Nvidia's Balance Sheet props up much of the AI boom. Jensen Huang's exit tests how much of that rested on one man

Central banks run on credibility as much as on capital. A governor who promises to do whatever it takes seldom has to prove it, because markets take him at his word. Nvidia, which investors have taken to calling "the Central Bank of AI", has worked in much the same way. On November 12th the man whose word underwrote it, Jensen Huang, resigned as chief executive of the company he co-founded 33 years ago. For investors the question is no longer only who will design the next generation of chips. It is who now stands behind the promises.

Mr Huang's departure, effective immediately, followed a week of damaging reports. The first concerned an affair with Alice Huang, a businesswoman whom American intelligence suspects of spying for China. She has worked for Megaspeed, a Chinese firm that buys Nvidia's products. The second concerned an affair with a senior Nvidia employee. Mr Huang has denied compromising national security or company secrets, and the board has hired an outside law firm to investigate. Colette Kress, the chief financial officer, is now interim boss.

How the bank works

Nvidia sells most of the world's AI accelerators. Over the past three years, though, it has become much more than a vendor. It has taken stakes in its customers, backstopped data-centre leases and helped to finance purchases of its own chips. It has put at least $30bn into OpenAI alone. It owns part of Perplexity, an AI search firm, and holds Equity in xAI, Elon Musk's AI venture, now folded into SpaceX, which declared in August that it would buy only Nvidia's chips. Its obligations to other firms are estimated at around $400bn.

The logic is circular and, while a boom lasts, virtuous. Nvidia invests or lends. The recipient buys graphics processors. Nvidia books the revenue, its share price rises and its balance sheet swells. Then it invests or lends again. In the fiscal year when Ms Kress joined, in 2013, Nvidia's Revenue was about $4bn. In fiscal 2026 it was $216bn. Its shares have risen roughly 14-fold since ChatGPT appeared in late 2022.

Vendor financing is not new. Telecoms-equipment makers lent heavily to their customers in the late 1990s, and the unwinding was brutal. What sets Nvidia apart is scale, and the fact that much of its web of customers was built personally by one man. Mr Huang seeded the market with buyers beyond the big cloud providers: specialist "neoclouds", start-ups and national AI projects. He vouched for them, and they in turn vouched for demand. "He's connected financially to everybody," says one executive recruiter.

Testing the guarantee

The market's first verdict was harsh. Nvidia's shares fell by as much as 14% on November 12th, pushing its Market Value below $4trn. The stock is heading for a sixth straight weekly decline, its longest losing run since 2012, and Nvidia has already slipped behind Apple in the ranks of the world's most valuable firms.

The companies most dependent on Nvidia's chips, and often on its money, fared worse. Nebius fell by 17%, though its own disappointing guidance that day played a part. CoreWeave dropped by nearly 12% and IREN by 8%. In effect these firms are leveraged bets on continued access to Nvidia's products and goodwill. When the guarantor wobbles, so do they.

Not a bank, but not nothing

The comparison with a central bank has limits, and they offer some comfort. Nvidia takes no deposits, so it cannot suffer a run in the classic sense. Officials at the Federal Reserve are watching closely but see no imminent threat to the financial system. A former bank regulator who lived through the 2008 crisis says the firm is not too big to Fail "in a bank sense". If things go wrong, losses would fall on shareholders and on lenders to AI companies, not on taxpayers.

The subtler danger is a quiet tightening of credit. Even a modest change in Nvidia's appetite for financing its customers would ripple outwards. A finance chief turned interim chief executive, facing congressional scrutiny and an internal investigation, has every reason to be cautious with the balance sheet. If she is, the marginal buyer of GPUs loses its financier, and some Demand that looked organic may prove to have been partly self-funded.

There are already hints that Capital is becoming choosier. OpenAI has raised $15bn at a valuation of $1.25trn, according to reports, after originally seeking up to $30bn at $1.4trn. Next week Anthropic, its chief rival, is due to go public, giving stockmarket investors their first chance to put a daily price on a frontier AI lab. Both events will show whether money is still flowing freely into the industry Nvidia has done so much to finance.

What to watch

The next test comes on November 17th, when Nvidia reports quarterly results. Ms Kress has sat beside Mr Huang on every Earnings Call for 13 years. This time she will be alone. Investors will look past the headline numbers to the plumbing: how large the firm's commitments and guarantees have grown, whether new investments in customers are slowing, and whether guidance assumes that the financing machine keeps running at full speed.

Institutions usually outlast their governors, and Nvidia's technology, customers and cash pile do not vanish with its founder. But a lender of first resort depends on the belief that it will keep lending. For three years that belief rested heavily on one man in a leather jacket. Nvidia now has to show that it rests on the company itself.