Research

AI data-center spending meets a 97% market bubble warning

A Cambridge and London Business School model assigned a 97% probability to bubble-like conditions in September 2026, matching 2007. The score signals resemblance, not a crash forecast.

170 Years of Market History Just Flashed a 97% Bubble Warning — the Last Time It Was This High Was 2007
170 Years of Market History Just Flashed a 97% Bubble Warning — the Last Time It Was This High Was 2007FotoRieth / pixabay
By Sophie Lindqvist6 min read

Why it matters

  • A model estimated a 97% probability of bubble-like conditions in September 2026, matching its 2007 reading.
  • Researchers analyzed 795,964 The Economist articles spanning 170 years.
  • The study compared warning signals with 22 documented bubbles across 10 countries.
  • The S&P 500 had fallen about 6% from its October 6 record; the Nasdaq Composite was down 12%.
  • Margin debt had reached a record $1.5 trillion earlier in 2026.

By Sophie Lindqvist

One model assigned a 97% probability that financial markets showed bubble-like conditions in September 2026, matching its reading around the 2007 financial crisis.

Researchers at the University of Cambridge and London Business School produced the estimate in Crying Wolf: Media Warnings and the Predictability of Financial Bubbles. They used AI to analyze 795,964 The Economist articles published over 170 years, then compared language about speculation, asset values and financial risk with 22 documented bubbles across 10 countries.

24/7 Wall St. published the analysis on November 18, 2026, as Wall Street spent much of the year debating whether AI had created another speculative mania. The source says technology companies were pouring hundreds of billions of dollars into data centers. For investors, the warning matters because it bears on what they pay for future growth, how much a portfolio depends on a few technology companies, and how leverage can magnify a sell-off.

24/7 Wall St. notes that bubbles are hard to recognize while underway. Rising share prices encourage investors to take more risks, while strong corporate profits can make unreasonable valuations look justified until prices collapse.

How did researchers build the warning?

The study focused on three kinds of language: speculation, asset valuations and financial risk. Researchers compared those signals with 22 documented market bubbles spanning 10 countries. The resulting scores measured how strongly conditions around each episode resembled conditions around known bubbles.

24/7 Wall St. presented the model’s readings as follows:

  • 1889 Baring crisis: 89%
  • 2000 dot-com bubble: 92%
  • 2007 financial crisis: 97%
  • September 2026: 97%

In plain terms, the model finds a historical pattern in market coverage, but it does not identify the next market peak.

What does a 97% reading actually mean?

The result came with a wide range. The study produced other estimates as low as 40% for the current market, which 24/7 Wall St. said underscored the uncertainty.

“97% doesn’t mean there’s a 97% chance stocks will crash,” 24/7 Wall St. wrote. “It measures the model’s assessment that current conditions resemble historical bubbles.”

The number therefore describes similarity, not a countdown. It does not identify when prices might fall, how far they might travel or whether a major decline will follow. The source also listed 97% as the model’s reading for the 2007 crisis, making September 2026 the highest comparison in its four-entry table.

The practical point is simple: 97% describes resemblance to a bubble, not a 97% probability of a crash.

Why do the warning and the market pull in different directions?

The study also found a poor record for published crash calls. Warnings in financial commentary implied an average 20% probability of a major market collapse within six months. Such declines occurred just 5.7% of the time, meaning commentators anticipated crashes roughly 3.5 times as frequently as they happened.

24/7 Wall St. invoked the old maxim: “As the old saying goes, economists predicted nine of the last five recessions.” The source treats that gap as an argument against selling stocks whenever someone predicts disaster.

But the researchers found that sustained warnings about speculation frequently coincided with genuine bubbles. The study linked continuing warnings to episodes before the 2000 technology collapse and the 2008 financial crisis, even though isolated predictions often failed.

The September 2026 signal also arrived alongside measurable weakness. According to MarketWatch, the S&P 500 had declined approximately 6% from its October 6 record, while the Nasdaq Composite had fallen 12%. The source says those declines do not establish that a crash is coming, but they suggest investors are becoming less willing to pay premium valuations for future growth.

24/7 Wall St. also points to a record $1.5 trillion in margin debt reported earlier in 2026. Borrowed money can amplify selling when prices decline, giving existing market weakness a potential mechanism for a sharper move.

The balance is clear: individual crash calls often mislead, while persistent warning patterns deserve attention when leverage is rising.

How does AI spending fit into the warning?

24/7 Wall St. says the biggest vulnerability is not necessarily AI itself. It is the price investors pay for companies expected to profit from AI, especially as hundreds of billions of dollars flow into data-center infrastructure.

Nvidia (NASDAQ:NVDA), Microsoft (NASDAQ:MSFT), Amazon (NASDAQ:AMZN) and Alphabet (NASDAQ:GOOGL) are building businesses around AI infrastructure, cloud computing and software. Unlike many dot-com companies in 2000, these businesses generate billions in annual profits.

Even profitable companies can become poor investments when expectations outrun future returns. The source says Michael Burry has questioned whether enormous AI infrastructure commitments will generate sufficient returns, particularly when customers can reduce spending faster than suppliers can cut their costs. His concern reinforces the parallel with earlier speculative excesses.

In plain terms, solid profits can support an AI investment, but only when the purchase price leaves room for weaker returns.

What should investors do with the signal?

The source offers modest guidance rather than a command to sell:

  • Review how concentrated a portfolio is across AI-related companies.
  • Avoid excessive borrowing.
  • Keep enough cash or bonds to cover near-term expenses.

24/7 Wall St. warns that selling everything because of one warning could prove as costly as ignoring it. A portfolio does not need a perfect market forecast to benefit from checking its exposure to a sector correction.

A $100,000 portfolio concentrated in AI stocks would lose $30,000 in a 30% sector decline. Diversification cannot eliminate market losses, but it can reduce dependence on one industry’s fortunes. The source also says history does not promise another 2008, though it offers reason not to assume the AI boom will continue indefinitely.

The practical response is to reduce portfolio fragility, not to make an all-or-nothing bet on a crash.

Sponsor material: The source also carried a sponsored Fisher Investments guide for 401(k) and IRA planning. It describes 401(k)s as powerful retirement tools while warning that poor fund choices, ignored small fees and rollover mistakes can quietly cost savers money for decades. The free guide covers rollovers, contributions, required withdrawals and mistakes that do the most damage to retirement accounts.

What should investors watch now?

The study does not supply a crash date, a target for the S&P 500 or a verdict on any individual AI stock. Its useful output is narrower: it identifies current conditions that resemble earlier bubbles while leaving substantial uncertainty over timing and severity.

That makes the next step practical rather than predictive. Investors still need to test their assumptions about AI-related profits, spending commitments and valuations against the possibility of a 30% sector decline. The next decision is not whether to abandon stocks on a 97% score, but whether their portfolios can survive that decline without forced selling.

Source: 247wallst.com

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.

777 articles

Related articles

  1. Gartner Warns $3.5 Trillion in AI Investment Could Turn Worthless by 2030
  2. Nasdaq 100's 5.5% Drop Puts $1 Trillion AI Bet on Trial
  3. AI Spending Fears Test a Market Still Near Its Highs
  4. Michael Burry Says Nvidia's Real Problem Is a $176 Billion Accounting Gap
  5. Forrester: AI bubble fears premature, but math is getting harder

« Previous articleNext article »