Last week, OpenAI founder Sam Altman sat down for an interview with venture capitalist Brad Gerstner and Microsoft CEO Satya Nadella. Both are investors in OpenAI, so it seemed like a friendly audience. But Gerstner posed a question that seemed to make Altman uncomfortable.

Since introducing ChatGPT three years ago, OpenAI has posted impressive growth, but Gerstner wondered whether the company was, nonetheless, getting ahead of itself.

“How can a company with $13 billion in revenues make $1.4 trillion of spend commitments?” Gerstner asked. The commitments in question are OpenAI’s agreements to purchase computing resources. In total, they’d cost more than 100 times its current revenue. Commitments that top $1 trillion would be significant for any company, but they’re of particular concern because OpenAI has yet to turn a profit.

Altman was quick to debate Gerstner. First, he said, “we’re doing well more revenue than that.” He dismissed what he called “breathless concern” over OpenAI’s finances, and he expressed frustration at Gerstner—who is himself an OpenAI investor—for even asking the question.

“Brad, if you want to sell your shares, I’ll find you a buyer…I think there’s a lot of people that would love to buy OpenAI shares.”

In recent months, investors have been asking questions like this with increasing frequency, concerned about the economics underpinning the AI economy.

For everyday investors, these questions are important because many of the largest public companies are now heavily dependent on AI spending. At the top of the list: Nvidia. Its graphics processing unit (GPU) chips power most AI-based computers. Last week, it became the first company ever to reach a market capitalization of $5 trillion. It now accounts for 8% of the total value of the S&P 500. As a point of reference, it’s now worth more than the total value of the UK stock market. As a result, this debate has taken on more importance, so it’s worth looking at both sides.

Concerns about the AI ecosystem start with the worry that there’s a circularity to the profits these companies are generating. A little while back, Nvidia announced an investment of as much as $100 billion into OpenAI, at the same time that OpenAI is spending billions on Nvidia’s chips. Nvidia has invested in more than 100 other AI-related companies over the past two years, helping further drive demand for its own chips.

OpenAI also signed a deal with AMD, another chip maker, to buy tens of billions of dollars of AMD chips. As part of that deal, OpenAI will become a shareholder in AMD. Those are just some of the very sizable deals that have happened this year. Other complicated and interrelated deals involve Elon Musk’s xAI and a newly-public company called CoreWeave.

Beyond the potential circularity of these arrangements, there’s a more fundamental question: Sensing a parallel to the technology bubble of the 1990s, some are asking whether today’s AI spending is all being put to good use. A key feature of the 1990s bubble was the over-building of fiber optic networks, with the result that much of it went unused and billions were wasted. In the 1990s, those miles of unused fiber came to be known as “dark fiber,” leading some to ask whether today there are “dark GPUs.” In other words, are there Nvidia chips that have been sold but that are sitting dormant in a data center somewhere?

On this question, opinions differ. At a recent conference hosted by the venture firm Andreessen Horowitz, the consensus was that the notion of dark GPUs is off the mark. The speakers felt that there’s actually a shortage of GPUs. But this is an open question.

Satya Nadella has acknowledged that Microsoft does have Nvidia GPUs “sitting in inventory that I can’t plug in,” due to other constraints. It’s not clear how many, but this is another data point to consider. If there are too many surplus chips out there, it means Nvidia’s future sales may come in lower than expected. A sales shortfall would pose a risk to any stock but would pose a very significant risk to a highflier like Nvidia.

What would be the impact on everyday investors?

In the past, there was the expression that, “if General Motors sneezes, the country catches a cold.” That is the concern with these deals, and it isn’t limited to Nvidia. The so-called Magnificent Seven stocks—Nvidia, Microsoft, Apple, Alphabet, Amazon, Meta and Tesla—now account for more than a third of the S&P 500’s total market value, up from less than 10% a decade ago. So if they stumble, the overall market will stumble. That’s the risk, and that’s why it’s fair that investors want to better understand these companies’ finances.

That said, some see these concerns as overblown. While AI is hardly perfect, it’s delivering tangible productivity improvements across many industries. Among other things, AI can now create video, build spreadsheets and write computer code. New AI “agents” can even be scheduled to take actions autonomously. I’ve tried this myself and found the results remarkable. These capabilities are expanding rapidly.

How will things turn out? The reality is that no one knows for sure. Partisans on both sides of this debate make valid points. But as always, risk management should be paramount. Nvidia and its peers have helped drive the stock market up over the past two years. But because of the resulting top-heavy nature of the market, now is a good time for investors to review their portfolios. Look to see how diversified you are beyond these big tech stocks.

Do you own mid- and small-cap stocks, which carry much less exposure to these AI risks? Do you hold international stocks? Most importantly, do you hold bonds or cash which could meet your expenses in the event of a market downturn?

While stocks are still doing very well, this is a good time to take inventory.​