Is the AI ​​Stock Bubble About to Burst?

The situation is starting to get a little absurd. Once again, technology stocks suffered heavy losses this week — the third recent wave of turmoil. They may have even presented a modest stabilization this Friday (6), but this instability seems more worrying than the previous episodes of nervousness.

Once again, but now with greater conviction, investors are asking the same question: "Is it for real this time? Is the AI ​​(artificial intelligence) bubble finally bursting?"

I don't claim to know the answer. Ask again in a week or so and we'll know if this swing will deepen and spread or if it's just another false alarm. But the fact that we're all asking the same question again says a lot.

It is clear that artificial intelligence technology works. There is no discussion about that. But the internet also worked very well — and yet it caused markets to collapse 25 years ago. (Sorry if this makes you feel old).

Any reasonable person can identify signs of unsustainable euphoria in AI-related assets, especially in stocks in the sector, which have a dominant weight in American and global markets. There are several possible sources of tension: high debt levels, increasingly complex financing structures, doubts about the real revenue generation potential, and difficult-to-explain examples of circular financing.

As the BIS (Bank for International Settlements) recently pointed out, almost 29% of investment agreements for AI companies involve other companies in the sector, and more than half of the resources invested in these companies come from other AI companies. In a crisis, circular finance can easily turn into a confrontation where everyone attacks each other.

All of this was already known and widely pointed out for months as a possible source of a market correction. What was never clear was what could trigger it.

An American policy might finally be that trigger. The initial reaction to the historic midterm elections held this week in the United States, which gave significant victories to the Democratic Party in Congress and in state races, was quite positive.

Stock markets rose around the world, partly on the relief of a clear and uncontested result, partly on the hope that stronger checks and balances on an unpredictable president could return some normality to the markets, and partly on the expectation that elections would hasten the end of the seemingly senseless war in Iran.

The price of oil fell to less than US$100 (R$550) per barrel, after reaching US$118 (R$649) at the end of last month. Voters from different parts of the United States sent a clear message of dissatisfaction with high energy prices, driven by the war, which make the cost of living — including fueling the car — more expensive. The market's assessment is that, one way or another, the war needs to end, and quickly.

But the good mood in the markets lasted just one day. This is because voters have also made it clear that they do not like artificial intelligence and, specifically, that they do not want large data centers, which consume enormous amounts of energy and water, built in their neighborhoods.

AI was on the ballot in several states, with public consultations on proposals to postpone or restrict the construction of these facilities. In almost every case, voters turned against big tech companies.

A much stricter regulatory and licensing environment lies ahead for an industry that supports an alarmingly large share of the American economy and global markets.

Frank Flight, strategist at Citadel Securities, stated in a note on Thursday (5) that, in theory, at least part of these projects will simply be transferred from less receptive states to more favorable ones. Still, they must come out of the paper.

Perhaps. But this remains a problem for projects financed with large amounts of debt. "Securing alternative locations, power supplies and approvals takes time. Therefore, this adjustment could still significantly delay deployment and increase costs," he said.

"We therefore expect some moderation in the short-term boost to growth and inflation arising from construction and the demand for resources associated with it, as well as a smaller impact on accumulated investment and financing needs."

Diversified technology companies, particularly those that have already secured capacity in data centers, are faring relatively well in this new wave of turbulence. But companies with businesses focused on expanding AI infrastructure have suffered significant losses.

"At these valuations, AI investment is like a shark that needs to keep swimming to survive," said Trevor Greetham, head of multi-sector investments at Royal London Asset Management.

Ahead of the midterm elections in October, Dario Perkins of research consultancy TS Lombard outlined a scenario for what could be a true collapse of AI investment.

Naturally, much of this analysis involves speculation, but he estimates that, in an extreme scenario, American stocks could lose at least 30% of their value, while credit losses would reach US$200 billion (R$1.1 trillion), accompanied by "a wave" of debt defaults linked to the technology sector.

In the case of stocks, "these are conservative estimates, which would take the technology's share of the American market back to levels prior to the launch of ChatGPT," he said.

Government bonds and stocks from markets outside the United States stand out as possible hedging alternatives, he added. It's a worrying analysis, but also a plausible one — and, tellingly, US Treasury bonds have already started to appreciate.

A quick end to the war in Iran remains the best possibility for markets to regain solid footing. But at some point, the warning triggered by investments in AI will no longer be just a false alarm.