On Thursday 5 November 2026, the S&P 500 fell 4.4% to 7,353.73, its worst day since the “Liberation Day” tariff sell-off of April 2025, according to the Associated Press. The Dow Jones Industrial Average lost 1,833 points, or 3.4%, and the Nasdaq composite fell 5.5% to 24,315.00, leaving it 11.9% below the record it set in October. The heaviest losses were in the companies most tied to artificial intelligence: Nvidia fell 9.3%, Advanced Micro Devices 17.3% and the cloud provider CoreWeave 9.8%.
The AP described AI stocks as having swung for months on worries that the construction of further AI data centres may slow because borrowing has become so expensive. Two days earlier, Democratic wins in the US midterm elections had, in the AP's words, raised the possibility that governments could force pauses in the construction of more data centres. Fewer new data centres would mean fewer purchases of AI chips, and chip makers were among the biggest fallers.
The market has since steadied. On the morning of 11 November the AP reported the S&P 500 up 0.9% and the Nasdaq up 1.4%, with the S&P 500 heading for its third gain in four days. CoreWeave rose 10.7% after reporting a smaller quarterly loss than expected and raising its 2026 revenue forecast. So the question many readers are asking is a fair one: has a scare about technology spending hit the market like this before, and what happened next? It has, and two cases from the last 26 years sit at opposite ends of the range.
For scale: how big was the fall?
A 4.4% fall in the S&P 500 is large but not unprecedented in recent memory. After the tariff announcement of 2 April 2025, the index fell 4.84% on 3 April and a further 5.97% on 4 April, according to Reuters and Barchart. The sell-off of 5 November was a single day of that size, not two in a row, and it came the day after what the AP called the market's best day in three months.
DeepSeek, January 2025: a record loss, then a record high
The closest recent precedent is the sell-off of Monday 27 January 2025. The sell-off followed the release of a low-cost AI model by the Chinese start-up DeepSeek, which raised doubts about how much the industry would need to spend on chips. Nvidia's shares fell about 17% that day. Bloomberg calculated the loss in market value at about $589 billion, the biggest one-day loss for any company in US stock-market history, more than double the previous record of about $279 billion. Reuters reported that the Nasdaq fell 3.1%.
The recovery was relatively quick. Nvidia set a new record close of $154.31 on 25 June 2025, above its previous record of $149.43 from 6 January, and briefly became the first public company to be valued at $4 trillion on 9 July 2025, according to the AP. In this case the fear that cheaper AI would mean less spending did not, over the following months, show up in the demand for chips.
The telecom bust, 2000 to 2002: when the build-out really did overshoot
The more sobering precedent is the fibre-optic boom of the late 1990s. According to the Federal Reserve Bank of Dallas, companies laid about $90 billion worth of fibre-optic cable between 1997 and 2001. Demand did not keep pace: the Dallas Fed, citing Merrill Lynch, put usage of long-distance backbone capacity below 3% in April 2001, and an FDIC paper in June 2002 said only about 2 to 5% of the fibre in the United States was actually lit.
The companies that had borrowed to build that network paid the price. Global Crossing filed for bankruptcy on 28 January 2002, which Reuters called the fourth-largest in US history at the time. WorldCom followed on 21 July 2002 with $107 billion of assets, then the largest bankruptcy in US history. The Nasdaq, which had closed at a peak of 5,048.62 on 10 March 2000, fell to 1,114.11 by 9 October 2002, a drop of 78.4%. It did not close above that 2000 peak until 23 April 2015, at 5,056.06.
What is similar, and what is different
- The fear is about building ahead of demand. Both the 2000s telecom bust and today's worries centre on expensive infrastructure built in the expectation of future use. In 2000 to 2002 that expectation proved far too high; in 2025 it did not.
- Borrowing matters. The telecom failures were companies that had borrowed heavily to build capacity. The AP links today's worries directly to the cost of borrowing, which is why companies that rely on debt to build data centres are watched closely.
- The trigger is different. DeepSeek was a technology shock about the cost of AI. The November 2026 sell-off combined interest rates with a political question about whether new data centres will be allowed, which is not something either precedent had.
- One day is not a trend. The DeepSeek fall was reversed within months; the telecom decline unfolded over two and a half years. A single sharp day, on its own, did not tell investors which of those paths was coming.
What to watch
The useful questions now are factual ones. Do the companies building data centres keep, cut or raise their spending plans in the coming quarters, and do results like CoreWeave's continue to show customers paying in advance? What do new state and local governments actually propose on data-centre construction, as opposed to what markets fear they might? And what happens to borrowing costs, which the AP identifies as a central part of the worry?
Past sell-offs do not tell us how this one will end, and the two cases above ended in opposite ways. They do suggest that the evidence to follow is whether real demand keeps up with what is being built. Signal Intelligence will keep tracking those facts as they come in.