which it could sell the GPUs, help the customer finance the infrastructure, and then share in the revenue generated by that same compute capacity. At the same time, NVIDIA’s commitments reduced the risk that lenders and investors took on when financing projects that need billions of dollars before they start generating revenue.
The problem is that the greater the influence the leading supplier of AI accelerators has over what infrastructure gets built, how it’s financed, and who can use it, the more questions arise about its position in a market where it already enjoys enormous presence.
NVIDIA Wanted to Earn Both When It Sold the GPU and When It Was Used
The company officially unveiled this new model on 07/01/2026.
Its pitch was fairly simple. There are startups, model developers, and companies with compute demand, but the new cloud providers need enormous amounts of capital to buy accelerators and build data centers.
NVIDIA could use its balance sheet and financial position to help unlock those projects.
The company publicly described a model based on revenue sharing and credit support. Cloud providers would buy NVIDIA infrastructure and sell compute capacity built on it. NVIDIA would collect its usual revenue from selling its products and, under certain conditions, a share of the revenue generated by that capacity.
The idea was gradually turning NVIDIA into something more than a manufacturer.
It would no longer just be selling accelerators to whoever could pay for them. It could help a project secure financing, back part of the risk, and later collect revenue tied to how that infrastructure was used.
Among the first names announced were Sharon AI and Firmus. NVIDIA presented both companies as initial participants in the new model.
According to The Wall Street Journal, the initiative has been paused after concerns about possible antitrust implications surfaced inside NVIDIA. The program isn’t said to be permanently canceled and could be reworked or folded into another initiative.
An NVIDIA spokesperson has maintained that the model unveiled in July still exists and continues to evolve in the face of strong demand.
The Problem Starts When NVIDIA Can Decide Who Gets to Use the GPUs
The antitrust question doesn’t stem solely from NVIDIA helping finance its customers.
According to the report, some of the tension centered on how much control NVIDIA wanted to retain over how that capacity was used.
The company reportedly told some providers that the accelerators could only be rented out to approved customers, and showed a preference for spreading capacity across several smaller AI companies rather than concentrating it with a single large customer.
That detail considerably changes how the model should be read.
A manufacturer that simply guarantees financing to facilitate a sale has a fairly recognizable commercial relationship. But if the leading GPU supplier for AI is also involved in financing the deal, shares in the revenue, and has a say in who can subsequently rent that capacity, its influence stretches across several layers of the market.
And that’s where the potential regulatory problem comes in.
There’s no indication that authorities have declared the program illegal, nor that any antitrust ruling exists against these agreements. What does exist, according to the reporting, are internal concerns at NVIDIA about the regulatory exposure the model could create.
That distinction matters.
The $36 Billion Showing Up in NVIDIA’s Books
NVIDIA’s latest financial results offer the first look at the scale these mechanisms had reached.
In its filing for the second quarter of fiscal year 2027, ended 07/26/2026, NVIDIA disclosed $36 billion in commitments related to AI cloud deals. These typically run for around six years.
The expected breakdown is:
| Fiscal year | AI cloud agreements |
|---|---|
| 2028 | $6 billion |
| 2029 | $8 billion |
| 2030 | $7 billion |
| 2031 | $6 billion |
| 2032 and beyond | $9 billion |
| Total | $36 billion |
The structure has one particularly interesting feature.
Cloud providers buy NVIDIA data center infrastructure, and the company commits to it through cloud services agreements. However, those providers can stop supplying that capacity to NVIDIA and sell it to third parties instead when they find more favorable economic terms.
NVIDIA’s commitments shrink as the capacity is used by outside customers or by the company itself for research and development.
In addition, when certain conditions are met, NVIDIA can share in the revenue the cloud provider earns from those third parties.
The company itself acknowledges in its financial filings that a change in market conditions could negatively affect its results.
But NVIDIA Has Another $20 Billion Tied to Data Centers
The $36 billion isn’t the only figure that matters here.
NVIDIA also discloses another $20 billion in commitments tied to data center leases for third parties that haven’t yet begun.
Combined, those two categories bring future commitments to $56 billion.
It’s worth not confusing that figure with an even larger one that shows up in the company’s recent filings.
NVIDIA also holds guarantees covering land, power, and data center construction. The maximum gross exposure disclosed for certain AI clouds reaches $3.5 billion.
On top of that comes the massive deal announced in August with SB Energy for the PORTS-Pike campus in Ohio.
NVIDIA has committed credit support to initially secure around 4.25 GW of IT capacity, earmarked for NVIDIA infrastructure under 20-year lease agreements for OpenAI. NVIDIA’s maximum obligation under the initial guarantees is capped at $105 billion, though OpenAI has agreed to reimburse and indemnify NVIDIA for any amounts it might have to pay the lessor under those agreements.
The maximum gross exposure NVIDIA currently shows across its guarantees breaks down as follows:
| Guarantees | Maximum gross exposure |
|---|---|
| Land, power, and construction for AI clouds | $3.5 billion |
| SB Energy / PORTS-Pike | $105 billion |
| Total | $108.5 billion |
That doesn’t mean NVIDIA has spent $108.5 billion, or that it necessarily will. This is maximum exposure under guarantees subject to conditions, not an immediate bill.
That distinction is essential for understanding the numbers.
NVIDIA Is Using Its Balance Sheet to Create Customers That Can Afford NVIDIA
This is one of the more interesting angles in the entire AI infrastructure boom.
NVIDIA closed its second fiscal quarter with $96.2 billion in revenue, up 106% from a year earlier. Data Center alone contributed $89 billion, up 117% year over year.
Demand remains extraordinarily strong.
But building the facilities needed to absorb that demand requires land, power connections, buildings, cooling, networking, and tens of thousands of accelerators.
Not every customer has the balance sheet of Microsoft, Amazon, Google, or Meta.
NVIDIA is using part of its enormous financial strength to solve that problem: helping new buyers build infrastructure that, in turn, will need large amounts of NVIDIA products.
From an industrial standpoint, that makes plenty of sense.
From a financial standpoint, it creates much more complex relationships between supplier, customer, financier, and end demand.
And from a regulatory standpoint, it can look even more delicate if NVIDIA also has a say in which companies end up using the capacity.
The Risk of So-Called Circular Financing
Deals like these also explain why investors and analysts are paying closer attention to what’s being called circular AI financing.
The concept doesn’t automatically imply anything improper.
The problem arises when a technology supplier finances, guarantees, or invests in the companies that then buy that same technology. Part of the supplier’s demand can end up being tied, directly or indirectly, to its own financial support.
In NVIDIA’s case, the relationships differ from deal to deal and shouldn’t be lumped together as if they were equivalent.
An investment in a startup isn’t the same as a lease guarantee. A capacity purchase agreement isn’t the same as directly financing a GPU purchase either. The scrutiny isn’t limited to cloud financing, either: NVIDIA’s reported $12.9 billion move to acquire Hugging Face shows the company pushing further into every layer of the AI stack, not just deals with compute providers.
But the growth of these structures makes it harder to answer a simple question: how much AI infrastructure demand would actually exist without the financial backing of the ecosystem’s own suppliers?
The answer matters because NVIDIA is building capacity for an industry growing at an extraordinary pace.
The company also raised its supply and capacity commitments to $279 billion, up from $119 billion the previous quarter, driven mainly by memory and manufacturing contracts intended to cover expected demand for its current and future architectures.
Pausing the Program Doesn’t Mean NVIDIA Is Giving Up on Financing AI
Here too it’s worth separating the headline from the reality.
Based on the available reporting, NVIDIA hasn’t abandoned its strategy of using its balance sheet to accelerate AI infrastructure.
What it has apparently paused is part of the model unveiled in July, built around credit support and revenue sharing with AI clouds.
The SB Energy deal remains a separate structure. Investments, supply contracts, capacity commitments, and other financial mechanisms are also still in place.
NVIDIA itself maintains that its new compute-access model is still in effect and continuing to evolve.
The pause may simply end up producing a different version in which NVIDIA has less say over the commercial decisions of cloud providers.
But the episode shows how much the company’s business has changed.
For decades, a chipmaker could focus on designing a product, having it manufactured, and selling it.
AI infrastructure is forcing NVIDIA to also think about gigawatts of power, data centers, twenty-year leases, financial guarantees, customer financing, and future capacity utilization.
The success of its GPUs has turned precisely those variables into part of the problem.
NVIDIA needs enough infrastructure to appear so it can install all the accelerators it’s able to sell. Helping finance that infrastructure can speed up growth, but the more it’s involved with the companies that buy, finance, and lease its own products, the more attention it will also draw from investors and regulators.
Pausing the revenue-sharing program doesn’t prove the model is illegal, nor does it necessarily point to an investigation. It does show that even inside NVIDIA, people are starting to weigh the limits of a scenario in which the biggest beneficiary of AI infrastructure construction is also increasingly involved in making its financing possible.
Frequently Asked Questions
Has NVIDIA canceled its financing agreements with AI providers?
No. According to the reporting, NVIDIA has paused part of its credit-support and revenue-sharing initiative while it reviews the structure. The company says the model still exists and continues to evolve.
How much has NVIDIA committed to AI cloud providers?
As of 07/26/2026, NVIDIA disclosed $36 billion in AI cloud agreements, plus another $20 billion in data center leases for third parties that hadn’t yet begun.
Does NVIDIA really have $108.5 billion committed?
That figure is the maximum gross exposure of certain guarantees, including $105 billion tied to SB Energy and $3.5 billion in other AI cloud guarantees. It isn’t money already spent.
Why could this model worry regulators?
According to The Wall Street Journal, concerns emerged inside NVIDIA about how much influence the company could exert over its customers’ commercial operations, including who gets allocated capacity. Nothing in the reporting indicates a ruling that declares these agreements illegal.
