← All articles

Market Basics · 6 min read

AI Bubble? What Happened to Tech Stocks in Previous Crashes

By George Matis · Regimic ·

On 22 October 2026, the Nasdaq fell 2.6%, its biggest drop since June. The semiconductor index (SOX) fell 6.7% in a single day. The trigger was Meta: it raised its 2026 investment plans again, to $135–145 billion, and its stock fell almost 7%. Amazon, Nvidia, Alphabet and Microsoft fell with it.

The question is being asked more and more often: is artificial intelligence a bubble, like the internet in 2000? Nobody knows for sure. But we do know what happened to tech stocks the previous times the market turned sharply.

How much is being spent on AI

The numbers are unprecedented. According to Moody's, the big cloud companies will spend about $785 billion on investment in 2026, approaching $1 trillion in 2027. Amazon alone plans about $220 billion, Alphabet $195–205 billion, Microsoft about $175 billion and Meta $135–145 billion. Gartner estimates that total AI investment worldwide will reach $2.7 trillion in 2026.

The market's question is not whether AI is useful. It is whether this money will bring in enough revenue, fast enough.

Three signs that worry the market

  • Investment is exceeding cash. In the third quarter of 2026, Meta invested $34.9 billion, more than the $32.8 billion it generated from its operations. It also announced it will raise up to $95 billion to fund AI.
  • A few companies carry the whole market. The gap between the regular S&P 500 and its version where every company has the same weight is the widest in 35 years (UBS, May 2026). Three companies, Alphabet, Amazon and Meta, explain about 70% of the increase in earnings expectations for 2026.
  • Interest rates are rising. The Fed raised rates in September and held them at 3.75%–4.00% on 28 October, with two members dissenting. The 10-year U.S. Treasury yield is at its highest since 2002. High rates hurt most the companies that expect their profits in the future.

2000–2002: the dot-com bubble

The Nasdaq peaked at 5,048.62 on 10 March 2000 and fell 78% by October 2002. It got back to that level only in 2015.

The most telling example is Cisco. In 2000 it was the company selling the equipment the internet was being built on, much like Nvidia today for AI. It peaked on 27 March 2000 and lost about 90% of its value. It closed at a new record only in December 2025, at $80.25, more than 25 years later. The internet succeeded. But anyone who bought Cisco at the top waited a quarter of a century.

2008: the crisis that did not start in tech

In 2008 the cause was banks and mortgages, not technology. Even so, the Nasdaq fell about 48% from 2007 to March 2009. When the economy falls into recession, tech stocks fall along with everything else, however good the companies are.

2020: the fast crash

In February and March 2020, as the pandemic hit, the Nasdaq fell sharply. By June 2020 it was already back at a record, within about three months. Central banks cut rates to zero and demand for technology soared, as everyone worked and shopped from home.

2022: when rates went up

In 2022 the Nasdaq fell 33%, its worst year since 2008. There was no big collapse of companies. The reason was that the Fed raised rates very fast to stop inflation. Meta and Tesla lost about two-thirds of their value and Amazon about half. The Nasdaq got back to its high in early 2024.

What is different from 2000

  • Valuations are much lower. In late 2025, the Nasdaq-100 traded at about 26 times expected earnings. At the dot-com peak it was over 80 times.
  • The big companies make real profits. In 2000 many internet companies had no revenue at all. Today Nvidia, Microsoft and Alphabet are among the most profitable companies in the world.
  • But there are similarities too. Huge investment in infrastructure before the revenue shows up, borrowing to pay for it, and a market that rests on a few names. Some smaller AI companies trade at valuations that recall 2000.

What the previous times suggest

Three patterns repeat:

  • A real technology does not protect against a crash. The internet changed the world, yet the Nasdaq took 15 years to recover.
  • How long the fall lasts depends on the backdrop. In 2020 rates fell and the recovery came in months. In 2022 rates rose and it took more than a year. In 2008 a recession came.
  • The "suppliers" of the revolution are not safe. Cisco sold to every player on the internet, and it still fell 90%.

Four episodes are not a statistical sample, and AI is not the internet of 2000. History does not tell us whether there is a bubble. It tells us which questions to ask.

What to watch

Five questions, and how to read the answer:

  • Investment versus revenue: is AI revenue growing as fast as AI spending? If revenue keeps up with investment, the bubble risk shrinks, because the money is paying off. If investment runs far ahead of revenue, the risk grows, as it did with the internet companies in 2000.
  • Borrowing: are companies funding AI from their profits or with debt? As long as they pay from profits, a disappointment costs less. If debt rises, the risk grows: debt has to be repaid even if revenue comes late, and with high rates it costs more.
  • Market breadth: are other stocks rising too, or only a few big names? If gains spread across more sectors, the market rests on firmer ground and the risk shrinks. If everything depends on a few companies, a stumble by one of them can pull down the whole index.
  • Rates and inflation: the next CPI data comes out on 10 November and the Fed decides again in December. If inflation eases and rates stop rising, the risk shrinks, as in 2020. If inflation persists and the Fed raises rates further, the risk grows, as in 2022.
  • The market's mood: the big tech falls coincided with a change in the overall mood, from risk-on to risk-off. As long as the mood stays risk-on, falls usually remain corrections. If it turns risk-off, the risk of a bigger fall increases.

None of these alone says whether there is a bubble. But when several point the same way, the picture becomes clearer.

Where Regimic fits

Regimic does not say whether there is a bubble or which stocks to buy. But it shows what, as we saw, mattered most in every previous crash: the overall market backdrop. Whether the market is risk-on, neutral or risk-off, and how markets reacted the previous times after big events, such as Fed decisions and inflation data.

Prices are nominal. The Nasdaq's 2007–2009 decline and its 2024 recovery date are based on monthly averages. Data as of 28 October 2026.