Signs that the AI ​​bubble has begun to deflate

Andreas Cervenka

This is a commentary piece. The analysis and viewpoints expressed are those of the author. 

Published 18.49

AI is the largest investment project in history.

Will it pay off?

An increasing number of people are asking themselves this question.

If the answer is no, things could get rather uncomfortable.

“Most people have no idea what they are investing in. It’s like a casino.”

These words do not come from some disgruntled AI skeptic; rather, they come from the CEO of one of the companies that has benefited the most from the AI ​​hype: chipmaker Broadcom.

At a conference this week, Hock Tan explained that many of those pouring money into AI have unrealistically high expectations.

He further noted that companies allowing their employees to freely experiment with expensive AI tools are wasting resources—likening the situation to children receiving money from their parents, only to gamble it all away on arcade games in a matter of minutes.

The fact that even someone at the very forefront of the AI ​​movement is questioning this massive expenditure speaks volumes.

What is happening within the field of AI is historic in many respects.

Never before has so much money been invested in such a short time into something that no one can be entirely certain will benefit humanity as a whole.

After all, several leading AI companies have even warned that we could face extinction. The figures are difficult to grasp. A recent analysis from the American Brookings Institution projects that AI investments in the United States alone, between 2025 and 2032, will total $10 trillion—approximately 100 trillion Swedish kronor. This averages 3.6 percent of U.S. GDP annually—a significantly higher figure than the investment poured into the development of railways, highways, telecommunications infrastructure, or power grids. The electricity demand for all these new data centers will be equivalent to the total consumption of all residential homes in the United States.

According to Wall Street analysts, the major tech companies—including Amazon, Alphabet, Meta, Microsoft, and Oracle—are expected to invest over 9 trillion kronor this year, with that figure rising to 13 trillion kronor next year.

This has triggered an unprecedented profit boom for the companies that supply the equipment required for AI.

Ultimately, however, AI services must generate sufficient revenue to justify these massive, multi-trillion-kronor outlays. Otherwise, the result could be the largest "ghost town" in human history.

According to Brookings, by 2032, annual spending on AI services will need to reach the equivalent of 9 percent of U.S. GDP for the investments in data centers to yield a reasonable return. That equates to just over 37 trillion kronor—more than what Americans spend on food. It sounds optimistic, but who knows? 

The pressure to generate revenue is particularly intense because a large portion of the invested capital is borrowed. According to the bank Morgan Stanley, this figure alone amounts to 5.7 trillion kronor for the year 2026.

If AI fails to generate enough cash flow to cover interest payments and loan repayments, losses will quickly mount.

So, how strong is the actual demand for AI? A key indicator is the performance of the major, pure-play AI companies, such as Anthropic and OpenAI—and here, questions have begun to arise.

This week, leaked information revealed that Anthropic’s revenue grew by 50 percent between the second and third quarters, reaching approximately 170 billion kronor. While this represents an incredibly rapid increase, it still marks a slowdown compared to the beginning of the year. Furthermore, this pace is slower than what many of Anthropic’s investors had anticipated as recently as this summer.

Expectations are exceptionally high, as Anthropic plans to go public in the near future with a valuation of 20 trillion kronor.

OpenAI, the creator of ChatGPT, has already postponed its initial public offering—officially citing the ongoing debate surrounding AI safety. However, even at OpenAI, growth has turned out to be lower than previously stated.
 
This week, the share prices of AI-related companies have retreated, with chipmakers like Nvidia seeing the most significant declines. After all, lower demand ultimately points to reduced sales down the line.

Japan’s SoftBank—one of OpenAI’s largest backers—has seen its shares fall by 13 percent.

An even clearer sign of unease has emerged in the credit market, among those who have lent money to fuel the AI ​​boom.

Bonds issued to finance several of the largest data centers in the United States are now trading at levels suggesting that investors do not expect to recover their full principal, the *Financial Times* reports.

Cloud services provider CoreWeave, which has borrowed large sums at interest rates as high as 10 percent, is among the companies hit hardest. Its bonds have lost value, its stock has plummeted, and the cost of insuring the company’s debt against default has climbed to levels that signal genuine concern that the firm could go under.

On Thursday, the stock recovered slightly after Nvidia CEO Jensen Huang spoke positively about the company during a visit to Stockholm.

Nvidia is simultaneously a major investor in, supplier to, and customer of CoreWeave—just one of many examples of the circular flow of capital that characterizes the entire AI boom.

The web of financial interconnections between tech giants, AI firms, and chipmakers is reminiscent of the classic game of pick-up sticks.

This also means that, in the event of a genuine crisis, virtually every company involved in AI would likely be dragged down with it.

We are not there yet, and a great deal could still happen. Especially considering the pace at which AI is developing.

But there is no doubt that the stakes are high.

Several trillion times higher than in a casino.