Is it possible to capitalize on a bubble in AI? The market has already found a way

Three years have passed since OpenAI introduced the ChatGPT chatbot based on generative AI technology. It has generated euphoria, predictions of radical changes in the economy and daily life, an explosion of capital investment, a stock market boom, and, more recently, talk of a bubble. While money is still pouring into the AI field , some investors have begun to hedge against the AI bubble.

The Nasdaq-100 index of major U.S. technology companies more than doubled after the release of the first ChatGPT model. But the shares of companies in one way or another related to AI, from chip maker Nvidia to Constellation Energy, the largest nuclear power plant operator in the US, rose much more strongly.

A small number of stocks and investment ideas have trillions poured into them, says Eric Clark, portfolio manager at Rational Dynamic Brands Fund(as quoted by Bloomberg). When there is even a hint that companies associated with one of the ideas are having problems or that valuations have become so excessive that they can't continue to grow as before, all investors will "run out of there at the same time," he believes.

The first such hints have already appeared. Oracle's stock almost doubled in value from the beginning of the year to its peak in September thanks to its active entry into the AI sector. The company has received impressive orders for cloud computing (including $300 billion from OpenAI).

On December 10, after the market closed, it released its report for the second quarter of fiscal 2026, in which it reported an impressive 438% year-over-year increase in capacity orders, now totaling more than half a trillion dollars ($523 billion). But the market was not impressed, and the next day investors dropped the company's securities by almost 11%: Oracle's revenue did not reach the expectations of Wall Street, in addition, the market is increasingly concerned about the growth of its debt load. To fulfill its orders it will need significant investments in data centers, the funds for which Oracle began to attract by issuing bonds for tens of billions of dollars. Since the peak in September, its securities have already collapsed by more than 40%.

Morgan Stanley analysts are advising investors to hedge against the AI bubble by buying credit default swaps (CDS), a market-based insurance that provides a payout if a borrower defaults on its debt obligations.

But it is possible to make money on swaps without default, which Oracle is still far away from. The cost of Oracle's CDS rose from less than 40 basis points in the summer to 141 b. p. (i.e. the cost of five-year insurance against the company's default on its $10 million bond is now $141,000 a year). This is the highest level since April 2009.

Morgan Stanley notes Oracle's "lack of funding, growing balance sheet, risks associated with capital expenditures and [equipment] obsolescence, credit downgrades, counterparties, and more". Most of these risks relate to Oracle's growing financial obligations (in the form of bonds and data center leases), which the investment bank predicts will nearly triple in the next three years.

Morgan Stanley believes the CDS price could soon rise to 150 bps. "and continue to grow in the longer term". In 2026, the bank's analysts believe it could approach the 200 b. p. p. - level reached during the 2008 financial crisis.

The CDS market is becoming an important indicator reflecting investors' opinion about the bubble in the AI sector, the Financial Times writes. The volume of swaps traded by a group of technology giants related to AI has grown by 90% since September, according to the US clearing company Depository Trust & Clearing Corporation.

This is how investors insure themselves against possible defaults due to the growth of loans that companies raise for AI development.

"Those who work with credit are smarter than those who work with stocks, or at least they're worried about the right thing - how to get their money back," Kim Forrest, chief investment officer at Bokeh Capital Partners, told Bloomberg.

Meta Platforms issued $30 billion worth of bonds in October. After that, its shares fell by almost 22% in less than a month (however, then part of the fall was recovered). CDS for its debt (usually issued by banks) also appeared, and active trading in them began.

Swap trades on individual companies rose significantly this quarter, especially for tech giants building huge data centers, Nathaniel Rosenbaum, a credit strategist at JPMorgan, told the FT.

A top manager of a large American investment company also pointed out the same thing: "How to protect yourself and make a hedge? The most common way is a CDS basket on technology companies". According to him, people are increasingly active in forming CDS baskets on large technology companies or trading swaps on Oracle and Meta separately.

Meta, Amazon, Alphabet and Oracle this fall issued bonds for $88 billion to finance AI projects. According to JPMorgan's forecast, by 2030, investment-grade companies will raise $1.5 trillion for these purposes, much of it for data centers.

OpenAI alone has announced plans to spend $1.4 trillion, but its revenue is far below its operating expenses, and the company forecasts positive cash flow only in 2030, Bloomberg notes. For now, it is raising enough money - for example, it received $40 billion this year from Japan's Softbank Group and other investors.

And in September, Nvidia, the leading manufacturer of graphics processors for AI, decided to invest $100 billion in it. But this seemingly positive fact caused additional concern among investors. Market veterans began to recall how, during the dot-com bubble of the late 1990s, Internet companies invested in each other: as a result, their financial position seemed strong, but their business model remained unprofitable.

The share of the largest technology companies, which are now all actively engaged in AI, in the stock market is large - the growth in capitalization of the S&P 500 in the last three years was largely provided by techno-giants like Alphabet and Microsoft, and companies benefiting from the costs of AI infrastructure: Nvidia, Broadcom and energy suppliers, including Constellation Energy. If their shares stop rising, indexes will also stop rising, Bloomberg does not rule out.

"The correction in these stocks will happen not because growth rates will come down, but because those rates will stop accelerating further," Samir Bhasin, a principal at Value Point Capital, told Bloomberg.

Technology companies seem expensive, but their valuation ratios are still well below what they were during the dot-com boom.

The ratio of capitalization to projected earnings (P/E) of companies in the Nasdaq-100 is now 28, but at the peak of the bubble a quarter century ago it was over 89. Many companies then were young and not profitable at all. Today's market leaders are long-established large companies with huge cash flows and profits.

True, up until now much of these profits have been earned on intangible assets (software, cloud computing, advertising), but now even Alphabet and Amazon are releasing chips, and everyone is investing in very costly data centers, the cost of which will then need to be written off through depreciation.

These still aren't the ratios seen during the dot-com bubble, says Tony Despirito, chief investment officer at BlackRock:

"That's not to say there aren't some areas of [excessive] speculation or irrational euphoria, but I don't think this euphoria is about the AI-related companies in the Magnificent Seven." It includes Alphabet, Amazon, Apple, Meta Platforms, Microsoft, Nvidia, Tesla.

Some valuations do suggest that the value of tech giants has reached bubble levels, argues Ruchir Sharma, chairman of Rockefeller International.

In the bubbles seen in various markets since the 1970s, inflation-adjusted quotes have risen 10 times in 10 to 15 years, and tech stocks have recently reached that threshold, he writes in a column in the FT. And studies of bubbles over the last century indicate that the probability of collapse starts to exceed 50% when a manic sector outperforms the broad market by more than 100% over two years.

AI-related stocks are now close to that mark, Sharma notes.

The excesses and curiosities that characterize the late stage of a bubble are multiplying. Shares of the until recently little-known Indian company RRP Semiconductor rose 55,000% in the 20 months through Dec. 17 to become this year's best-performing stock. This despite the fact that it is unprofitable and has only two full-time employees, according to its annual report. RRP said in 2024 that it is leaving the real estate sector for semiconductor manufacturing. Amid the AI boom and related chips, the company's stock has become super popular on social media among a growing army of Indian private traders, Bloomberg writes.

To curb the excitement, the exchange was forced to limit trading in RRP shares to one day a week.

"However, history shows that there is no specific moment at which a bubble bursts under its own weight," Sharma writes.

This article was AI-translated and verified by a human editor