The Bubble Warnings Are Getting Louder. What Would You Do With a Forecast?

Last Thursday was a reminder of how fast the mood can turn. The Nasdaq fell 694 points, or 2.6%, to 26,008.86, its steepest drop since early June. The S&P 500 fell 1.21% to 7,663.43, and the Philadelphia semiconductor index lost 6.7% in a single session.¹

The trigger was Meta. It beat Wall Street’s estimates and still fell 7%, after reporting its first cash outflow in more than a decade because of AI spending. The same day, reports of renewed U.S.-Israeli attack plans and a deadly attack in Saudi Arabia pushed Brent to $113, a five-month high.² Investors were being asked to weigh AI spending, oil and war in the same session.

The warnings are piling up

Thursday did not come out of nowhere. On October 7, the day after the S&P 500 set a record, the 30-year Treasury yield touched 5.71%, its highest since 2002.³ The same day, Ray Dalio said artificial intelligence shows the hallmarks of a classic bubble nearing its bursting point, driven by debt and rising rates.⁴

Two days later, Joachim Klement of Panmure Liberum told Bloomberg that “the AI bubble will either burst in 2027 or in 2028.” His 2027 target for the S&P 500 is 5,000, which implies a 36% drop and would be the worst crash since 2008.⁵ On October 13, the International Monetary Fund used its Global Financial Stability Report to warn of an AI-driven correction and of circular financing among AI companies.⁶

None of these people know exactly what comes next. That is the point. The market is at a stage where serious, credible voices disagree sharply about the next few months, and the swings in between are large enough to hurt. Some of them may be wrong. But the number of serious warnings in a single month is itself a sign of how uncertain the outlook has become.

✓ Trusted partner One financial technology firm is offering a free look at a forecasting system it says projects where thousands of American stocks are headed weeks in advance, with no email or credit card required →

The question every investor is really asking

Most investors spend their time looking at what already happened. Price charts, earnings reports and news stories all describe the past. The money is made on what happens next.

The money is made on what happens next.

That is why the ordinary investor holding through a week like last week is not really asking whether AI is a bubble. They are asking a narrower and more useful question: what are the stocks I own likely to do over the next few weeks, and should I be holding, adding or getting out? Answering it well would change how they handle a week like last week. Someone who knew a stock was likely to recover within weeks would hold through the drop. Someone who knew it was likely to keep sliding would sell before the damage got worse.

For most of market history, the honest answer was that nobody could know. That answer is starting to change, not because markets have become simpler, but because prediction itself has become a different discipline.

AI is getting good at the unpredictable

Look at what predictive AI is doing outside of finance. This month, researchers from Emory, Beth Israel Deaconess and Stanford published a study showing that an AI score built from a single overnight heart recording could predict heart failure. Each step up in the score raised the risk by 1.69 times in one group and 2.33 times in another, across 22,386 patients.⁷

Earlier this year, models built at the University of Michigan-Dearborn flagged more than 99% of failing power lines and more than 97% of the areas that would lose power in a simulated blackout on a test grid.⁸ These are systems with millions of moving parts, the kind of problems people once assumed were too chaotic to forecast. What these systems share is the ability to find patterns in more data than any person could process, and then to check those patterns against what actually happens.

✓ Trusted partner See the research behind this story →

Finance is no exception. A Harvard-led study published through the National Bureau of Economic Research found that a neural network could predict about 71% of the trades made by professional mutual fund managers. The 29% it could not predict were more often linked to outperformance.⁹ Even the professionals are more predictable than they think.

A separate study by researchers at Princeton, Harvard and Stanford found that additional training more than doubled a small AI model’s stock-forecasting score, enough to edge past a simple momentum strategy.¹⁰ Not every attempt works. In one public test, eight leading AI chatbots were each given $10,000 to trade tech stocks for two weeks, and their combined portfolio lost about a third.¹¹ General-purpose chatbots are not forecasting systems. Purpose-built ones, trained and tested on market data, are a different thing.

Investors are already using it, carefully

Ordinary investors have noticed. In a SoFi survey of 1,000 U.S. adults conducted in February, 75% of investors said they use AI, and 86% of those users have acted on what it told them. Only 26% said they trust it a lot.¹² People want the help, but they are not willing to take it on faith.

That gap between use and trust is healthy. A forecasting tool should earn trust by showing its work. The useful ones make specific, checkable predictions, and let people test them before relying on them. A vague promise to beat the market is easy to make. A precise forecast for a specific stock on a specific date is not, because anyone can see whether it came true.

With bubble warnings getting louder and daily swings getting larger, a tool that shows where your stocks are likely headed over the next few weeks is no longer a luxury. It is the most practical kind of help there is, as long as you can see it work before you trust it. That is a much higher bar than most market commentary ever has to clear.