Analysis: Is Nvidia too big to fail? Its AI financing web raises the stakes

The swirl of sex-and-spying allegations around Nvidia CEO Jensen Huang is raising a worrying question: Could trouble at the chipmaker spread through the tightly connected financial network Nvidia has built to support the artificial intelligence boom?

Serious problems with Nvidia could do far-reaching economic and financial damage. The company has made hundreds of billions of dollars in financial commitments to other firms to help the industry grow, earning it the nickname the “central bank of AI.” The Federal Reserve and other financial institutions are closely watching the AI buildout to monitor for the possibility that an unexpected event could destabilize the global financial system.

Even back in July Fed officials were discussing the possibility of of a “broad-based repricing of assets” in the event of problems in the AI sector.

A rerun of the 2007-2008 financial crisis isn’t likely, according to analysts and current and former financial officials. Nvidia isn’t “too big to fail in a bank sense at all,” said Sheila Bair, a former government regulator who ran the Federal Deposit Insurance Corp. Unlike a troubled bank, Nvidia isn’t vulnerable to a sudden loss of confidence and would still produce a product customers want to buy even if something went wrong. And its massive equity gives it a cushion.

But regulators need to watch carefully for exposure by the banks and the wider financial system to Nvidia and other firms like it, Bair said.

“Nvidia going down would be a pretty seismic event,” she said.

Investors still have high hopes for Nvidia, which reports earnings on Tuesday. Analysts expect its revenue to more than double in the current fiscal year, and then to roughly double again over the following two years, according to LSEG data. But with Huang under scrutiny by the FBI, Congress, and Nvidia’s board over his relationship with a woman who The New York Times reported was suspected of spying for China, analysts see a wide range of outcomes as suddenly possible, including the potential removal of Huang, who co-founded the company in 1993.

The company didn’t immediately respond to a request for comment. Nvidia’s board has expressed confidence in its CEO.

Nvidia’s stock fell 30% from its Oct. 6 peak through Tuesday’s close, amid a broader reassessment of the AI trade prompted by Democrats’ surprising victory in the Nov. 3 midterms. Its stock recovered Wednesday, rising more than 4% as of midday. Still, people who closely track the financial flows underlying AI say that even sharper losses by Nvidia likely wouldn’t trigger the kind of systemic problems that regulators guard against.

And yet the risks flowing from AI are keeping some finance officials up at night. They are watching for the possibility that many of the factors that have helped the AI boom could reverse simultaneously. If the value of Nvidia’s GPU chips fell, for instance, it could prompt investors to ask whether companies that borrowed money to buy them would be able to pay off their debts. 

In that case, the financial assistance big tech firms have extended to others in the industry could suddenly become drags on their growth. Tech companies would have credit problems at the same time that their equity is falling. Problems among giant tech firms wouldn’t stay contained to tech investors’ portfolios, but could cause problems across the economy.

“I think that’s a real concern because both the economy and financial markets are standing on the same leg, which means a drawdown could have reinforcing effects between the two,” said Jason Wu, an assistant director at the International Monetary Fund who oversees the Global Markets Analysis unit.

Wu was the lead author of a chapter of the IMF’s Global Financial Stability report published in October that warned about the potential for growing leverage among AI firms and investors to trigger broader problems. The so-called hyperscalers — the biggest tech firms — have grown from just 1% of bond issuance in the $15 trillion investment-grade debt market to 10% in just two years. Hedge funds and retail investors are making risky bets even riskier by taking on debt to buy stocks or investing in funds that have embedded leverage.

Nvidia is at the center of all of that. The company has repeatedly stepped in to help backstop smaller AI companies known as neoclouds, which build data centers using Nvidia chips and lease capacity to other firms. Among other obligations, Nvidia disclosed in July that it has committed to buy up to $36 billion in cloud services from those companies. That helps Nvidia ensure that there will be more customers for its AI chips, and it helps those companies find easier financing.

Wu and others refer to these kinds of arrangements as circular financing. When business is growing, it can be productive for bigger firms to help their customers buy their products. But if Nvidia couldn’t honor its obligations, it would ripple through the firms that borrowed the money and the banks that lent to it. Relatively opaque private credit funds have also lent heavily into the AI boom, and how they would fare is a subject of debate among regulators.

Nvidia didn’t reveal its specific partners, but some neoclouds such as CoreWeave and IREN have separately disclosed their relationships with the bigger firm. 

Many neoclouds have been under stress in recent weeks. The cost of insuring CoreWeave debt, as measured by credit-default spreads, rose by 94% — nearly doubling — from Oct. 1 to Nov. 5, before settling, according to LSEG data. The price of an IREN convertible bond maturing in 2029 fell 23% from Nov. 3 to Nov. 5, amid a 26% decline in the company’s share price. 

Nvidia’s other obligations include a $105 billion phased guarantee to a major data center project in Ohio planned by OpenAI and SB Energy, subject to certain conditions. Nvidia promised to backstop up to $125 billion of a $500 billion credit program by Wall Street financial firms. 

These obligations come in different forms and are unlikely to come due all at once, if ever. Stijn Van Nieuwerburgh, a Columbia Business School finance professor, estimates them at $400 billion overall. 

That is a huge sum, but still a small fraction of the $4 trillion market capitalization Nvidia has even after its stock decline. “The company would have to get to the brink of extinction before all these guarantees come into jeopardy.” 

“I don’t want to say it’s a safe system, but I do want to say there’s a lot of equity in the system right now,” Van Nieuwerburgh said.

But in the past few days, Nvidia’s troubles have separated it from the pack. Most neoclouds’ equity prices have recovered this week as Nvidia’s share price has fallen. That suggests investors see Nvidia’s problems as specific to the company, for now. 

Meanwhile the basic laws of economics are driving other firms to try to move into Nvidia’s lucrative chips business. AMD, Broadcom, Intel, Google, Microsoft, and OpenAI all have efforts underway to build their own AI chips, Van Nieuwerburgh said. 

Over time, they could replace Nvidia’s products. That could cap the economic fallout of any potential troubles at the company.

And yet it’s impossible to entirely write off the financial risks of the AI trade suddenly hitting a wall, added Van Nieuwerburgh, who spoke to me over Zoom from the sidelines of a bimonthly meeting of finance officials in Basel, Switzerland. 

Central bankers “are clearly very, very, very anxious about this and thinking hard about these questions,” he said.