AI cloud stocks tumble, putting Nvidia’s financing guarantees under pressure
The chipmaker has promised to buy unused computing capacity and support major data-center projects. Those commitments can become costly if customers cannot pay.
Loading page…
The chipmaker has promised to buy unused computing capacity and support major data-center projects. Those commitments can become costly if customers cannot pay.
Listen to this story
Nvidia’s customer-financing model is under scrutiny as AI cloud providers face a sharp rise in borrowing costs. Its disclosed commitments include $36 billion in AI-cloud backstops and $20 billion in future data-center leases, generally payable only after defaults. The largest guarantee—up to $105 billion—supports OpenAI’s planned Ohio data center; if OpenAI cannot pay, Nvidia may need to assume the lease, find another tenant, or cover a sale shortfall. Strong demand and Nvidia’s cash generation may help, but exposure also depends on customer credit and GPU resale values.
CoreWeave has about $51.6 billion in total debt, including $35 billion in notes, bonds, and loans; some 2031 and 2032 bond yields topped 15%.
Nvidia has committed to buy up to $6.3 billion of CoreWeave computing capacity if no other buyer takes it.
The Ohio guarantee can end if OpenAI posts its own collateral or reaches a sufficiently high credit rating.
Nvidia’s financial support for the AI buildout is facing a market test. CoreWeave, Nebius and IREN shares have each fallen at least 30% in October, while yields on some CoreWeave bonds topped 15%. The pressure reaches beyond those cloud providers: Nvidia has promised to buy unused computing capacity and backstop infrastructure serving its customers.
CNBC’s October 24 report puts the chipmaker’s exposure at the center of the selloff. CoreWeave alone carries about $51.6 billion in total debt. The spread on its credit-default swaps—contracts investors use to insure against a default—has climbed past 11%, another sign of concern about its ability to meet obligations.
Nvidia does more than supply equipment to neoclouds, the companies that rent out AI computing capacity. It also invests in customers and uses its investment-grade credit standing to help projects borrow at lower rates. That support can make construction possible before operators have enough money to finance it themselves.
The protection takes different forms. Nvidia may guarantee a minimum resale value for chips or a minimum price for renting their computing capacity if a lender must recover its money. At CoreWeave, Nvidia has committed to buy up to $6.3 billion of compute if no other buyer takes it.
These promises are not all immediate cash expenses. For the vast majority of the arrangements described by CNBC, Nvidia pays only if project participants default. Its disclosed commitments include $36 billion in AI cloud backstops and $20 billion in future data-center leases that it expects to assign to third parties.
Nvidia’s largest single guarantee covers up to $105 billion for chips, land and power at OpenAI’s planned data center in Pike County, Ohio. The agreement requires Nvidia hardware for at least 20 years. If OpenAI misses payments or becomes insolvent, Nvidia has several ways to meet its obligations.
The guarantee can also end: OpenAI could post its own collateral or achieve a sufficiently high credit rating. That makes Nvidia’s eventual exposure dependent on both the customer’s finances and what the facility could fetch from another tenant or buyer.
Jensen Huang argues that financing, not customers, is the missing ingredient. Speaking to CNBC on October 22, Nvidia’s chief executive called neocloud demand “off the charts” and said working capital was needed to fuel growth. He also said CoreWeave did not need Nvidia’s backstop and that its cost of capital was falling.
Nvidia has substantial cash generation behind that position: it produced $70 billion in free cash flow in the first half of this year. But the downside depends partly on hardware values, not just today’s demand. Nvidia is seeking insurers to absorb some risk from a potential decline in GPU values.
Moody’s vice president David Gonzales offered a qualified counterweight to the market anxiety. He expects chips to retain values above their contractual minimums in the near term, which could limit losses in many structures. Moody’s does not count guarantees as debt when rating corporate buyers, but Gonzales said it continually monitors how much exposure investment-grade AI companies can handle.
Loading discussion...
Join the conversation
Explain when supporting customers becomes too much financial risk.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.