Nvidia’s AI Financing Role Faces First Major Stress Test as Neocloud Shares and Credit Conditions Weaken
Nvidia’s role as AI financier faces its first major stress test as neocloud shares fall and credit costs rise, with $36 billion in AI cloud backstops and a $105 billion OpenAI guarantee.
The pressure centers on a strategy by CEO Jensen Huang to use Nvidia’s investment-grade credit rating not only to sell chips but to help customers fund purchases and projects. Nvidia has provided backstop guarantees on large projects for OpenAI so buildouts can be financed at lower interest rates, and it has taken equity stakes in dozens of companies. Its portfolio was worth $99 billion as of the second quarter, including large investments in OpenAI and Anthropic.
In a July quarterly filing cited by CNBC, Nvidia disclosed $36 billion in off-balance-sheet AI cloud agreements, or backstops, for neoclouds. It also has a guarantee of up to $105 billion for chips, land and power at a proposed 10-gigawatt OpenAI data center in Ohio, and it signed as a tenant for $20 billion in future data center leases that will later be assigned to third parties. For the vast majority of these deals, Nvidia will not have to pay unless project participants default.
Market conditions are testing that structure. CoreWeave, the largest of the neoclouds, has seen the spread on its credit default swaps rise past 11%, meaning investors are paying more to insure against default. The shares of CoreWeave, Nebius and IREN have fallen at least 30% this month. Nvidia declined to comment for the CNBC story. Nvidia wrapped up a third straight down week on Wall Street on Friday, and its stock is on pace for its worst month since November of last year.
David Trainer, founder and CEO of financial research firm New Constructs, said the risk is tied to demand for computing capacity. “If the demand for compute doesn’t meet these really super lofty expectations, we’re going to have a problem,” Trainer told CNBC.
Huang reiterated his confidence in the market this week. Speaking to CNBC on Thursday on the sidelines of Nvidia’s GTC conference in Berlin, he said demand for neoclouds is “off the charts.” “What is missing in the middle is the working capital to fuel that growth,” Huang said, adding that there is “a scarcity of capital” because demand spans so many sectors of AI.
Other Nvidia-backed neoclouds have run into challenges. Australian AI cloud provider Firmus, an Nvidia investee, abruptly withdrew its IPO plans this month. U.K.-based Nscale and SoftBank-backed SB Energy, which Nvidia also backs, filed to go public in the U.S. in September but have not provided updates on their plans in weeks, CNBC reported.
Nvidia remains the dominant supplier of graphics processing units used to train and deploy AI models. Revenue in its latest fiscal year reached $216 billion, roughly eight times its level when OpenAI launched ChatGPT in late 2022. Its newest systems are far larger and more expensive than the graphics cards it sold before the AI boom: the flagship Vera Rubin product is a seven-foot rack with 72 GPUs, weighs as much as a pickup truck and costs more than $7 million, according to a Morgan Stanley estimate cited by CNBC.
Competition is emerging from Google, which rents out tensor processing units inside its cloud business, and Advanced Micro Devices, which launched its first AI rack, called Helios, in July. Cash remains Nvidia’s other advantage. The company generated $70 billion in free cash flow in the first half of this year, while AMD totaled just over $4 billion and Google parent Alphabet went cash flow negative in the second quarter to fund its buildout.
Nvidia is using that cash to offer a fuller platform of software, services, integrated hardware and, for many customers, financing assistance. Investors told CNBC that a central question is how Nvidia can use its cash and $5 trillion market value to get more chips into customers’ hands, and how much risk it will assume in the process.
An August filing cited by CNBC outlined more than $500 billion in off-balance-sheet commitments over the coming years, with a majority maturing through fiscal 2029. That same month, Huang announced a partnership with six large asset managers on a $500 billion financing push designed to treat compute infrastructure as an asset class. The agreement promised financial firms up to $125 billion in residual value guarantees on GPU financings, in addition to the equity stakes Nvidia has taken.