NVIDIA's AI Financing Role Faces First Stress Test as Neocloud Stocks Plunge
nashnova research
Nvidia's off-balance-sheet guarantees for AI cloud clients have reached $36 billion, plus up to $105 billion in chip-and-power backstops for an OpenAI data center; neocloud stocks plunged over 30% this month, testing whether Nvidia's 'sell GPUs + help clients borrow' model can hold.
How much has Nvidia guaranteed — and for whom?
Nvidia's latest quarterly filing shows $36 billion in off-balance-sheet AI cloud backstop guarantees — commitments to cover losses if neocloud clients default on their financing.
On top of that, Nvidia backstops up to $105 billion in chips, land, and power for a planned 10-gigawatt OpenAI data center in Ohio, and has signed $20 billion in future data-center leases to be transferred to third parties.
This means → Nvidia has shifted from "sell chips, collect cash" to "help clients borrow, then sell chips." Its revenue increasingly depends on whether those clients can repay their loans.
In most scenarios Nvidia pays nothing unless a project counterparty defaults — but if multiple clients stumble at once, paper commitments turn into real losses.
Why did neocloud stocks crash together?
CoreWeave, Nebius, and IREN — three neoclouds in Nvidia's portfolio — each fell more than 30% this month. All three are burning cash.
CoreWeave's credit-default-swap spread — a gauge of how likely the market thinks a company is to default — has spiked above 11%. In plain terms = investors see CoreWeave's default risk as high enough to pay steep premiums for protection.
Nvidia-backed Australian AI cloud firm Firmus abruptly pulled its IPO this month. UK-based Nscale and SoftBank's SB Energy filed for US listings in September but have made no progress since.
This reflects a broader loss of confidence in the neocloud model — not just isolated company trouble, but a collective question about whether this business category works.
What is Jensen Huang's response?
At the Berlin GTC conference this week, Huang told CNBC that neocloud demand "far exceeds expectations" and the real bottleneck is working capital, not orders.
His words: "What's missing is the working capital to support growth, because demand across every area of AI is enormous and capital is extremely scarce."
This means → Huang frames the problem as "too much demand, not enough money" rather than "weak clients." But the market is answering with falling stock prices and widening CDS spreads — a different verdict.
What gives Nvidia the confidence to do this?
Nvidia generated $70 billion in free cash flow in the first half of this year — dwarfing AMD's $4 billion-plus and Alphabet's negative cash flow over the same period.
Nvidia's investment-grade credit rating lets it backstop mega-projects like OpenAI's data center, pulling down borrowing costs. In plain terms = Nvidia lends its good credit so clients can borrow more cheaply.
As of Q2, Nvidia holds equity stakes in dozens of companies — including large positions in OpenAI and Anthropic — with a portfolio valued at $99 billion.
Do pricier products deepen the financing dependency?
Nvidia's latest flagship, Vera Rubin, is a rack standing roughly two meters tall, packing 72 GPUs, weighing as much as a pickup truck, and priced above $7 million per unit according to Morgan Stanley.
This means → as unit prices leap from tens of thousands to millions of dollars, clients cannot pay out of pocket — they must finance. Nvidia's dependence on client borrowing capacity rises in lockstep with product prices.
David Trainer, founder and CEO of financial research firm New Constructs, warned: "If compute demand doesn't meet these very high expectations, we have a problem."
Nvidia's stock is down 7.6% this month, posting three straight weekly declines and on track for its worst month since November last year — whether neoclouds can regain capital-market trust is the key test for this model's sustainability.
市场有风险，内容仅供研究参考，不构成投资建议。