Find this research useful? Add aibubblequestion.com to your Google Preferred Sources (opens in a new window), one click, affects only your own results.

The AI Bubble Question
All pages

Foundations

What Is a Bubble?

A bubble is not just 'prices went up a lot.' It is a repeatable pattern with phases you can name, measure, and compare across two centuries.

Updated

Here is the working definition this site uses: a bubble is a period when the price of an asset rises far above what its realistic future earnings can support, powered by borrowed money and by the belief that prices will keep rising, followed by a collapse when that belief breaks. The rise can take years. The collapse usually takes months.

Financial historians describe the same arc again and again, from railways in the 1840s to crypto in 2021 (Stanford GSB (opens in a new window), Quinn & Turner (opens in a new window)). This site draws it the same way on every case study:

The arc of a typical bubble

InflationYears of quiet, real growth

Mania1-3 years of frenzy

PeakOften within 18 months of rate rises

Trough1-13 years later

RecoveryYears to decades

Fall: 50-90% is typical

Peak to trough: Months to a decade

Back to even: Often 10-25 years

The five phases

1. Inflation

Something genuinely new appears: a railway, a fiber network, a chatbot that writes. Early investors do well because the growth is real. Prices rise for good reasons, and this phase can look identical to a healthy boom, because at this point it is one.

2. Mania

The story takes over from the numbers. Money arrives faster than the underlying business can absorb it. Capital spendingMoney a company spends on long-lived physical things: buildings, machines, chips, data centers. Spent now, paid back (hopefully) over years.Full definition in the glossary accelerates, new investment vehiclesThe flow of newly created investments being sold to the public: IPOs, new bonds, new funds. Issuance waves tend to crest right around market tops.Full definition in the glossary multiply, ordinary people pile in, and borrowed moneyMoney investors borrow from their brokers to buy stocks. It amplifies gains on the way up and forces selling on the way down.Full definition in the glossary amplifies everything. Skeptics start publishing warnings, and get ignored, often for years. In several historical cases, circular financing arrived at this stage and stretched the mania 2 to 4 years past what independent demand supported.

3. Peak

Peaks rarely announce themselves. Historically the most common companion is a central bank raising interest ratesWhen a central bank raises interest rates or drains money from the system to slow the economy. Most historical bubbles peaked during or within 18 months of tightening.Full definition in the glossary: in six or seven of eight major bubbles since the 1840s, the peak came during or within roughly 18 months of a tightening cycle (NBER (opens in a new window)). The warning lights page tracks the indicators that historically preceded peaks and what they read today.

4. Trough

The fall is measured as a drawdownHow far a price falls from its peak to its lowest point, in percent. The dot-com crash was a 78% drawdown for the Nasdaq.Full definition in the glossary: peak to troughFrom the highest point to the lowest point of a decline. Used to measure both how far prices fell and how long the fall took.Full definition in the glossary, the Dow lost 89% after 1929, the Nasdaq 78% after 2000, and the Nikkei about 80% after 1989. Leverage turns falling prices into forced selling, which is why the way down is so much faster than the way up.

5. Recovery

The most underappreciated number in bubble history is how long "back to even" takes: 25 years for the Dow after 1929, about 15 for the Nasdaq after 2000, 34 for the Nikkei after 1989. Markets recover; the people who bought the top often don't stay around long enough to see it.

Fifteen bubbles, one table

The dataset below condenses the research corpus behind this site. Dates and percentages are best-estimate ranges; for several older episodes, sources genuinely disagree, and the ranges reflect that rather than false precision. {rows.map((r) => ( ))} Drawdowns are for each episode's primary asset or index. Scroll sideways on small screens. Estimates for the 19th-century and Florida episodes diverge between sources; see Sources for the full citation list. How these historical patterns compare to the AI boom specifically is weighed on Conclusions.

Fifteen documented bubbles, roughly 1840-2022
Bubble Years How far prices fell Time back to even Was the underlying thing real? Source
British Railway Mania 1844-1850 50-70% (rail shares) A decade or more; some lines never Yes: the network drove growth for a century CEPR (opens in a new window)
US railroads / Panic of 1873 1870-1879 25-30% broad; 40-60% rail securities Mid-1880s or later; some roads never Yes, over 20-40 years Wikipedia (opens in a new window)
Florida land boom 1924-1928 70-90% (prime Miami lots) Decades; many parcels never (in real terms) Eventually, after very long lags NBER (opens in a new window)
1929 US stock bubble 1927-1932 89% (Dow) 25 years (Nov 1954) Yes: the underlying industries endured market history (opens in a new window)
Nifty Fifty 1967-1974 60-70% (leading names) Roughly 7-10 years; many far exceeded old highs later Yes: many became giants; the price was the problem Stanford GSB (opens in a new window)
Poseidon nickel (Australia) 1969-1971 Over 95% (Poseidon NL) Never Partly: the nickel was worth far less than implied Wikipedia (opens in a new window)
Hunt Brothers silver 1979-1980 75-80% in about three months 31 years to touch $50 again (2011) Silver stayed useful; the corner-driven price didn't Britannica (opens in a new window)
Japan's asset bubble 1986-2003 About 80% (Nikkei, over 13 years) 34 years (Nikkei reclaimed its 1989 high in 2024) Yes: Japan's industry stayed world-class market history (opens in a new window)
Nordic banking bubbles 1986-1993 50-70% (banks); 30-50% (property) Roughly a decade, after painful reforms Yes: the systems were rebuilt and thrived PIIE (opens in a new window)
Asian Financial Crisis 1993-1998 50-80% local currency; often 80-90% in dollars 5-10+ years for many markets Yes: the export economies endured PIIE (opens in a new window)
Dot-com bubble 1998-2002 78% (Nasdaq) About 15 years (2015) Emphatically: the internet was the future, just not at those prices Britannica (opens in a new window)
Telecom / fiber bubble 1998-2002 80-95%; several giants to zero A decade or more; the sector consolidated Yes: the 'wasted' fiber later carried the modern internet Richmond Fed (opens in a new window)
US housing / subprime 2002-2012 27-35% nationally; over 50% in some cities About 10 years (2016-2017) Housing is real; the lending behind it wasn't sound overview (opens in a new window)
Chinese A-shares 2007 2006-2008 About 70% in roughly a year Not durably held for over a decade The economy was real; the market was policy-driven and volatile chronology (opens in a new window)
SPAC / meme / crypto complex 2020-2022 70-99% depending on the asset Mixed: Bitcoin partly; most SPACs and meme stocks never Partly: a few durable pieces inside a lot of froth Wikipedia (opens in a new window)

Drawdowns are for each episode's primary asset or index. Scroll sideways on small screens. Estimates for the 19th-century and Florida episodes diverge between sources; see Sources for the full citation list.

The recurring finding: the technology is usually real

Read the last column again. Railways carried Britain's economy for a century. The "worthless" 1990s fiber later carried streaming and cloud computing. Even the 1929 crash destroyed prices, not the industries underneath. In almost every major bubble, the technology was real; the prices and the timing were the problem.

That cuts both ways for AI. It means "AI will change everything" is not evidence against a bubble; investors in 1845 and 1999 were broadly right about the technology and still lost most of their money. And it means a crash, if one comes, would say little about whether AI ultimately matters. The historical pattern is that value arrives, but often years later than the spending assumed, and often to a different set of owners than the ones who paid for the build-out.