When the artificial intelligence sector’s overwhelmingly dominant chipmaker, Nvidia, announced quarterly results this week, many took this as a reason for confidence in the economic future of AI. Not me. My primary reflex, in fact, was fear.

On their face, Nvidia’s results were unambiguously spectacular. Reported net profit doubled from a year ago to roughly $60 billion, compared with $6.2 billion just three years ago. The company’s revenues have doubled accordingly, soaring to $96.22 billion. Although it is already the world’s largest company in terms of its $5 trillion-plus stock market value, Nvidia said that it expected next fiscal year’s revenues to rise by a further 70 percent.

I am not rooting here for a comeuppance for Nvidia. Whatever I might think of AI as an intellectual problem or public policy matter, my pocketbook, like that of many millions of Americans (and others) who now own small portions of the company as direct shareholders or as retirement account and fund investors, depends on it sustaining its value. Beyond personal finance, though, as a global financial danger, Nvidia’s outsized heft has become a big and multidimensional problem.

In the narrowest terms, our sudden collective Nvidia dilemma is almost child’s play to describe. Nvidia’s chips are expensive and in extremely high demand, but the companies buying them may not eventually be able to afford them.

As the stock investor advisory service the Motley Fool noted on Wednesday, “The company makes its own best products obsolete on purpose. It ships a new chip architecture almost every year. Each replacement is so good that people fight to buy it at a higher price.”

At first glance, this model sounds perfect. This year and next, just four companies—Amazon, Google, Meta, and Microsoft—are projected to spend $1.5 trillion building data centers, which are sure to be full of Nvidia’s chips. According to some estimates, those four companies, combined with Nvidia, constitute around 23 percent of the country’s entire stock market in terms of worth. The only problem is that few of Nvidia’s largest customers that are racing to build these enormously expensive centers are making money, despite their colossal ongoing investments.

To keep the party going, Nvidia has begun entering into co-financing and investment arrangements with some of its largest chip customers—becoming, in effect, their partial underwriters. At present, the biggest example of this phenomenon seems to be one of the largest and most ambitious global players in AI, OpenAI.

After Nvidia announced its latest results, CEO Jensen Huang was at pains to give this financial support as positive a spin as possible. “The big picture is that we’re going through this platform shift and it affects every computer company. These will be some of the most consequential technology companies in history.”

Maybe so, but OpenAI and some of its peers have so far struggled to monetize AI in a way that keeps pace with the costs of their continuing hardware investments. Citing Naveen Chhabra, a principal market analyst at Forrester, a research firm, the Wall Street Journal recently reported that “despite surging demand for Nvidia’s products, there remains a persistent gap between how much big tech companies are spending on data centers and other computing infrastructure and how slowly profits are being generated by the end products of AI.” Meanwhile, one of OpenAI’s front-line competitors, Anthropic, has labored to find paying customers for its most advanced AI models, which of course rely on Nvidia’s most powerful chips.

A principal reason for this is that Chinese AI companies have figured out a way to turn U.S. export restrictions on Nvidia’s chips to their advantage, building far cheaper AI models that are built with less expensive, domestically produced chips and open-source architecture. Chinese companies have faced difficulties making AI pay for itself, too, but this means they are working from a lower cost basis, and one that undercuts their U.S. competition.

This has had the ironic effect of causing Nvidia to invest in developing so-called open-weight AI models of its own that will potentially keep costs down and compete with China’s open-source pioneers, such as DeepSeek. The problem is that this has the potential of undercutting the big U.S. AI companies that Nvidia has increasingly begun to support. If it is smart strategy for Nvidia to cover its bets, this is hardly reassuring about the overall health or direction of the United States’ AI industry.

It is precisely the growing incestuousness of the AI sector that worries me most, with Nvidia becoming the indispensable banker and potential savior to a growing number of AI companies and the data centers that they rely upon to host their services.

This dynamic reminds me, above all, of other financial arrangements in another country in another era: That country is Japan, where I arrived as a reporter for the New York Times in 1998, right on the heels of the bursting of an economic bubble that it has struggled to recover from ever since. If one is looking for a perfect analogy, this is not it. However, these eras share enough in common to warrant deep concern.

By the height of its boom years, toward the very end of the 1980s, Japanese assets rose to dizzying and irrational heights. After years of climbing, the Tokyo stock market rose more than 60 percent between 1988 and 1989 alone, and real estate prices inflated accordingly. By comparison, the U.S. S&P’s market value has only come close to doubling since the present bull market began in 2022. The market’s present forward price-to-earnings ratio of roughly 23 may be very high by historic U.S. standards, but it is still only about a third of where Japan’s ratio stood in December 1989.

Nonetheless, a central feature of the Japanese economic system back then bears eerie resemblance to what is happening in AI today. Japan’s big industrial groups held shares in each other’s stocks, which helped bid up their values. They also cross-financed each other and refrained from selling stocks even in the face of market downturns. Financing, in other words, became more a function of business relationships than objective measures such as cash flow or other sober, reality-based market analysis.

For those who think that what occurred in Japan could never happen in the United States, it is worth considering that the AI boom is still in its infancy, as is the type of circular financial architecture that has begun to characterize Nvidia’s relationship to some of its most important customers. With each new boffo quarterly report, Nvidia assuages doubt and acquires new believers. But make no mistake: We, meaning the entire global economy, are along for a ride, which may or may not end well. That’s how big the company has become. That’s how central AI has recently become to U.S. economic performance. And that’s how important the U.S. economy is in today’s world.

Another reason to be wary is that despite predictions of utterly transforming employment, revolutionizing medicine, and solving the most difficult mathematical problems, among other prodigies, the jury remains out on whether the AI revolution will pay for itself in the real economy. This is a question that sharply divides leading economists. Some, such as Massachusetts Institute of Technology Nobel Prize laureate Daron Acemoglu, estimate that it will boost U.S. productivity by less than 1 percent over the next decade, while others, such as the University of Virginia’s Anton Korinek, say that it may radically boost annual GDP growth.

The house of cards that Japan’s bubble was built upon was made of different stuff. But what we know beyond a doubt is that when its securities market burst in December 1989, it would not be until February 2024—34 years later—that stock valuations there matched their previous peak. Whether the circular financial architecture taking shape around Nvidia ends differently is a question no quarterly earnings report, no matter how bullish, can answer. For most of us, there is little more we can do than hold our breath.