The AI Bubble: Much Ado About Something?

The nervousness of financial markets regarding artificial intelligence (AI) and fears of a bubble has grown significantly in recent weeks due to several moves targeting Nvidia —the company that symbolizes the current boom and the first in history to surpass 5 trillion dollars in market capitalization. First came the short selling by Michael Burry, the financier made famous by his accurate predictions on subprime mortgages that triggered the major global financial crisis since the Great Depression; then the sale of SoftBank’s entire nearly 6-billion-dollar stake by Masayoshi Son, someone who certainly understands technology; and finally the sale of a much smaller stake held by Peter Thiel’s hedge fund, another figure well-versed in bits and money. This is why, much more than usual, everyone was watching the quarterly earnings of the chip giant led by Jensen Huang with obsessive attention. Yet once again, the company surprised everyone with results even better than analysts had expected (projecting 65 billion dollars in revenue for the last three months of the fiscal year, ending in January, compared to the market’s expectation of 62.1 billion —a figure still stellar considering that only three years ago, during the same period, Nvidia recorded revenues of 7.64 billion).

Naturally, we can expect equally intense scrutiny for upcoming earnings reports —not only for the San José company but others as well— and perhaps in the meantime a financial meteor will strike the markets, or there will be a soft landing that brings tech stock prices back to normal relative to earnings (although it should be noted that on average, we are not seeing record-high ratios). Still, the anxiety that has taken hold in recent weeks —more in the media than in financial markets, to be honest— deserves some reflection.

Last week, the George Mason economist Tyler Cowen wrote in his op-ed for The Free Press that he considers concerns about a potential AI stock bubble unfounded or at least premature. He noted that “there are strong biases toward negative news in the media. People click on articles suggesting that AI is a bubble, and much of the media responds by producing that kind of content, whether it’s true or not.” Probably a consequence of AI itself and the data-driven approach of many online publications. We might therefore be witnessing a bubble about the AI bubble driven by AI itself, quite a paradox!

And yet, before shouting about another media bubble —which indeed cannot be ruled out given that equity markets have fallen only a few percentage points over the past weeks (not even 10% in total, compared with much larger gains since the start of the year)— one must ask what is so unique about the underlying technology and why it produces such oscillating feelings of euphoria and disappointment in the stock market.

The real issue is that even top experts —scientists or business leaders who earned their vast wealth through technology— hold completely opposite views regarding AI’s future trajectory. Some foresee the imminent arrival of AGI (Artificial General Intelligence) or superintelligence; others believe it will never come, or in any case, as Keynes put it, that when it arrives “we’ll all be dead.” This uncertainty is worsened by the lack of objective, universally recognized parameters that rigorously define AGI or superintelligence.

The fact remains that while Burry, Son, and Thiel are selling Nvidia (though, for example, SoftBank was quick to announce it would reinvest the proceeds into other AI companies), Berkshire Hathaway, Warren Buffett’s financial arm —one who has based his financial strategy on long-term investment, awarded by huge returns on average— has in recent weeks purchased Alphabet shares (Google’s parent company) worth 4.3 billion dollars. A strong bet on a company that has shaped the history of AI more than any other and that just days ago launched Gemini 3.0, its new foundation model competing with ChatGPT, a launch that has thus far been extremely well-received by tech experts.

In any case, it’s worth clarifying that behind the concerns expressed in the media, more than high price-to-earnings ratios, what weighs most is the scale of capital investment that tech companies are undertaking— a financial exposure that may be too large to be justified by growing but still limited revenues associated with AI services. According to a recent Morgan Stanley report, major tech companies are expected to spend 2.9 trillion dollars by 2028, financing not even half of it with their own cash reserves. It’s no coincidence that in recent months there has been an explosion in bond issuance by tech companies. By November, they had already issued more than 100 billion dollars in bonds in 2025 alone —most of it since September— while in previous years they hadn’t even approached the 50-billion mark. Cash flow has suffered, collapsing from the stellar levels of recent years.

According to some estimates, if these enormous investments —in data centers, chips, and everything needed to power AI— do not generate revenue that grows to at least 600 billion dollars annually by 2030, the whole system could fall apart. And today we are far from those levels. OpenAI, the company whose ChatGPT is the most successful AI app at least in the consumer market, is expected to close this fiscal year with 20 billion dollars in recurring revenue. Even though Altman has declared the company will generate several hundred billion by 2030, there’s currently no certainty this will happen, especially as competition grows from U.S. rivals, Chinese open-source models, and other countries eager to join the race, if only for reasons of technological sovereignty.

Two more technical factors must also be factored in.

First, an increasing number of tech companies, including well-known ones, are turning to creative (and inherently opaque) financial instruments to protect their balance sheets. We are still far from the situation seen during the subprime mortgage securitization era, but the risk is that we are moving increasingly in that direction.

Second, the collateral used in debt operations —meant to guarantee lenders in case of default— is primarily chips, the famous GPUs mostly produced by Nvidia, which have (so far) brought it enormous success. The problem is that, unlike the railroad bubble of the late 19th century —which burst twice, by the way— here the level of technological obsolescence is very high, meaning the value of this collateral is destined to decrease significantly over time. Today few believe Big Tech companies could fail, but it is no coincidence that the bonds of the only one currently with negative cash flow —and expected to remain so for years, Oracle— have suffered, offering a much higher risk premium compared to others and to U.S. Treasury bonds.

Moreover, far more money is at stake than in the past, even compared with the dot-com bubble 25 years ago. Back then, 17% of American household wealth was invested in the stock market; today it’s 21% (and up to 35% for those over 70), in a country that has grown substantially in both population and wealth. We’re talking about an enormous 42 trillion dollars held by U.S. households, plus pension funds and investment funds, with obvious real effects on the economy. The Economist has calculated that if the depth of a market correction were similar to that of the dot-com bust, reduced household wealth could lead to a drop in consumption of around 500 billion dollars, 1.6% of U.S. GDP. And the effects would be felt worldwide, both indirectly (e.g., exports to the U.S. or overseas investments by U.S. Big Tech) and directly (foreign investors own 18 trillion dollars of U.S. equities).

Still, it should be noted that unlike the dot-com bubble, which funded the premature, lavish build-out of fiber-optic infrastructure while many leading startups had bogus business models, this time we are dealing with infrastructure that can barely keep up with demand and companies that generally swim in profits, albeit with some exceptions. This is why, rather than Nvidia, it might be more appropriate to focus on OpenAI and Anthropic, companies born from the generative-AI wave, products of the hype of recent years, and still searching for a viable business model. They are not (yet) publicly traded, but the tech giants have invested massive resources in them and through them.

The real test of a potential AI bubble would therefore be whether OpenAI and Anthropic —today losing enormous sums relative to their still rapidly growing revenue— will manage to break even in the coming years. If they do, then the nervous chatter of recent weeks about an AI bubble might indeed prove to be a media-financial bubble. As we await Nvidia’s next quarterly earnings.

* Co-Founder and partner Techno Polis