Nvidia says AI laboratories that it supports financially will generate roughly a quarter of its business next year. The chipmaker has already committed close to $50 billion to customers that buy its hardware and is helping assemble financing platforms intended to raise more than $500 billion from outside investors.
Chief financial officer Colette Kress gave the figures during Nvidia’s August 26 earnings call. She acknowledged that some observers would describe the structure as circular financing, although Nvidia argues that it is helping young AI companies overcome a shortage of capital rather than manufacturing demand.
The distinction matters because the same money can move through several parts of Nvidia’s ecosystem before returning as revenue.
Nvidia is financing the buyers of its own chips
The basic loop begins when Nvidia invests in an AI laboratory. That backing can provide cash directly or help the company secure credit for a data center. The facility then buys Nvidia processors, converting part of the financing into sales for the chipmaker. Stronger revenue can support Nvidia’s valuation and cash position, giving it more capacity to invest again.
Kress said Nvidia has formed partnerships with Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs and KKR. The planned financing platforms are meant to bring more than $500 billion of external capital into AI infrastructure.
That figure is not money already available. Nvidia’s results describe the partnerships as subject to definitive agreements, meaning the binding contracts remain unsigned. It is therefore a target or commitment rather than completed financing.
Nvidia has also secured land, power and construction capacity with SB Energy for sites that will use only Nvidia equipment. The first phase supports 4.25 gigawatts and is intended for OpenAI. Kress said OpenAI’s existing and planned commitments amount to about 12 gigawatts of Nvidia compute through 2030. An unnamed second laboratory will receive Nvidia credit support covering nearly two gigawatts.
Cloud guarantees give Nvidia two revenue paths
The company is extending similar support to smaller cloud operators. Nvidia promises to rent part of an operator’s capacity, creating a minimum revenue stream that lenders can use when assessing the project. If demand pushes earnings above that floor, Nvidia receives a share of the additional income.
Kress described this as a structure in which Nvidia is paid twice: first for the equipment and later through rental income. She emphasized that outside lenders still evaluate every project and that Nvidia itself is not making the loans.
The company also argues that its hardware limits the downside. Customers are either investment grade or backed by an investment-grade sponsor, according to Kress. If one fails, Nvidia says the processors can be moved to another buyer.
That protection depends on demand remaining stronger than supply. A customer failure could otherwise damage both sides of the arrangement: Nvidia could lose the value of its investment and the associated sale. Reassigning expensive equipment is plausible while buyers are waiting, but the earnings call did not establish how resilient that market would be in a downturn.
Agent demand supports an aggressive growth forecast
Nvidia says the laboratories need help because their demand for computing exceeds what their finances can support. As young companies, they lack the long contracts and credit histories that infrastructure lenders normally require.
The chipmaker expects agentic AI to intensify that pressure. Kress told analysts that an agent can require between 15 and 100 times as much computing power as a person using the same system. CEO Jensen Huang said he believes AI became predominantly agentic during the past month, although Nvidia supplied no data to substantiate that claim.
On that outlook, Nvidia guided to $108 billion in revenue for the current quarter and preliminarily projected roughly 70 percent growth in the year ending January 2028. Kress said supply, rather than demand, limits that estimate.
There are constraints inside the boom. Kress warned that memory prices are rising faster than expected and projected gross margins falling to 74 percent this quarter before reaching 71 to 72 percent in the fourth quarter. AI construction is helping create the memory scarcity that is pressuring Nvidia’s own economics.
The financing strategy can accelerate infrastructure and chip sales at the same time. It also makes Nvidia’s results more dependent on customers whose growth the company is actively funding. The next earnings report, scheduled for November 17, will offer another measure of whether that loop is broadening demand or concentrating risk.
